Data Sources for Traffic Analysis in Urban Canyons—The Comprehensive Literature Review

Abstract

We propose a comprehensive literature review based on big data and V2X research to find promising tools to detect vehicles for traffic research and provide safe autonomous vehicle (AV) traffic. Presented data sources can provide real-time data for V2X systems and offline databases from VATnets for micro- and macro-modeling in traffic research. The authors want to present a set of sources that are not based on GNSS and other systems that could be interrupted by high-rise buildings and dense smart city infrastructure, as well as review of big data sources in traffic modeling that can be useful in future traffic research. Both reviews findings are summarized in tables at the end of the review sections of the paper. The authors added propositions in the form of two hypotheses on how traffic models can obtain data in the urban canyon connected environment scenario. The first hypothesis uses Roadside Units (RSUs) to retrieve data in similar ways to cellular data in traffic research and proves that this source is data rich. The second one acknowledges Bluetooth/Wi-Fi scanners’ research potential in V2X environments.

1. Introduction

Today, dense urban areas are recognized as tall buildings with minimal distance between the frontages. Such a way of urban planning reminds one of canyons. Buildings can interfere with the coverage of GNSS and BTSs (Base Transceiver Stations) of streets. However, as smart cities develop with IoT devices, the devices used by the city infrastructure can disturb itself on the same frequencies or spectrums, which the researchers consequently acknowledge. A wide variety of V2X and IoT-based systems used in cities can provide valuable data for transport modeling. The pinpoint positioning of autonomous vehicles can provide sufficient data for micro-modeling. Therefore, models will be used to further improve the capacity of the infrastructure. In Section 5 we proposed two hypotheses based on the review. Both hypotheses consider the usage of a connected environment.

In the review, the recently commonly used IoT technologies described and proposed in the literature are reviewed, such as Bluetooth, DSRC, IEEE 802.11p WAVE, Cellular V2X (considering usage of LTE, 4G, 5G), and Wi-Fi, as well as state-of-the-art technologies that are not in use in transportation modeling nowadays, however, with research potential such as ZigBee and LoRa, proposed in state-of-the-art literature. The technology selection does not include every technology but considers technologies with promising capabilities for big data research in transport modeling.

The purpose of the review in Section 4 is to find patterns in big data research for traffic modeling purposes; the data sources reviewed are Bluetooth, Wi-Fi, Cellular Data, GNSS data, Taxi data, E-scooter fleets, and Shared Bike fleet data.

V2X can consider many forms of communication, including Vehicle-to-Grid (V2G), Vehicle-to-Home (V2H), Vehicle-to-Building (V2B), Vehicle-to-Infrastructure (V2I), Vehicle-to-Load (V2L), and Vehicle-to-Vehicle (V2V) [1]. The focus of this paper is placed mainly on V2X itself, V2V, and V2I as promising sources for traffic engineering and transport modeling. Due to the number of abbreviations used in the paper, the abbreviations used in the paper are given directly in the text and also listed in the Abbreviations section at the end of the paper.

2. The Review Methodology and Structure

2.1. Meaning of Nomenclature in the Review

Firstly, a precise definition of the nomenclature used for the review should be made. Some terms used could be wrongly understood from a different point of view, as the review considers multidisciplinary research. An Autonomous Vehicle (AV) refers to a fully automated vehicle, according to the Society of Automotive Engineers (SAE), including level 5 [2], German BASt (Bundesanstalt für Straßenwesen—Federal Road Office)—level 5 [3], for NHTSA (National Highway Traffic Safety Administration), also corresponding level 5 [4]. As a traffic model, the micro and macro models are under consideration.

2.2. Selection Procedure and Classification of Articles

As mentioned in Section 1, not all of the potential technologies are listed in this paper. As traffic engineers, the authors focused on technologies known to have potential in traffic engineering or that are promising data sources for transport modeling in the future.

