Towards a Smart and Transparent Road-Based Vehicle Speed Detection System in Tanzanian Highways: A Review of Methods, Technologies, and Systems

Abstract

Accurate and transparent vehicle speed data are crucial for enforcing speed limits and other important applications. However, attaining the required levels of accuracy and transparency remains a challenge that needs to be addressed. The potential for further improvement is brought by technological advancements. To address this, it is necessary to understand the current developments in speed detection methods, technologies used in speed detection systems, and challenges of existing systems. This work reviews vehicle speed detection methods and provides a guideline for selecting an appropriate method. This work also reviews technologies for implementing smart systems and proposes an integrated approach for enhancing intelligence, interconnection, and transparency. Not only this, but this work also evaluates existing vehicle speed detection systems and highlights the need for further research. Furthermore, this work proposes a conceptual framework that integrates the Internet of Things, Artificial Intelligence, cloud computing, and blockchain technologies to enhance vehicle speed detection systems, particularly for developing countries. The Internet of Things facilitates data collection and transmission, ensuring system interconnectivity, while Artificial Intelligence is used for data pre-processing in cloud computing to improve system intelligence and scalability. Meanwhile, blockchain guarantees data security and transparency. A proof-of-concept demonstrator was implemented to validate the proposed conceptual framework. Evaluation results demonstrate an auspicious performance regarding end-to-end data delivery and transmission latency. This work provides both theoretical and practical insights regarding smart and transparent vehicle speed detection systems.

1. Introduction

Road traffic accidents are among the leading causes of human deaths and major contributors to disabilities. About 92% of road traffic deaths occur in middle- and low-income countries [1]. Tanzania, in particular, is ranked 10th in position for road accidents globally. The latest data show that about 18,054 road traffic accident deaths were recorded in the year 2020, which is equivalent to 6.12% of total deaths [2]. Road accidents are caused by automotive, environmental or infrastructure, and human factors. Studies indicate that human factors are the major contributors to road accidents [3,4,5,6]. Factors such as speeding, drunk driving, and mobile phone usage are reported to cause the majority of road accidents in Africa [7,8,9], and Tanzania in particular, at about 72% [10]. In Africa, of all the human factors, speed contributes to a minimum of 30% of deaths caused by traffic accidents [11,12].

Developed and developing countries are addressing speeding by enforcing speed limits. The speed limits are enforced in high-risk areas, such as schools, markets, and village or town centers, as well as in dangerous zones, including sharp corners, winding areas, slippery roads, and steep slopes. Unfortunately, the majority of drivers do not respect road signs regulating speeds [13,14,15]. Therefore, speed enforcement mechanisms are put in place to ensure speed limits are properly followed [16,17].

In Tanzania, both traditional and technological mechanisms are used for speed limit enforcement. Public transportation has stricter arrangements, monitored throughout the journey using logbooks and speed governor methods. With the logbooks, travel time between traffic police checkpoints is monitored manually. Technologically, both vehicle-based and road-based mechanisms are employed for speed control. In vehicle-based approaches, speed-governing devices are installed in buses and cargo trucks to monitor and report speed violations to authorities. In road-based methods, speed guns are used by traffic police officers at random risk or dangerous zones to capture speeding vehicles. If a violation is detected, the officer immediately stops the vehicle or sends photographic evidence to officers ahead for appropriate sanctions.

As the logbook method is performed manually, it is tedious and time-consuming. The speed governors are tempted to not work completely or properly, such as sending false information to the authorities. On the side of handheld speed guns, they are not fully automatic because of the manual practice of tracking and sanctioning speed violators. This practice is time-wasting to road users, costly to authorities, and limited by time and weather factors. Another limitation of speed gun is that they are standalone devices and cannot send data to a central system.

Therefore, the current road-based methods for vehicle speed detection in Tanzania are inefficient and manually operated [18]. The methods are also not transparent and exposed to favoritism, forgery, and corruption. Not only are the methods limited by time and weather conditions, but they also depend on traffic officer commitment. There exist off-the-shelf solutions, but they are expensive to acquire and maintain, making them out of reach to developing countries like Tanzania [19]. Therefore, a need for innovating an efficient, transparent, and affordable solution is evident.

To implement such a solution, a method for measuring vehicle speed and capturing evidence photos is needed. There is no perfect choice because each method has advantages and disadvantages. Modern non-intrusive methods such as radar-, laser-, and vision-based methods are nowadays used in favor of traditional intrusive methods. It is possible to use a single method or combination of methods to deliver the required functionalities. If a single method suffices, it implies weaknesses are not affecting results or can be addressed. Otherwise, a combination of methods is necessary to complement each other. For example, one method may be simple but affected by weather conditions while the other may be robust but complicated. Another advantage of the combined approach is performance optimization. A method may deliver the desired functionalities, but combining with another may make the system more efficient and reliable or provide simplicity. For example, a vision method is capable of measuring both speed and identifying vehicle details like number plate, color, make, and even model [20]. Identifying the most appropriate method for an application context can be a challenging and tedious task, and if not carefully considered it can result in inefficient system implementations. Unfortunately, to the best of the authors’ knowledge, this study could not find any relevant and recent work that provides guidelines for selecting road-based vehicle speed detection methods.

Implementing a smart and transparent system requires more than just speed measuring methods. Technologies for extracting license plate numbers from captured images as well as solutions for communication, storage, and processing are also essential. With the rapid advancement in information and communication technologies, it is important to understand their capabilities, appropriate use cases, and selection criteria. Unfortunately, this work could not identify any recent literature providing clear and specific guidelines for selecting technologies to implement smart and transparent road-based vehicle speed detection systems.

There exist some efforts performed by previous researchers in Tanzania, such as [11,18,21], and elsewhere, such as [22,23,24,25], towards automating vehicle speed detection. These studies have proved the concept of smartly detecting vehicle speed without relying on handheld speed guns. However, vehicle speed detection remains a challenge that requires further attention due to the limitations of the existing solutions and the obvious opportunity for further improvement brought by technological advancements. For instance, the previous works predominantly employ the Internet of Things (IoT) in conjunction with cloud computing. However, the centralized nature of cloud computing raises questions regarding security and trust. Specifically, the reliance on a central entity affects both integrity and transparency [26,27]. Furthermore, the desired quality factors such as efficiency, cost, and simplicity are yet to be fully realized. This situation indicates that more work needs to be conducted. Aligning with the need, this study proposes to explore how emerging technologies, IoT, Artificial Intelligence (AI), and blockchain can be leveraged to implement a vehicle speed detection solution feasible for developing countries like Tanzania.

To implement such a solution, a comprehensive review of vehicle speed detection methods and a thorough technology analysis are necessary for identifying the most suitable methods and technologies. The main objective of this work is to review the methods, technologies, and existing systems to propose an implementation of a smart and transparent vehicle speed detection system for Tanzania. Specific objectives, research questions, and study methodologies are tabulated in

Table 1. Study objectives, research questions, and methodologies.

S/n	Objectives	Research Questions	Methodology
i.	To review vehicle speed detection methods.	What are the strengths and weaknesses of the existing vehicle speed detection methods?	Literature review
ii.	To review technologies for implementing vehicle speed detection systems.	Which technologies can be used to implement smart and transparent vehicle speed detection systems?	Literature review
iii.	To review existing vehicle speed detection systems.	What are the strengths and weaknesses of existing vehicle speed detection systems?	Literature review
iv	To propose the implementation of a smart and transparent vehicle speed detection system compatible with developing countries’ environment.	How can the weakness of the existing vehicle speed detection systems be addressed to implement a system compatible with developing countries’ environments?	Literature review and brainstorming with domain and technology experts.

, while Figure 1 depicts the study flow chart.

Generally, this work contributes to the ongoing efforts towards achieving total automation of vehicle speed detection feasible for developing countries like Tanzania. It establishes a solid foundation, offering both theoretical and practical insights for the development of smart and transparent vehicle speed detection systems. The specific contributions of this work include the following:

• Developed a taxonomy and a summarized guideline for selecting vehicle speed detection methods.
• Proposed new technological integration of smart and transparent vehicle speed detection system implementation.
• Identified research gaps for addressing existing systems’ limitations and meeting efficiency and transparency requirements.
• Conceptualized and demonstrated how emerging technologies (IoT, AI cloud computing, and blockchain) can be leveraged to implement smart and transparent vehicle speed detection systems.

The rest of the paper is organized as follows: A review of methods and technologies is presented in Section 2 and Section 3, respectively. Section 4 presents reviews of existing road-based vehicle speed detection systems. Section 5 proposes an implementation of a smart and transparent system for improving vehicle speed detection in Tanzania, while the conclusion and future works are presented in Section 6.

2. Review of Methods for Vehicle Speed Detection

There exist many methods for measuring vehicle speeds; the methods range from traditional ones such as pneumatic tubes to emerging modern methods such as vision-based ones. Unfortunately, there is no agreement on the best method; thus, choosing a method for an application context is not a straightforward mission due to various factors (such as cost, accuracy, reliability, and simplicity) and conditions (such as weather, road type, etc.). Therefore, it is important to evaluate the available methods to be able to choose the most appropriate one for a specific requirement. The available methods are generally classified into two broad categories, road-based and vehicle-based [28]. The methods can also be classified according to their working methodology as either active or passive. Active methods emit signals and monitor how the signals are affected by the moving vehicles to estimate speeds. Examples of active methods are active infrared, radar, and laser. Passive methods, for example, video-based and passive infrared, wait for the moving vehicles to pass by to estimate speed by using respective mechanisms [29]. The literature review presented in this work focuses on speed measurement methods according to their placements. Figure 2 presents the taxonomy of vehicle speed measuring methods according to their placements.

2.1. Road-Based

Road-based data collection methods operate from either within the road surface area (interferes with traffic flow) or at the roadside (not interfering with traffic flow). There are two broad categories of road-based speed methods, intrusive and non-intrusive [30]. The methods are categorized by their nature of operation in terms of placements and functionalities. Measurement devices of both types can either be temporarily or permanently installed; however, sub-surfaced intrusive methods are usually permanently installed [31].

2.1.1. Intrusive Methods

These are the conventional methods whereby measuring devices are placed on top of the road surface or buried within the road surface. Because of their placement locations, they tend to affect the normal flow of vehicles. Two neighboring devices are used to estimate speed by using distance and travel time [31]. The common intrusive methods are pneumatic tubes, piezoelectric sensors, and inductive loops [32]. The intrusive methods are well-established, stable technologies that have been useful for monitoring road traffic parameters such as vehicle speed, count, and classification. The problem with these methods is that they tend to disrupt normal traffic flow during their installation, operation, and maintenance. The devices for measuring speed are also exposed to damage risks if installed on poor road surfaces or during road maintenance operations [33].

The first intrusive method is pneumatic tubes, which use a pair of nearby rubber tubes installed on the road surface across road lanes. The pneumatic method is used only temporarily due to the delicate nature of the tubes [31]. Vehicle speed is estimated by using the principle of pressure changes produced by the vehicles passing over. The tubes are connected to electronic devices that are activated each time a vehicle crosses the tubes. The time taken to receive activations from the two tubes and the distance between them is used to compute speed. The method is also useful for other functionalities such as counting and categorizing vehicles [34]. Pneumatic tubes are lightweight and portable, which makes their installation and maintenance easy and cost-effective, and means they are sensitive to detect motorcycles, consume low power, and are waterproof. Disadvantages are their visibility to influence drivers and vulnerability to damages [35].

Piezoelectric sensors are the second intrusive method, working on the principle of conversion of mechanical energy to electrical energy. The sensor is installed inside a groove cut within the road tarmac surface. Vehicles passing over the sensor exert pressure, which causes potential differences; the amplitude and frequency of the potential difference signal are used to measure the speed and weight of the moving vehicles [36]. Piezoelectric sensors are one of the cheapest intrusive methods in terms of maintenance cost and service time, and they are also compatible with road surfaces; thus, installation is simple. However, the method may be sensitive to temperatures and influence driving behaviors because of its visibility [25].

The third intrusive method is the Inductive Detector Loop (IDL), which consists of wire coils buried within the road surface. They can be installed in single- or dual-loop configurations. It is not possible to perform direct speed measurements with single loops; therefore, the dual loops, which are installed in pairs, are used for direct speed measurement [37]. As the vehicle passes over the dual IDL electromagnetic fields, the metal content of the vehicle affects the inductance of the loop, and the vehicle is detected at each point. The distance and time to travel between the two points are used to estimate vehicle speed. Count and occupation data are also collected using this method [38]. IDLs are cheap and fast in data processing and are unaffected by weather conditions. However, poor response of the IDL at certain times affects vehicle detection as well as accuracy when operating under congestion. Maintenance cost is also a concerning factor with IDL.

Table 2. Summary of intrusive methods.

Technique	Working Principle	Advantages	Disadvantages
Pneumatic tubes [31,32,35]	Based on the principle of pressure changes caused when a vehicle passes over a pair of tubes. Speed is estimated based on time taken and distance between the tubes.	Also counts and categorizes vehicles. Lightweight and portable. Cost-effective, sensitive to detect motorcycles, consumes low power, and is weatherproof.	Vulnerable to damages and requires periodic maintenance. Tubes are visible, influencing driving behaviors.
Piezoelectric [32,36,39]	Works on the principle of conversion of mechanical to electrical energy. Speed is measured from the amplitude and frequency of the potential difference produced by the pressure exerted by the passing vehicle.	Cheapest in terms of maintenance cost and time. Easy to install.	The method is sensitive to temperature and influences driving behavior because it is visible to road users.
Inductive Detector Loop [32,37,38,40]	As the vehicle passes over the dual IDL, the metal content of the vehicle affects the inductance of the loop and is detected. Distance and time to travel between the two IDL points are used to calculate speed.	Also, collects count and occupancy data. IDL is cheap, fast in data processing, and not affected by weather conditions.	Poor response of the IDL at times. Accuracy issues when operating under congestion and maintenance costs.

provides a summary of the reviewed intrusive vehicle speed detection methods.

2.1.2. Non-Intrusive Methods

In non-intrusive methods, the measuring devices are installed above road lanes or at the roadside, where they can monitor more than one road lane. Compared to intrusive methods, these methods do not affect traffic flows during installation, operation, or maintenance and are packaged into sensor units for remote observations [33]. Data communication is achieved through wireless technologies like WiFi, cellular, or Low-Power Wide-Area Networks (LPWANs) such as Long Range (LoRa) [41] depending on location and requirements. Data can also be stored locally in the sensor units for backup. The widely used non-intrusive methods for speed measurement are vision-, radar-, and laser-based [34,39,42]. The methods can be used in either fixed or mobile modes. Fixed mode is when the sensor unit is mounted above or at the roadside. Mobile mode can either be a sensor device temporarily installed at a specific location for a certain period and then moved to another location or can be handheld guns.

Vision-based is the first non-intrusive method, which has recently become very promising for vehicle speed measurements. Its effectiveness has been fostered by technological advancements in cameras, image processing techniques, and computing devices. Various image processing techniques have been employed to analyze video frames to detect and track vehicles and be able to measure their speed [24,29]. This method possesses great challenges and opportunities to be addressed and exploited, respectively. The main opportunity is cost reduction, brought about by the possibility of using already existing traffic cameras. However, this method faces challenges, including high computational cost [43] and not being mature enough to handle accurate speed detection at night time and in challenging weather conditions [20].

Microwave detectors or Doppler radar is second in this category. The name radar (radio detection and ranging) is derived from the work it performs: detection of objects’ locations and their distance from the radar device. The name microwave is also used because it works in the microwave wavelength [39]. It is also called Doppler because it uses the Doppler principle to compute vehicles’ instantaneous speeds [36]. Unlike vision-based, radar methods are not affected by weather conditions and can work during the day and night. It is a matured technology used since World War II for military applications; it is also the easiest and fastest method for detecting vehicle speed [44]. Furthermore, radar methods are some of the most accurate and can capture the speed of vehicles traveling in both directions of the road. They support both fixed and mobile modes. If in fixed mode, radar recorders require no human support and are not easily noticed to influence driving behaviors. However, mobile speed radar guns are labor-intensive and are affected by the behaviors of humans operating the device [45].

Laser-based is also an intrusive method that is useful for measuring speeds and counting and classifying vehicles. The laser technology uses laser beams to capture vehicle instantaneous speeds. The device emits a beam of laser that is reflected to the device, whereby the time taken by the beam to travel and return can be translated to speed. The advantage of laser technology is reliable and durable measurements. Unlike radar, while using a laser speed gun, it is possible to target the exact vehicle without capturing untargeted vehicles. However, laser technology can be more sensitive to weather conditions such as humidity and precipitation [45,46].

Other non-intrusive methods are ultrasonic and infrared (active and passive). Ultrasonic and active infrared-based methods operate on the principle of change in sound and light signals [34], respectively. Passive infrared requires two sensors to measure vehicle speed using the distance and time taken to travel between them [25,47]. These methods are inexpensive options and are mostly used in vehicle detection [19]. The performance of these methods is, however, affected by environmental conditions and not so well suited for speed measurement. A summary of the reviewed non-intrusive vehicle speed detection methods is provided in

Table 3. Summary of non-intrusive methods.