Keywords used for the review in Section 3:

• v2x Bluetooth
• v2i Bluetooth
• v2x ZigBee
• v2x urban canyon
• Wi-fi urban canyon v2x
• LoRa urban canyon v2x
• DSRC v2x
• IEEE 802.11p urban canyon v2x
• Cellular v2x
• LoRa v2x

Keywords used for the review in Section 4:

• Cellular data record big data transportation
• Cellular data record OD matrix
• Bluetooth big data transportation
• Bluetooth OD matrix
• Wi-Fi big data transportation
• Wi-Fi OD matrix
• Wi-Fi big data transportation
• Wi-Fi OD matrix
• GPS big data transportation
• GPS OD matrix
• Taxi big data transportation
• Taxi OD matrix
• E-scooter big data transportation
• E-scooter OD matrix
• Bike big data transportation
• Bike OD matrix

2.3. Publication by Year

The paper includes publications from 2009 to 2025. Figure 1 shows a chart showing the number of articles by year included in the review. The authors’ goal was to achieve a review of state-of-the-art technologies, and the structure of the publication years indicates that. The most publications were from the year 2024—20 positions, and 2020—19. The approximately two-and-a-half-year period, as at the time of writing this paragraph, is August 2025 up to 2023, consisting of 30.23% of the articles. The almost five-year publishing period from 2020 to 2025 consists of 62,02% of the articles. Figure 1 shows publication number in the horizontal axis and the years published in the vertical axis.

2.4. Paper Classification by Research Method

In this subsection, research is divided into four groups: experimental papers that concern real-world text, papers that use simulations to evaluate proposed V2X-related methods, papers that are related to big data research in traffic engineering, and the fourth group—others that are not suitable for classification to previous groups, such as databases, law documents, other types of documents that are not scientific work, web pages, and one study applicable to the discussed technology, without V2X purpose. For the first group, “Experimental V2X” articles with field tests among the simulations presented are included. The numbers of all groups are presented in Figure 2. The highest publication number is in the third group that includes big data traffic research, with a number of 52. However, the first two groups concerning V2X research consist of 71 papers.

2.5. The Review Methodology

The review methodology was based on keywords listed in Section 2.2; the search engine used was mostly Google Scholar as it gave much better results than the authors’ university finding tools. The parameters used was “Sort by relevance”, “Any type”, “Include citations”. However, in case of older papers found, the option “Since 2021” was used. Paper language was restricted to only English-language publications.

2.6. Structure of the Paper

The paper has six sections. The comprehensive review evolves from V2X technology solutions and data acquisition by big data traffic research in cities to form hypotheses about the future of traffic modeling in urban canyon connected environments using data-rich V2X systems. Section 1 is an introduction about the research area. Section 2 covers the methodology and structure of the review and papers. Section 3 is a review about IoT technologies used, and proposed, in V2X systems with focus on urban canyon application and obstacles. Section 4 is a review about traffic modeling big data solutions. Both Section 3 and Section 4 end with tables summarizing the technologies. Section 5 covers two hypotheses about traffic modeling future applications supported by V2X system data analysis. Section 6 is the conclusion of the review.

3. Recently Used IoT Technologies Capable of V2X Systems in Urban Canyons

Historical trends in the automotive industry have predicted rapid growth in the percentage of AVs in traffic [5]. Therefore, to ensure safe and fluent traffic, many technologies are proposed in the literature for V2X purposes. In this review, and reviews in the following sections of the paper, the focus is put not only on types of communication and system characteristics, but also on how the technologies perform in the urban canyons. For example, a real-world simulation performed in [6] prove that GNSS-based systems are not suitable for urban canyons. The study recognized a drop in the Position Dilution of Precision, higher error, and multipath bias in urban canyon scenarios, sharply reducing the effectiveness of V2X. Ref. [7] has the same conclusion, also supported by RTK (Real-Time Kinematic) tests, by using it in open road scenarios—the mean horizontal error was 0.49 m and maximum horizontal error 0.83 m—and in urban canyon scenarios, where values increased to 1.46 m and 3.68 m. That could be enough for lane-level accuracy. However, e.g., under the viaducts, the position error could be several times higher, making a GNSS-only-based solution inadequate for urban canyons. In [8], ray-tracing-based 3DMA GNSS and factor graph optimization reduced positioning error from >30 m to <10 m in urban canyons; however, it states that standard double difference methods cannot eliminate location-specific multipath/NLOS errors. The PhD dissertation in [9] proposes Ultra-Wideband ranging for GNSS, resulting in a horizontal error of 0.6 m to 1.1 m in an urban canyon scenario. Finally, ref. [10] summarizes that GNSS technologies are not suitable for urban canyons and proposes other solutions, such as V2X-based localization, waveform-based positioning, sharing one’s own in-vehicle GNSS measurements, inertial data, and V2X-observed measurements for improving position accuracy or usage of passive RF (Radio Frequency) backscatter tags on infrastructure. At the end of the section, findings are summarized in

Table 1. Summary of Section 3 findings about each technology.