Technique	Working Principle	Advantages	Disadvantages
Vision-Based Method [34,39,42]	Leverages advancements in cameras, image processing techniques, and computing devices to analyze video frames to track vehicles and measure their speed.	Cost reduction and the possibility of using already existing traffic cameras	High computational cost. Not matured technology, accuracy issues at night time, and affected by harsh weather conditions.
Radar-Based Method [36,39,43].	Works by sending radio signals, which are reflected to the source device upon hitting the target object. The difference in returning signals is used to compute vehicles’ instantaneous speed by using the Doppler principle.	Matured technology, not affected by weather conditions, can work during the day and night. Easiest, fastest, and one of the most accurate methods capable of capturing the speed of vehicles traveling in both road directions.	It captures all vehicles in range, including untargeted vehicles.
Laser-Based Method [45,46].	Uses laser beams to capture the instantaneous speed of the vehicle. The device emits a beam of laser that is reflected. The time taken by the beam to travel and return is translated to speed.	Reliable and durable measurements. Useful for measuring speed, counting, and classifying vehicles.	Sensitive to weather conditions such as humidity and precipitation.
Ultrasonic-Based Method [19,47]	Works in the same way as a radar and laser. The fundamental difference is that it uses sound waves instead of radio or laser.	These methods are inexpensive options and are mostly used in vehicle detection.	Performance is affected by environmental conditions and is not well suited for speed measurement.
Infrared (Active and Passive)-Based Method [19,25,47]	Active infrared works in the same way as ultrasonic but uses light signals. Passive infrared uses two sensors; distance and time taken to travel between them are used to measure speed.	These methods are inexpensive options and are mostly used in vehicle detection.	Performance is affected by environmental conditions and is not well suited for speed measurement.

.

2.2. Vehicle-Based

These methods are known as In-vehicle Monitoring Systems (IVMSs) because they measure speed and other behaviors using mobile devices that are placed inside vehicles [13]. Floating Car Data (FCD) is a term also referring to this category of methods. Vehicle-based methods are alternatives to or rather complement road-based methods for measuring vehicle speed, location, and direction in either positional or continuous settings.

Positional speed measurement can be achieved using Radio Frequency Identification (RFID) technology, whereby vehicles are equipped with RFID tags and RFID readers are installed at specific road points [31]. When a vehicle passes a reader, its data gets captured. The time taken to travel between two adjacent readers is used to compute speed. Continuous speed measurement can also be achieved using the Global Positioning System (GPS) or the Global System for Mobile Communication (GSM), utilizing mobile phones or GPS modules as road sensors [28]. Vehicle-based methods enable low-cost speed measurements over a wide geographical area by exploiting the already existing GSM and GPS infrastructure. This is achieved by equipping a vehicle with either a mobile phone or a GPS module [48]. The disadvantage of RFID technology is its complexity and low penetration rate [49]. In addition to requiring a vehicle to be equipped with an RFID tag, it requires physical installation of RFID readers at selected road points, making deployment tedious. Generally, vehicle-based methods are limited to particular vehicles such as taxis, trucks, and buses and need to seek the consent of the driver or owner. These methods also possess privacy concerns [50]. However, the willingness of private vehicle owners and drivers to be used as probe vehicles can be influenced by incentives.

Table 4. Summary of vehicle-based methods.

Technique	Working Principle	Advantages	Disadvantages
RFID-Based Method [31,49].	Vehicles are equipped with RFID tags. Readers are installed at road points to detect tags. Travel time between two readers determines speed.	Low-cost and suitable for positional speed measurement.	Requires more tedious work to install and maintain RFID readers at road points.
GPS-Based Method [28,48]	Uses a GPS module equipped within the vehicle for continuous speed measurement.	Low-cost, uses existing GPS network, and provides location and road speed limit information.	Suitable for specific vehicle types such as trucks. Requires consent of driver or owner and raises privacy issues.
GSM-Based Method [28,48]	Uses a mobile phone connected to a GSM network for continuous speed measurement.	Low-cost, uses the existing GSM network.	Limited to particular vehicles such as taxis. Needs consent of the driver or owner and raises privacy concerns.

summarizes the reviewed vehicle-based speed detection methods.

3. Review of Technologies for Implementing Smart and Transparent Vehicle Speed Detection Systems

Smart systems are characterized by perception, communication, embedded knowledge, reasoning, learning, and control capabilities. Self-organization and context awareness are considered emerging properties [51]. These capabilities enable these systems to understand a situation, make predictive or adaptive decisions, and take smart actions [52,53,54]. Intelligence and interconnection must be achieved to implement smart systems capabilities. AI and IoT are important technologies for this achievement [55].

AI is an intelligence demonstrated by machines with the ability to think and act like human beings [54]. It is an umbrella term for various areas such as Machine Learning (ML), Natural Language Processing (NLP), Computer Vision (CV), Speech Recognition, Robotics, and Expert Systems [56]. These areas can also be categorized as interactive AI such as chatbots, text AI such as speech-to-text conversion, functional AI such as robotics, analytic AI such as sentiment and predictive analysis, and visual AI such as computer vision [57]. The enormous applications of AI include healthcare, education, agriculture, gaming, social media, marketing, finance, and travel and transportation [58]. In transportation, AI is one of the key technologies for implementing intelligent transportation systems (ITSs) alongside IoT, cloud computing, and blockchain [59,60,61,62,63,64]. Speed monitoring and control (speed management) is a function of ITSs within the Intelligent Traffic Management (ITM) area [61]. Thus, AI can be leveraged to implement smart speed detection systems. ML and CV are the branches of AI that are widely used for vehicle speed detection [65,66,67]. ML and CV techniques perform various functions such as object detection, segmentation, tracking, and feature extraction to extract useful information from images and videos [68], typically vehicle number plate and speed. For example, ref. [48] applied Deep Learning (DL) models with CV approaches to propose a system for vehicle speed estimation.

IoT can be defined as a network of physical objects collectively called things [69]. These things are embedded with electronics, sensors, actuators, and communications to facilitate data collection, processing, and sharing without human involvement [70,71]. The goal of IoT is to connect all things to the internet beyond the usual things like computers and mobile phones [72]. The interconnection of all things through IoT brings the physical world and the digital world together [72,73]. However, there is no global consensus on IoT architecture, and different architectures have been proposed, such as three-layer, four-layer, and five-layer [74]. The four-layer IoT architecture produces a good balance between the perception, network, service, and application layers [55]. This architecture is used in this work as a reference architecture.

IoT uses a wide range of sensors, actuators, and controllers (collectively called embedded devices) depending on the application domain [75]. Sensors convert physical activity into analog or digital signals, while actuators convert the digital signal into physical action [76,77,78]. The controllers are responsible for managing IoT operations and are either microcontroller- or microprocessor-based [79]. Communication technologies can be short-range, such as Bluetooth and RFID, medium-range, such as Wi-Fi and Zigbee, or long-range, such as cellular and LPWAN [80,81]. Apart from range, communication technologies have different cost, speed, and power requirements [55]. The service layer provides data storage and processing services, using cloud computing technology, which may be private or a service [82]. Free and commercial IoT cloud service providers are available, such as Arduino IoT Cloud, ThingsPeak, Google Cloud Platform (GCP), Amazon Web Services (AWS), etc. The choice of a cloud platform depends on factors grouped into quality, technical, and economic. The factors can be classified further, for example, quality (reliability and usability), technical (storage capacity and processing performance), and financial (acquisition and maintenance) [83]. Application layers include user interfaces such as web and mobile interfaces or notification mechanisms such as text messages, email, and phone calls. IoT applications can be found in smart transportation, home automation, security and public safety, healthcare, logistics, smart environments, and agriculture [84].

Cloud computing has been widely used to provide data storage and processing services for data collected by IoT devices. Cloud computing poses challenges such as single point of failure, transparency, security, and privacy concerns. On the other hand, blockchain is a decentralized, distributed, and peer-to-peer (P2P) technology designed to be more immutable, transparent, and trustworthy [84,85]. As a result, this technology is vital in tackling these security concerns and challenges [83,86,87]. Therefore, there has been a shift in the IoT research field from the IoT–cloud computing partnership to the IoT–blockchain partnership [88]. However, cloud computing is still a very important technology that is unlikely to be completely replaced by blockchain. Instead, they can be used together to complement each other. The level of cooperation between the two technologies depends on the application environment and business requirements.

In traditional vehicle speed detection systems, data collection is the duty of sensor nodes, while data storage and processing duties are facilitated by cloud computing. However, with the emergence of blockchain, storage and processing duties can be distributed between the two technologies. For example, data processing can be performed with cloud computing, while data storage can be performed through blockchain or any other possible configuration.

Although it is possible to achieve business requirements with one technology alone, balancing between centralization and decentralization is key to achieving optimum results. These possible improvements have made the integration of IoT, cloud computing, and blockchain an emerging research area that has recently attracted the attention of many researchers, including in this work.

Thus, this work proposes a combination of IoT and AI alongside cloud computing and blockchain technologies to be used to provide the needed interconnection, intelligence, and transparency factors for vehicle speed detection systems. IoT and cloud computing have been widely used; however, the benefits of blockchain are yet to be exploited in this domain. IoT interconnectedness and multidisciplinarity provide the whole system infrastructure to interconnect all the elements involved. Cloud computing will perform data pre-processing by using AI techniques to give meaningful information. Blockchain will facilitate transparent and secure data storage among stakeholders.

4. Review of Existing Road-Based Vehicle Speed Detection Systems

Vehicle speed detection is a crucial research area in intelligent transportation systems (ITSs). This section reviews the literature to explore various methodologies, technologies, and implementations of road-based vehicle speed detection systems, as summarized in Table 5.

Vehicle speed data are essential in different fields of applications, such as traffic management and control, urban planning, and environmental and epidemiological studies. They play a crucial role in modeling noise and air pollution and assessing potential health impacts to support environmental regulations [89]. There are typically two types of vehicle speed detections, instantaneous and average speeds. Instantaneous speed is the actual speed a vehicle travels at a specific time, while average speed is the speed computed by how much distance a vehicle has traveled during a certain time [43]. In traffic management and control, these speed types serve different purposes; instantaneous is used for real-time speed detection ideal for speed limit enforcement. On the other hand, the average speed is used for speed monitoring over long distances and is applicable in speed enforcement, traffic analysis, fuel consumption control, etc. However, both instantaneous and average speed measurements are crucial in vehicle speed detection. Instantaneous speed can be measured by road-based methods such as radar and laser guns. Average speed is commonly measured by speed cameras using computer vision techniques. Overspeed detection is achieved by comparing the measured speed (instantaneous or average) with a pre-set speed limit.

An intrusive-based system is developed in [90] to detect instantaneous speed by embedding a network of acceleration sensors in the road pavement. Using several algorithms, these sensors detect vibrations generated by passing vehicles to extract vehicle speed and other traffic information. Similarly, a piezoresistive-based sensor system is introduced by [91] for measuring instantaneous vehicle speed. The sensor is installed on the roadway, while the supporting electronics are positioned on the pavement edge. The challenge with these systems is that they disrupt normal traffic flow during installation, operation, and maintenance and may also be damaged if deployed on poor road surfaces.

The research in [11] focused on monitoring and analyzing drivers’ speeding behaviors in dangerous zones by deploying off-the-shelf Average Speed Cameras (ASCs). These cameras monitor vehicle speeds and transmit data to a central server. While the system demonstrated the feasibility of smartly detecting speeding vehicles, its high acquisition cost remains a limitation. However, the study highlighted the need for further research to develop more affordable and efficient vehicle speed detection systems. The proposed system by [18] features an algorithm for multi-vehicle speed estimation designed for multi-lane roads, utilizing image processing and CV techniques. The system is client–server-based, where the client is deployed at the roadside to detect speed violations. When a violation occurs, the client captures images of the vehicle, stores them locally, and then transfers them to a central server. While the client’s implementation details are not provided, it appears to be computer-based. Otherwise, the study is primarily experimental to evaluate the algorithm’s effectiveness. Further enhancements are required to improve the system’s performance, particularly for deployment in remote areas. In addition, the study by [24] proposes an integrated framework for vehicle speed estimation using a Light Detection and Ranging (LiDAR) sensor deployed at the roadside. This study demonstrates that LiDAR is an effective method for detecting vehicle speeds in urban environments due to its high accuracy. However, it has limitations, such as restricted detection range and occlusion challenges. The study suggests integrating multiple sensors to mitigate these challenges.

A real-time speed estimation system utilizing RFID technology is presented in [92]. The work integrates a modular neural network (MNN) to tackle accuracy issues associated with a single camera. A similar RFID-based system is presented in [93], whereby [25] proposes an IoT-based speed measuring system that uses a pair of IR sensors to determine vehicle speed based on the distance between sensors and travel time. RFID technology is employed for vehicle identification, GPS for location tracking, and Wi-Fi for communication. However, the system has drawbacks: RFID requires vehicle involvement, IR sensors are affected by weather conditions, and Wi-Fi limits its applicability to urban areas. Moreover, the system integrates road-based and vehicle-based methods, but the latter depends on the driver or owner’s cooperation, exposing it to possible forgery and manipulation. Similarly, an IoT-based system presented in [22] employs two cameras deployed at the entry and exit locations of the measurement region. It uses a pair of Raspberry Pi modules: one functions as a live-streaming web server for video recording and the other as a webcam server to detect speeding vehicles. The system uses image processing and CV techniques to compute average speed and to extract vehicle number plate details. It is suitable for urban environments; however, its implementation seems complicated and computationally expensive. Additionally, ref. [94] introduces an IoT-based system for estimating instantaneous speed using magnetic sensors, while a portable system for instantaneous speed estimation using impulse radio ultra-wideband (IR-UWB) radar and ARM Cortex-M7 is presented in [95]. These works illustrate how emerging technologies combined with robust sensing technologies can provide affordable, efficient, lightweight, and easy-to-install alternatives to conventional speed detection systems.

Other IoT-enabled systems [96,97] focus on detecting overspeed for law enforcement rather than gathering general-purpose data. These systems are hardware-dominated and notification-based, and data are not available for analysis. In particular, a system is described in [98] to detect vehicle speeds using a Programmable Logic Controller (PLC), two laser sensors, an HMI (Human–Machine Interface), a camera, and a CMR (Communication Module Radio) to detect vehicle speeds. If an overspeed is detected, the camera captures an image of the vehicle, stores it on an SD card, and sends an SMS notification to authorities for enforcement. On the other hand, the system described in [99] is typically vehicle-based, consisting of a microcontroller, an IR speed sensor, an LCD, and a buzzer. These components are all installed on the vehicle to measure speed and notify the driver if it exceeds safe limits. Moreover, similar IoT-enabled, vehicle-based systems are presented in [100,101,102,103,104,105,106,107].

Classical computer vision techniques are still utilized in vehicle speed estimation, as witnessed in recent studies that continue to implement deterministic, rule-based image and video processing algorithms. The study in [23] presents an algorithm for estimating vehicle speeds using digital image processing. It incorporates background subtraction, binary image formation, and grayscale image processing to track the travel time between two reference points for speed calculation. Similarly, ref. [108] introduces a real-time vehicle speed detection method using image and video processing. It isolates moving vehicles through masking and frame subtraction. Speed is determined by measuring the time interval between frames and the distance traveled by detected objects. While these works demonstrate promising precision and accuracy, they require enhancements to support license plate identification. Additionally, the system in [109] employs CV methods using a single camera configuration, augmented by image processing, headlight recognition, and vanishing point algorithms to estimate vehicle speeds during night time. Moreover, ref. [110] proposed a model that performs uncalibrated detection and tracking, tackles occlusion issues, and enhances speed estimation accuracy using a homography transformation regression network. Not only that, but the work in [111] also introduces a real-time vehicle speed measurement system employing image processing techniques such as morphological operations and binary logic. The system adapts bounding box sizes based on vehicle dimensions and operates within a defined Region of Interest (ROI) using a two-line approach.

In contrast, recent advancements in CV have seen widespread application of ML and DL models for vehicle speed detection. For instance, an intelligent vehicle violation detection system is proposed in [112] to address low accuracy, poor stability, and slow detection speed issues. The system employs CV for pre-processing image data from the Bit Vehicle model dataset, Kalman filtering for vehicle tracking, and human–computer interaction (HCI) technology to create a user-friendly interface. Similarly, ref. [113] presents an approach for determining vehicle speed using an SSD (Single-Shot Multi-Box Detector) model for detection, combined with the Dlib library and DeepSORT for tracking. The system computes speed based on relative velocity, demonstrating reliable detection and robust tracking. In addition to this, the DL-based model, You Only Look Once (YOLO), is widely employed for vehicle detection and tracking, whereas speed estimations are achieved in combination with other approaches. The employed approaches for speed estimation are the main distinguishing factors between studies. The study in [114] presents a Logistic Vehicle–YOLO (LV-YOLO) approach for speed detection, which involves segmentation using U-net to isolate vehicles in frames and detection for identifying trucks along with speed estimation based on the Boxy Vehicle dataset. Meanwhile, studies in [115,116,117] enhance speed estimations by utilizing version 3 of the YOLO (YOLOv3) model in combination with various algorithms such as the Kalman filter and Euclidean. Further enhancements in vehicle speed detection have been achieved in studies [118,119,120] through the application of the YOLOv4 model.