Technology	Strengths	Weaknesses	Range	Data Rate	Latency	Cost	Scalability	Urban Canyon Suitability	Maturity	Research Trends/Gaps
Bluetooth	Low energy, adaptive frequency hopping, good short-range performance	Limited range, low throughput at distance	~300 m static, ~1 km dynamic	Up to ~992 kbps	~100 ms	Low	Medium	Moderate—short range avoids interference	Mature	Integration with smartphones, hybrid V2X systems
Wi-Fi	High bandwidth, widely available	Poor NLOS performance, interference in urban areas	~40–100 m in NLOS	High	Variable, often >100 ms	Medium	Medium	Limited—signal attenuation, interference	Mature	Beacon stuffing, GNSS support, hybrid systems
ZigBee	Low power, multi-hop support, good short-range accuracy	Low bandwidth, interference on 2.4 GHz	~300–600 m	Low	Moderate	Low	Medium	Moderate—suitable for short-range V2X	Mature	Hybrid with LoRa, GNSS-free localization
LoRa	Long range, low power, good NLOS performance	Low data rate, speed limitations	~160–500 m (urban), >10 km (UAV)	Very low	High	Very Low	High	Good—especially for non-safety apps	Emerging	SF tuning for urban NLOS, UAV applications
LTE (4G)	Reliable, widely deployed	Performance drops in dense urban canyons	~800 m+	Medium	~10–50 ms	Medium	High	Moderate—LOS helps, but reflections degrade	Mature	Fusion with INS, V2I/V2V hybrid systems
5G/NR-V2X	High speed, low latency, AI optimization, mmWave support	Needs dense infrastructure, blockage sensitivity	~200 m (mmWave), varies	Up to 25 Mbps uplink	~5–10 ms	High	High	Good—with extra power and beamforming	Emerging	Deep learning for NLOS detection, private 5G for SAE L3+
C-V2X	High reliability, GNSS-independent localization, LiDAR fusion	Signal loss in deep urban canyons	~150 m (RSU-based)	High	<10 ms	High	High	Strong—RSU-based systems perform well	Emerging	RSU density optimization, hybrid with DSRC
DSRC/IEEE 802.11p	Low latency, proven V2V/V2I use	Blockage and reflection issues in urban canyons	~300 m	Medium	<100 ms	Medium	Medium	Moderate—needs RSU density increase	Mature	mmWave integration, beam-forming, digital twin modeling

.

The first technology searched in the review is Bluetooth. It is proposed as a safe and proven low-range communication technology [11]. Bluetooth version 5 allowed consideration of the use of V2X. According to [12], the version provided twice the speed, four times the higher range, eight times the advertising capacity, and was optimized for low energy. The low range of Bluetooth could be a disadvantage of the proposed communication system. However, in urban canyons, long-range signals could be disturbed by objects and tall buildings. The disadvantage is less serious, and Bluetooth can be considered. Decentralized, ad hoc networks known as piconets, which operate across 79 distinct frequencies, are resistant to interference from dense urban radio signals. According to [13], because of Adaptive Frequency Hopping, which avoids interfered channels, only half of the 79 channels are used in classical Bluetooth, and the authors proposed a method reducing occupancy by up to 50%. Bluetooth (BT) on its blog [14] shows readiness for applications in the V2X environment. State-of-the-art literature tested BT in V2X applications and estimated a range of 300 m used statically and in real-world highway scenarios around 1 km while driving [15]; however, this was done with degradation of throughput values, from 992.5 kbps at 0 m to 128.9 kbps at 100 m. In a further study [16], effective speed for sending and receiving messages is only 20 kmph using a stationary roadside device located 1 m above the ground, with a mobile device inside the vehicle. In [17], experimental results from dynamic testing demonstrate that reliable data reception is achievable at velocities up to 70 km/h, primarily due to the high frequency of advertising transmissions. Bluetooth is also proposed as a part of V2X systems [18] as a way of transferring messages from OBU to apps in smartphones of road users, where for safety-critical messages the DSRC is proposed. An open-source simulation platform using DSRC based on IEEE 802.11bd is proposed in [19]. The platform recognizes urban canyons in the detailed high-frequency model 3GPP NLOS. Another form of DSRC communication is between vehicles and pedestrians’ smartphones. In a simulated scenario [20], the system provides a Collision Detection Algorithm with safe delays (under 100 ms) even under congested scenarios.