On the other hand, the study in [121] introduces a speed detection system specifically tailored for highway traffic. It leverages the YOLOv5 model for vehicle detection and the Deep Simple Online and Realtime Tracking (SORT) model for tracking. The Shifted Window Transformer (Swin Transformer) block is integrated to enhance detection accuracy. Vehicle speed is determined by averaging instantaneous speeds across multiple frames. Similarly, more YOLOv5-based works are discussed in [122,123,124], while ref. [125] presents a vehicle speed detection mechanism employing a Convolution Neural Network (CNN)-based Hybrid Vehicle Detection Network (HVD-Net) alongside SORT detection and tracking. These works claim to perform well in highway applications. Additionally, the proposed framework in [126] estimates vehicle speed using an onboard camera. Vehicles are detected and tracked with a YOLOv7 model and a tracking algorithm, and speed estimation is performed through a linear regression model. The system is vehicle-based, designed specifically for autonomous vehicles, but it can easily be adapted for road-based use by simply stopping the measuring vehicle or by attaching the camera to a pole or building. Another YOLOv7-based work for vehicle speed detection is presented in [127]. Furthermore, a vehicle speed detection system using the YOLOv8 model for vehicle detection and tracking in video feeds is proposed in [128], aiming to enhance road safety and traffic management in metropolitan areas. Similarly, the study in [129] evaluates different YOLOv8 variants, Nano, Small, and Medium, for vehicle speed estimation using existing cameras. More works incorporating YOLOv8 for detection, and ByteTrack for tracking, are presented in [130,131], whereby speed estimations are achieved via intrusion lines and linear regression, respectively. Other studies based on YOLOv8 are presented in [132,133]. These studies collectively underscore how advancements in deep learning techniques are consistently enhancing vehicle speed detection.

Interestingly, Unmanned Aerial Vehicles (UAVs) in combination with CV and DL have recently been utilized for monitoring vehicle speeds. A system tailored for UAVs has been proposed in [134] for real-time traffic surveillance capable of detecting, tracking, and estimating vehicle speeds using images captured from the air. Similarly, ref. [135] presents a UAV-based system for immediate detection of vehicle speeds. This system utilizes a Raspberry Pi equipped with a camera along with a Mobile Net-SSD deep learning model to detect vehicle speeds from a UAV operating at various altitudes and angles. More UAV-based vehicle speed detection systems are presented in ref. [136,137]. UAVs serve as effective tools for aerial traffic surveillance, providing real-time visual data that enhance traffic management within smart city settings [138]. However, UAV-based systems face numerous challenges, such as susceptibility to weather conditions, limited battery life, and limitations in detection range that impact their practicality [139]. In addition, demands for processing, regulatory restrictions, and issues related to privacy pose significant barriers to widespread acceptance.

In recent years, the vision-based speed detection method has shown great promise in the vehicle speed detection domain. This is due to the continuous advancements in the CV and ML fields [140]. Additionally, it is cost-effective compared to other methods as it leverages existing camera infrastructure [141,142]. However, it is still immature, challenged by weather conditions, and requires artificial lights at night [24]. Although it is possible to use a single camera, the method mostly requires double installation of cameras at the entry and exit points of the measurement region to compute average speed [22]. Furthermore, it involves complicated image or video processing activities to detect speed and extract vehicle details [128]. Moreover, achieving precise speed measurement with this method remains challenging due to the inherent limitations of cameras in estimating range with precision. On the other hand, data-driven approaches that rely on ML need an advanced and expensive configuration to record videos in real traffic scenarios because the camera must be aligned with a precise sensor to generate accurate ground truth speed data [143]. These weaknesses raise concerns about the suitability of the vision-based system in aiding speed limit law enforcement to fine, suspend driving licenses, and even to jail speed violators. Thus, the usage of high-precision and instantaneous speed detection methods such as radar and lidar is vital for more robust results [132]. Radar is a well-established technology offering high accuracy and monitors the speed of vehicles in both road directions. It also remains unaffected by weather conditions and operates effectively during the day and at night [44,45,98]. In addition, its ability to detect instantaneous speed eliminates the requirement for installing sensors twice.

On the other hand, the existing IoT-based [22,25,96,97,98,99,100,101,104,105,106] systems, client-server-based [11,18,112] systems, and UAV-based [134,135,136,137] systems predominantly employ cloud computing for data storage and processing. Although this combination has been auspicious, the centralized nature of cloud computing raises questions regarding security and trust—specifically, the need for trusting the central entity and raising concerns about integrity and transparency. As a result, there has been a remarkable shift in IoT research from the IoT–cloud computing partnership to the IoT–blockchain partnership. However, integrating IoT with blockchain is not a straightforward mission due to the diverse nature of the technologies. It involves addressing several challenges such as the limited storage and computing capabilities of IoT devices compared to the high storage and computation demands of blockchain technologies [144,145,146,147].

Consequently, the IoT–blockchain combination has become an emerging research area requiring further exploration. In addition to this, there are no universally agreed architecture frameworks for their integration [148]. The existing studies principally focus on specific domains, such as [144,149,150]. To the best of this study’s knowledge, no prior research work has explored the merging of IoT with blockchain in the context of vehicle speed monitoring. This study is motivated by current research trends and the growing need for more efficient and transparent vehicle speed detection systems in modern transportation [151]. Therefore, further research is necessary to explore the potential of integrating IoT and blockchain for enhancing vehicle speed detection. In light of this, this study proposes the development of a smart and transparent vehicle speed detection system that combines IoT, AI, cloud computing, and blockchain technologies, as conceptualized in the following section.

Table 5. Comparison of reviewed works.

Reference	Speed Type	Detection Method	Technologies	Storage and Processing	Comments
[90,91]	Instant	Acceleration/piezoresistance sensors	Wireless Sensor Network (WSN)/embedded system	Centralized/Remote	These systems may disrupt traffic flow during installation, operation, and maintenance and may be damaged if deployed on poor road surfaces.
[11]	Average	Vision	IoT and AI	Centralized	Proves the concept of smartly detecting vehicle speed on the highways of Tanzania. However, not feasible due to cost.
[18]	Instant.	Vision	AI in a client–server architecture.	Centralized	Proves the concept of detecting speed on urban roads of Tanzania. However, the implementation of client nodes is not disclosed.
[24]	Average	LiDAR	AI	N/A	Demonstrated the usefulness and limitations of LiDAR sensors for vehicle speed measurement.
[92,93]	Instant	RFID	IoT and AI	Centralized	RFID technology faces several obstacles in vehicle speed detection such as privacy issues and dependence on vehicle involvement.
[25]	Average	Infrared, RFID, and GPS.	IoT	Centralized	Suitable for urban environments. However, combining road- and vehicle-based methods is susceptible to forgery.
[22]	Average	Vision	IoT and AI	Centralized	Suitable for urban environments but looks complicated and computationally expensive.
[94,95]	Instantaneous	Magnetic sensor	IoT, IoT, Lightweight	Remote/Cloud	Illustrates how emerging technologies can provide practical, affordable, and simple-to-install alternatives to conventional speed sensors.
[96,97,98]	Average/ Instantaneous	Radar/potentiometer/laser	IoT/Automation	Centralized/ Remote	Primarily designed for law enforcement to detect overspeed, not collecting general-purpose data.
[99,100,101,102,103,104,105,106,107]	Average/ Instantaneous	GPS, radar sensors, and accelerometers	IoT/AI/embedded systems	Centralized/ Remote	Vehicle-based systems depend on the trust and cooperation of drivers and are limited to particular vehicles such as trucks and buses.
[23,108,109,110,111]	Average	Vision	Classical CV		Demonstrate promising precision and accuracy. Require enhancements to support license plate identification.
[112,113]	Instantaneous	Vision	AI, HCI	Remote/Centralized	Demonstrate application of non-YOLO-based deep learning algorithms and libraries for vehicle detection, tracking, and speed estimations.
[114,115,116,117,118,119,120]	Instantaneous/Average	Vision	AI	N/A	Demonstrated applications of deep learning models, YOLOv3 and YOLOv4, for speed determination.
[121,122,123,124]	Averaged-Instantaneous	Vision	AI	N/A	Demonstrated improved accuracy in vehicle detection and speed estimation by leveraging the YOLOv5 model.
[125]	Average	Vision	AI	N/A	Acknowledged vision-based speed detection is a complex task as its accuracy depends on detection, tracking, and speed estimation schemes.
[126,127,128,129,130,132,133,151]	Average	Vision	AI	N/A	Highlights the role of advanced deep learning techniques in improving vehicle speed detection.
[134,135,136,137]	Instantaneous	Vision	AI, UAV	Centralized (Edge Computing)	UAVs face various challenges, yet they are effective tools for delivering real-time visual information for enhancing vehicle speed detection.
This work	Instantaneous	Radar and Vision	IoT and AI alongside cloud computing, and blockchain	Centralized processing and decentralized storage.	Proposes integration of IoT, AI, cloud computing, and blockchain technologies for implementing smart and transparent road-based vehicle speed detection systems.

Table 5. Comparison of reviewed works.

Reference	Speed Type	Detection Method	Technologies	Storage and Processing	Comments
[90,91]	Instant	Acceleration/piezoresistance sensors	Wireless Sensor Network (WSN)/embedded system	Centralized/Remote	These systems may disrupt traffic flow during installation, operation, and maintenance and may be damaged if deployed on poor road surfaces.
[11]	Average	Vision	IoT and AI	Centralized	Proves the concept of smartly detecting vehicle speed on the highways of Tanzania. However, not feasible due to cost.
[18]	Instant.	Vision	AI in a client–server architecture.	Centralized	Proves the concept of detecting speed on urban roads of Tanzania. However, the implementation of client nodes is not disclosed.
[24]	Average	LiDAR	AI	N/A	Demonstrated the usefulness and limitations of LiDAR sensors for vehicle speed measurement.
[92,93]	Instant	RFID	IoT and AI	Centralized	RFID technology faces several obstacles in vehicle speed detection such as privacy issues and dependence on vehicle involvement.
[25]	Average	Infrared, RFID, and GPS.	IoT	Centralized	Suitable for urban environments. However, combining road- and vehicle-based methods is susceptible to forgery.
[22]	Average	Vision	IoT and AI	Centralized	Suitable for urban environments but looks complicated and computationally expensive.
[94,95]	Instantaneous	Magnetic sensor	IoT, IoT, Lightweight	Remote/Cloud	Illustrates how emerging technologies can provide practical, affordable, and simple-to-install alternatives to conventional speed sensors.
[96,97,98]	Average/ Instantaneous	Radar/potentiometer/laser	IoT/Automation	Centralized/ Remote	Primarily designed for law enforcement to detect overspeed, not collecting general-purpose data.
[99,100,101,102,103,104,105,106,107]	Average/ Instantaneous	GPS, radar sensors, and accelerometers	IoT/AI/embedded systems	Centralized/ Remote	Vehicle-based systems depend on the trust and cooperation of drivers and are limited to particular vehicles such as trucks and buses.
[23,108,109,110,111]	Average	Vision	Classical CV		Demonstrate promising precision and accuracy. Require enhancements to support license plate identification.
[112,113]	Instantaneous	Vision	AI, HCI	Remote/Centralized	Demonstrate application of non-YOLO-based deep learning algorithms and libraries for vehicle detection, tracking, and speed estimations.
[114,115,116,117,118,119,120]	Instantaneous/Average	Vision	AI	N/A	Demonstrated applications of deep learning models, YOLOv3 and YOLOv4, for speed determination.
[121,122,123,124]	Averaged-Instantaneous	Vision	AI	N/A	Demonstrated improved accuracy in vehicle detection and speed estimation by leveraging the YOLOv5 model.
[125]	Average	Vision	AI	N/A	Acknowledged vision-based speed detection is a complex task as its accuracy depends on detection, tracking, and speed estimation schemes.
[126,127,128,129,130,132,133,151]	Average	Vision	AI	N/A	Highlights the role of advanced deep learning techniques in improving vehicle speed detection.
[134,135,136,137]	Instantaneous	Vision	AI, UAV	Centralized (Edge Computing)	UAVs face various challenges, yet they are effective tools for delivering real-time visual information for enhancing vehicle speed detection.
This work	Instantaneous	Radar and Vision	IoT and AI alongside cloud computing, and blockchain	Centralized processing and decentralized storage.	Proposes integration of IoT, AI, cloud computing, and blockchain technologies for implementing smart and transparent road-based vehicle speed detection systems.

5. Proposed Smart and Transparent Road-Based Vehicle Speed Detection System

5.1. The Conceptual Framework

The proposed system is composed of four major parts: IoT, cloud, blockchain, and application subsystems. The conceptual framework for the entire system is illustrated in Figure 3.

In the IoT subsystem, a radar sensor is proposed for speed detection in combination with a camera for capturing vehicle images. Placement of the sensor is crucial because radar technology has good range; the sensor can be placed at the middle of measurement (i.e., risk or dangerous) section of the road. This approach allows several speed reads from both directions and provides ample time for the camera to capture evidence images. It works by the sensor detecting the Doppler shift frequency corresponding to the speed of the moving vehicle by using the Doppler effect. The velocity of the car in m/s can be computed using the Doppler effect [152], as shown in Equations (1) and (2), and converted into km/h as per Equations (3) and (4).

$$\mathrm{F}\mathrm{d}=2\times \mathrm{V}\mathrm{c}\times \mathrm{F}\mathrm{o}/\mathrm{C}$$

(1)

where

• Fd = measured Doppler shift frequency in Hz;
• Vc = velocity of the target in m/s;
• C = Speed of Light (3 × 10⁸ m/s);
• Fo = operating frequency of the sensor (e.g., 10.525 GHz for HB100).

From Equation (1),

$$\mathrm{V}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{m}/\mathrm{s}=\mathrm{F}\mathrm{d}\times \mathrm{C}/2\mathrm{F}\mathrm{o}$$

(2)

Converting speed in Equation (2) to km/h is performed as follows:

$$\mathrm{V}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{K}\mathrm{m}/\mathrm{h}=\mathrm{V}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{m}/\mathrm{s}\times 3.6$$

(3)

Replacing Vc in m/s in Equation (3) with the actual value as per Equation (2) is performed as follows:

$$\mathrm{V}\mathrm{c}\mathrm{i}\mathrm{n}\mathrm{K}\mathrm{m}/\mathrm{h}=\left(\mathrm{F}\mathrm{d}\times \mathrm{C}/2\mathrm{F}\mathrm{o}\right)\times 3.6$$

(4)

Thereafter, the vehicle image is captured, compressed, and encoded in Base64 format. A JSON object is then constructed to include both the image and vehicle speed data as fields of the same record. The record is temporally stored in the SD within the node. Finally, the JSON object is transmitted as a payload to the cloud server with the help of HTTP. The algorithm for implementing this subsystem is presented in Algorithm 1.

Algorithm 1: Algorithm for the IoT subsystem

Input: Radar Doppler shift frequency (Fd), timestamp, captured image

Output: JSON object transmitted to the cloud server

1. Initialize the radar sensor and camera module.

2. Continuously monitor radar sensor for Doppler shift frequency (Fd).

3. If a vehicle is detected:

a. Record timestamp

b. Compute speed (Vc) using the Doppler effect formula

c. Capture vehicle image.

d. Compress and encode the image in Base64 format.

e. Construct a JSON object:

{

“timestamp”: <time>,

“speed”: <V>,

“image”: <Base64_image>

}

f. Store JSON record temporarily on the SD card.

g. Establish an HTTP connection and transmit the JSON object as a payload to the cloud server.

4. End.

With these sensor options, lightweight and low-cost IoT nodes can be developed, due to the affordability of sensors and microcontrollers. For data communication, both cellular and Wi-Fi technologies are viable choices. While Wi-Fi is widely available in urban areas, cellular networks offer broader coverage, including remote locations. Cellular IoT is preferred as it operates on LPWAN technology based on GSM, the most widely deployed wireless communication system globally, making it suitable for remote areas. Other GSM-based technologies can be utilized when power consumption is not a challenge. In urban environments, Wi-Fi remains a feasible option. Multiple communication technologies can also be integrated across different system layers for enhanced connectivity.

In the cloud subsystem, there should be an API that listens for HTTP POST requests. Then, the server receives the JSON object, decodes the Base64 image, and decompresses and stores the image and speed data in the database for further processing. The photo needs to be processed by performing Automatic Number Plate Recognition (ANPR), also known as Automatic License Plate Recognition (ALPR) technology. This is an AI-based technology for image capturing, detecting, extracting, and reading number plates from the image [153]. It works by first detecting the license plate from the image. Thereafter, characters on the image are extracted using Optical Character Recognition (OCR) technology, which recognizes characters and numbers for further utilization. A standard tool for the implementation of ANPR is an Open-Source Computer Vision (OpenCV) and ML software library (such as OpenCV-4.11.0) that supports programming languages, mainly C++, Java, MATLAB, and Python; the main operating systems are Android, Linux, Mac OS, and Windows [154]. During the process, the image is first converted to grayscale. Then, bilateral filters and edge detection algorithms are applied to extract the number plate region, and the rest of the image is discarded. The number plate region is further processed to mine the details (numbers and characters) using tools such as the Google OCR algorithm, Google Vision API, and Plate Recognizer API. These details are then used to detect vehicle number plates [22]. Thereafter, speed and vehicle number plate information is sent to the blockchain via the HTTP protocol. Algorithm 2 outlines the generic procedure for implementing the cloud computing subsystem.

Algorithm 2: Algorithm for cloud subsystem

Input: JSON object {timestamp, speed, image}

Output: vehicle number plate, imageHash, new JSON object sent to Blockchain.

1. Start the API server to listen for HTTP POST requests.

2. Receive incoming JSON objects from the IoT subsystem.

3. Extract data fields: timestamp, speed, and Base64 image.

4. Convert base64 image to hash (SHA-256)

5. Decode the Base64 image and decompress it.

6. Store image and imageHash in the cloud database.

7. Perform ANPR on the stored image:

a. Convert the image to grayscale.

b. Apply edge detection and contour analysis.

c. Extract number plate region and process characters using OCR.

d. Store the extracted vehicle number in the database.

8. Construct a JSON object:

{

“timestamp”: <time>,

“vehicle_number”: <plate_number>,

“speed”: <speed>

“image”: <imageHash>

}

9. Send this data to the Hyperledger Fabric blockchain via the HTTPS.

10. If a transaction is successful, return a confirmation response.

11. If a failure occurs, retry or log the error.

12. End.

The blockchain subsystem is responsible for the distributed and transparent data storage across stakeholders’ premises to avoid forgery, corruption, and favoritism. The consortium blockchain network can be implemented using blockchain platforms such as Hyperledger Fabric (HLF) and Ethereum. Examples of stakeholders forming the consortium are the National Road Safety Council (NRSC), the Land Transport Regulatory Authority (LATRA), an authority for regulating land transportation of passengers and goods, the traffic police force (TPF), the Revenue Authority (RA), responsible for motor vehicle registration, and the National Roads Agency (NRA) responsible for managing national roads. These stakeholders need vehicle speed information to enhance the performance of their respective duties. Algorithm 3 presents the corresponding algorithm for this subsystem. The assumption is that the blockchain network has already been set up and started. The smart contract for storing data in the blockchain is already developed and deployed, and there is an HTTP connection between the cloud server and the blockchain network.

Algorithm 3: Algorithm for blockchain subsystem

1. Wait for incoming vehicle speed data from the cloud server

2. On receiving a new speed record:

a. Extract timestamp, vehicle number, speed, and image hash

b. Construct a transaction payload

c. Submit a transaction to the blockchain:

i. Invoke the smart contract function (recordVehicleSpeed)

ii. Pass arguments: timestamp, vehicle number, speed, image hash

d. Get transaction ID and confirmation from the blockchain

e. Log transaction ID for verification

3. Return transaction ID to the cloud system

4. Repeat for new incoming speed data

The application subsystem is implemented by each stakeholder separately depending on their distinct business requirements. The implementation can either be in the form of a mobile or web application or both. For example, mobile applications can be used by field officers such as traffic officers, drivers, and other enforcement officers. Web applications can be used in the office environment to analyze, generate, and visualize different reports. Algorithm 4 presents the algorithm-specific for implementing the speed law enforcement applications.

Algorithm 4: Algorithm for the police application subsystem

Use a police application to query the blockchain database for specific vehicle speed data.        def query_blockchain(vehicle_id):        vehicle_data = query_blockchain_data(vehicle_id)        return vehicle_data Check Speed Status: Analyse the speed data retrieved from the blockchain to determine whether the speed is normal or overspeed.        def analyze_speed_status (speed):           if speed > SPEED_LIMIT:              return “overspeed”               return “normal” Record Speed Status: Based on the analysis, record whether the vehicle was overspeeding or within normal limits.        def record_speed_status (vehicle_id, status):        save_speed_status (vehicle_id, status) Visualize Data: Present the vehicle speed, speed status, and other related data visually through the application interface, as stored in the blockchain.        def visualize_data(vehicle_data):        plot_vehicle_data(vehicle_data)

5.2. The Proposed System Proof-of-Concept Demonstration

We employed a combination of simulation and experimental approaches to establish a proof-of-concept demonstration, as illustrated in Figure 4. In this setup, the client computer, which emulates the IoT subsystem, sends data to the cloud server via an HTTP request. The cloud server then forwards the request to the blockchain server. Upon processing the request, the blockchain server generates an HTTP response, which is relayed back to the IoT subsystem through the cloud server.

The simulation approach provided data for integration verification, while experimentation validated the proposed cloud and blockchain algorithms. This demonstration architecture validates the overall conceptual design of the proposed system. To implement the functionalities of the IoT subsystem, we simulated about 10,000 data records, each containing number plate, speed, and timestamp fields. Figure 5a shows a portion of the generated sample data in CSV format, while Figure 5b presents the corresponding JSON version prepared for transmission to the cloud.

The data were designed to realistically reflect Tanzanian highway conditions, with speed values ranging from 20 km/h to 180 km/h and vehicle number plates conforming to Tanzanian formats (for both normal and special number plates). A Python script (developed using Python 3.13.2) was implemented with two main functionalities: first, to generate realistic vehicle speed data along with corresponding number plates, and second, to transmit the data to a cloud server via HTTP. The script was deployed on a client computer, effectively emulating the IoT subsystem functionality of real-time data collection and transmission to the cloud server. At the cloud, we implemented a RESTful API (FastAPI). Figure 6 shows the Swagger documentation of the API.

The API receives data from the client computer via the NGINX web server, stores data in the SQLite database, and forwards the data to the blockchain network. The cloud-based functionalities are hosted in the DigitalOcean cloud computing platform. Figure 7 demonstrates a sample of query results in the SQLite database, verifying successful cloud data delivery.

To demonstrate the functionality of the blockchain subsystem, we set up an HLF-based consortium blockchain network with two organizations and a single channel. A chaincode (smart contract) was developed and deployed on the network using JavaScript to facilitate data storage. We utilized the Fabric application gateway [155] to receive data from the cloud subsystem via the NGINX web server, which was implemented using the Express.js framework to listen for incoming HTTP requests. Upon receiving a request, the gateway triggers the chaincode (smart contract), which then executes a function to store data in the blockchain. The blockchain network was deployed on a Digital Ocean cloud server. Figure 8a,b show how the two organizations (ORG_1 and ORG_2) received the synchronized copy of the data from the cloud subsystem.

The core objective of this demonstration was to prove the concept of integrating IoT, cloud computing, and blockchain technologies for implementing vehicle speed detection systems. Therefore, we focused more on technology integration rather than on data collection. As a result, we adopted a hybrid approach involving experimentation and simulation to illustrate the practical implementation of this integration. The experimental component validated the proposed algorithms for implementing the integration, while the simulation provided the data necessary to evaluate the integration process.

5.3. Performance Evaluation of the Proof-of-Concept Demonstration

To evaluate the performance of the algorithms implemented in the proof-of-concept demonstrator, we conducted two experiments to analyze the impact of payload size (PS) and number of nodes (NN) on latency. These experiments were performed independently to isolate the effects of PS and NN on overall system performance, specifically measuring the time taken to transmit data from the client machine to the blockchain system. The experiments were executed from a client computer equipped with an AMD Ryzen 3 PRO 2300U CPU (2 GHz) and 12 GB RAM running a client script that sent requests to a server with a network bandwidth of 5.4 Mbps (download) and 9.6 Mbps (upload). The cloud subsystem was deployed on a Digital Ocean cloud platform with a basic droplet instance, featuring 1 virtual CPU(vCPU) and 1 GB of RAM.

5.3.1. Experiment 1: How Payload Size Affects Latency

In this experiment, we fixed the NN to twelve (12), representing the assumed maximum number of nodes in the selected road section. We then varied the PS, which is determined by the number of records (NR) transmitted in a single batch. Each record corresponds to the speed data of one vehicle and includes fields such as vehicle speed, license plate number, and timestamp.

To analyze the impact of NR on latency, we varied the number of records per transmission as follows: 5, 10, 20, 40, 60, 80, and 100. While, ideally, each vehicle’s data are captured and sent individually in real-time, we considered different scenarios:

• NR = 5: represents the typical transmission scenario, where up to five vehicles’ data are sent at a time.
• NR = 10: simulates data transmission after a brief network outage, where accumulated records are sent in bulk.
• NR = 100: represents the worst-case scenario, where data are transmitted after a prolonged outage.

The experiment was automated using a Python script that logged timestamps when data were sent from the client subsystem and when they were received in the blockchain subsystem. The difference between these timestamps was used to compute latency, following the sequential steps outlined here.

• Step 1: Computation of latency for each node i and record j using Equation (5).

  $${\mathrm{L}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{y}}_{ij}{=\mathrm{T}}_{\mathrm{blockchain}\text{-}\mathrm{receive},i,j}{-\mathrm{T}}_{\mathrm{client}\text{-}\mathrm{sent},i,j}$$

  (5)

  where
  • T_{client-sent,i,j} = timestamp when node i sends record j
  • T_{blockchain-receive,i,j} = Timestamp when blockchain received record j from node i
• Step 2: Computation of average latency per node for a given NR using Equation (5).

  $$\mathrm{A}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{L}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{y}\mathrm{f}\mathrm{o}\mathrm{r}{\mathrm{Node}}_i={\sum }_{j=1}^{\mathrm{NR}}{\displaystyle \frac{{\mathrm{L}}_{i,j}}{\mathrm{NR}}}$$

  (6)
• Step 3: Computation of overall average latency for the 12 NN with the help of Equation (5).

  $$\mathrm{A}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{L}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{y}\mathrm{f}\mathrm{o}\mathrm{r}\mathrm{N}\mathrm{R}={\sum }_{i=1}^{12}{\displaystyle \frac{\mathrm{latency}\mathrm{of}{\mathrm{Node}}_i}{12}}$$

  (7)
• Step 4: Finally, the conversion of latency computed in Equation (7) into milliseconds:

  $$\mathrm{L}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{y}\left(\mathrm{m}\mathrm{s}\right)=\mathrm{L}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{y}\left(\mathrm{s}\mathrm{e}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{d}\mathrm{s}\right)\times 1000$$

  (8)

The equations were implemented in Python using nested loops to simulate a scenario where twelve (12) nodes transmitted records over five (5) rounds of trials for each NR. The client script was executed on a desktop computer, generating a CSV file containing the overall averaged results. These results were computed using Equations (6) and (7) and are visually presented in Figure 9.

The results indicate that the highest observed latency in the worst-case scenario remained below one (1) second. Additionally, the findings suggest a near-direct proportional relationship between NR and latency. Figure 10 graphically represents this trend, showing how latency increases with the number of records per the fixed 12 NN.

5.3.2. Experiment 2: How the Number of Nodes Affects Latency

In this experiment, we kept the NR fixed at 100, assuming a worst-case scenario, and increased the NN up to a maximum of 18 instead of 12, to allow for potential future expansion of the NN. We varied the NN in increments of 1, 3, 6, 9, 12, 15, and 18 to study how latency is influenced by changes in NN. A Python script was used to automate the experiment, applying the formula for calculating the average latency for 100 NR in each of the NN values, as outlined in Equation (9). The resulting latency as visually presented in Figure 11 was then converted into milliseconds using Equation (8).

$$\mathrm{L}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{y}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{N}\mathrm{N}={\sum }_{j=1}^{100}\left({\displaystyle \frac{{\mathrm{T}}_{\mathrm{b}\mathrm{l}\mathrm{o}\mathrm{c}\mathrm{k}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{i}\mathrm{n}\text{-}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{e}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{d},j}-{\mathrm{T}}_{\mathrm{c}\mathrm{l}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{t}\text{-}\mathrm{s}\mathrm{e}\mathrm{n}\mathrm{t},j}}{100}}\right)$$

(9)

where

• T_{blockcchain-received,j} = timestamp when record j is sent by the client;
• T_{client-sent,j} = timestamp when record j is received by the blockchain.

Figure 11. Experiment 2 result data.

Figure 11. Experiment 2 result data.

The results indicate that latency increases as the number of nodes (NN) grows. With just one node, it takes less than a second to send 100 records, whereas with 18 nodes, it takes approximately six (6) seconds to send 100 records from each node. The relationship between latency and NN appears to be near directly proportional, except for two outliers at 12 NN and 15 NN, as shown graphically in Figure 12.

We conducted two experiments to evaluate the performance of the proposed integration by analyzing how the number of records (NR) and number of nodes (NN) individually impact system latency. The results demonstrate promising performance, even in worst-case scenarios, while employing a sequential execution approach under limited server resources. For instance, in Experiment 1, where 12 nodes each transmitted 100 records, the total transmission time remained under 1 s to reach the blockchain. Similarly, in Experiment 2, transmitting 100 records from 18 nodes to the same server requires approximately 6 s. These findings are based on a proof-of-concept demonstrator tested with a limited scope of 18 nodes and 100 records representing a selected road section. The results demonstrate the feasibility of integrating these technologies to develop a smart and transparent vehicle speed detection system. Moreover, they provide valuable insights for future advancements toward a fully functional system that can accommodate larger road networks with hundreds of nodes and thousands of records. In this study, a sequential execution approach was used to handle client requests, relying on nested loops. However, for future implementations, it is recommended to explore asynchronous and parallel processing approaches, which are more efficient for handling concurrent requests. Additionally, other performance metrics, such as failure rate and throughput, should be considered to further evaluate and optimize the system’s performance.

6. Conclusions and Future Works

Road accidents are becoming more frequent and often lead to fatalities and injuries, mainly due to overspeeding. Although speed limit laws are in place to control driving speed, efficient enforcement is essential. Vehicle speed data are critical for enforcing these regulations and enhancing overall road traffic management. This highlights the importance of vehicle speed data in intelligent transportation systems and the broader context of smart cities.

Road-based vehicle speed detection systems play a crucial role in collecting vehicle speed data. Their effectiveness depends on factors such as efficiency, transparency, and cost-effectiveness. However, the current road-based methods for vehicle speed monitoring in Tanzania are inefficient and manually operated. These methods are also not transparent and exposed to favoritism, forgery, and corruption. Not only this, but they are also limited by time and weather conditions and depend on traffic police officer commitment. On the other hand, the off-the-shelf technological solutions are expensive to acquire and maintain, causing them to be out of reach to Tanzania. As a result, vehicle speed detection is dominated by manual systems in the country and in other developing nations at large.

The concept of automating these systems is vital. Several studies have proposed and implemented various automatic methods. However, vehicle speed detection remains a challenge that needs to be addressed further due to the limitations of the existing systems. On the other side, improvement of these systems is necessary to ensure continuous effectiveness, an opportunity brought by emerging technologies. This study proposed and conceptualized a smart and transparent system by leveraging emerging technologies to improve vehicle speed data collection in Tanzania.

The unique feature of the proposed solution is the distributed storage of data for ensuring transparency, an essential functionality for addressing forgery, favoritism, and corruption problems. Additionally, the system offers improved accuracy and evidence-based data by leveraging radar- and vision-based methods. Other distinguishing features include low-cost and simple design achieved through an innovative approach of using cloud computing to interface Internet of Things nodes with the resource-intensive blockchain layer. This approach enables the use of lightweight and low-cost Internet of Things nodes to interact with the blockchain instead of requiring multiple costly edge or full nodes.

On the other hand, a key significance of the proposed system is its capability to collect speed data that can serve multiple purposes across various organizations. In Tanzania, different organizations rely on speed data for different purposes. Traffic police use them to enforce speed limits, while the Road Safety Council leverages high-level speed information to oversee traffic police operations, investigate accidents, advise the government on road safety policies, raise public awareness, and promote road safety research. Bus and truck owners utilize speed data to monitor driver behavior and ensure compliance with safety regulations. Likewise, the National Road Agency employs speed data for designing safer roads, conducting road safety audits, analyzing driving patterns to optimize traffic signs and speed limit mechanisms, and collaborating with other entities in accident investigations. Additionally, bus drivers find speed data valuable as evidence when resolving disputes with traffic officers, particularly in cases where they are accused of speeding.

Not only this, but this work has also laid the groundwork that provides both theoretical and practical insights into smart and transparent vehicle speed detection systems. These insights are expected to contribute significantly to addressing the weaknesses of existing systems and in meeting the cost and environmental requirements of developing countries. This work examined the methods used for detecting vehicle speed, the technologies enabling these methods, and the systems currently implemented for vehicle speed detection. Thereafter, this work proposed and conceptualized an implementation of a smart and transparent vehicle speed detection system by integrating the Internet of Things, Artificial Intelligence (alongside cloud computing), and blockchain technologies. We validated the feasibility of our proposal through a proof-of-concept demonstration, showing that these technologies can be integrated into a comprehensive speed detection system. Research questions, findings, and contributions are summarized in

Table 6. Research questions, answers, and contributions.

S/n	Research Question	Research Answers	Contributions
i.	What are the strengths and weaknesses of current vehicle speed detection methods?	Section 2 provides a detailed discussion of the operating principles, strengths, and weaknesses of vehicle speed detection methods; Figure 2 illustrates the taxonomy of vehicle speed detection techniques. Table 2 and Table 3 summarize road-based intrusive and non-intrusive methods, respectively, while Table 4 presents an overview of vehicle-based speed detection systems. These summaries serve as a reference for selecting appropriate speed detection methods.	Summarized guidelines for selecting road-based and vehicle speed detection methods. Taxonomy of vehicle speed measurement methods.
ii.	Which technologies can be used to implement a smart and transparent vehicle speed detection system?	An innovative integration of IoT and AI within cloud computing is proposed to enable system interconnection and intelligent functionalities, while blockchain is introduced to enhance data security and transparency, as discussed in Section 3.	Proposed technology integration for implementing smart and transparent vehicle speed detection systems.
iii.	What are the strengths and weaknesses of existing vehicle speed detection systems?	Accuracy concerns, lack of transparency, and high costs are the primary weaknesses of existing systems. These issues stem from the reliance on expensive, immature, and unstable speed measurement methods, and centralized cloud computing for data processing and storage, as detailed in Section 4.	A comparison of existing systems with the proposed systems highlights the research gap in ensuring accuracy, transparency, and cost-effectiveness by integrating emerging technologies and utilizing more accurate and reliable sensing technologies.
iv	How can the weakness of the existing vehicle speed detection systems be addressed to implement a system compatible with developing countries?	The Internet of Things is proposed for data collection and transmission, AI techniques in cloud computing for data pre-processing, and blockchain technology for security and transparency. Application layer for application-specific processing and sharing.	A validated four-layer conceptual framework for implementing a vehicle speed detection system. Algorithms for subsystem functionality implementation and integration.

.

However, the integration of these technologies is a complex task and remains an area that requires further exploration. This highlights the need for further research on integrating emerging technologies for enhancing vehicle speed detection. Specifically, this study identifies the necessity for future research to extend beyond the scope of this work by developing fully functional proof-of-concept and real-environment prototypes.