Initial research proposed 5G as a novel technological standard for cellular data transmission. Currently, 5G is widely developed in many countries, with significant improvement over the previous 4G standard for LTE technologies. An interesting finding is in study [21], where LTE and 5G were compared in an intersection; 5G had a lower packet delivery ratio in longer distances—over 800 m, due to signal strength issues. With sufficient infrastructure, LTE can match 5G performance. However, ref. [22] shows that NR-V2X performs with a 20% better packet reception ratio and has up to 95% more power consumption reduction. The dissertation in [23], aiming to optimize 5G usage, proposed a K-means clustering algorithm for a novel distance metric for segregating road users, an allocation scheme called Vehicular Frequency Reuse that significantly reduces handover rates (over 99% reduction) and enhances link reliability for high-speed road users. Also, the dissertation proposes AI-based systems to improve uplink transmission while minimizing power consumption. One advantage of 5G technology is its ability to operate in underutilized frequency bands, such as those around 3 GHz, which are less congested compared to bands commonly used by Bluetooth, Wi-Fi, and ZigBee. Study [24] proposes antennas operating at 400 MHz channel bandwidth at a practical range of ~200 m at 28/38 GHz. The proposed system can support up to 1 Mbps downlink and 25 Mbps uplink for vehicle speeds up to 250 km/h. Guidelines and architecture for 5G-based V2X in [25] proposed a requirement of 10 ms latency (only 5 ms for remote driving). Although 5G is reliable in urban canyon scenarios, it needs up to an additional 3 dB extra transmitter power [26]. Real-world tests conducted in [27] showed blockages impact 5G signal (at 28 GHz) with up to 7.5 dB loss for one car blockage, 2 dB loss through clear windows, but 15 dB through sun-protective film; reflection from walls adds only about 3 dB loss, and diffraction and foliage can cause loss over 20 dB. Three-dimensional ray-tracing matches measurement within ~4 dB. NLOS can be effectively detected using a high-accuracy deep learning model [28]. The state-of-the-art model achieved 93.5% accuracy in distinguishing between NLOS and LOS conditions, surpassing the performance of earlier approaches. Additionally, 5G technology can complement GNSS-based positioning in urban canyon environments. Paper [29] introduces a hybrid approach incorporating Kalman filters, multi-rate data fusion, and mmWave technology. The study concludes that for SAE Level 3 and higher, the deployment of private 5G networks is essential.

In [30], researchers focus on Cellular V2X strictly in an urban canyon scenario, forming them by containers. The empirical test showed severe packet loss occurring below −110 dBm received power in tested scenarios—vehicle behind a container, and containers representing an open-topped alleyway of the urban canyon. A deep learning approach is used in [31], with K-means for addressing variable fading channel conditions and a Genetic Algorithm for optimizing two parameters: number of reflection points and reflection coefficient to match the model for actual field measurements. A comparative analysis between low and high urban canyon environments revealed that the low urban canyon exhibits a greater number of reflection points and a higher reflection coefficient. In [32], urban canyons were classified in four groups: low—with a building height of 1 to 20 m, medium—from 20 to 60 m, high—over 60 m, and tall—over 100 m buildings. The simulation proves that the higher the buildings, the more severe the multipath effects, signal reflections, and greater shadowing are. This finding is also present in [33], with in-chamber tests supported by field measurements. The study provided a 3D geometric-based irregular channel model that reflects urban conditions. An independent GNSS-based CV2X-LOCA system is proposed in [34], maintaining less than 4 m accuracy in real-world scenarios even in tunnels, beside skyscrapers, and under elevated roads. The system is based on the RSU every 120–150 m. The simulation experiment showed accuracy up to 2.5 m, even in the high-speed (125 km/h) scenario. An example of a C-V2X application could be the Multi-Sensor LiDAR [35], where the system aggregates and fuses point clouds from LiDAR sensors in multiple vehicles using machine learning for pedestrian detection. For the urban canyon scenario, a comparison study [36] proposes a combination of C-V2X and DSRC. The review of existing technologies in this paper notes that most commercial devices support only one communication technology, and from the urban canyon point of consideration, all use GNSS.

For V2I in [37], Cellular V2X (4G and 5G) was used, and for V2V, DSRC was used in urban canyon scenarios. The study employed a fusion of Cellular V2I, V2V, and INS as an alternative to GPS, achieving robust and accurate positioning. The proposed method demonstrated an 18.7% improvement in accuracy compared to the state-of-the-art solutions available at the time of publication. An interesting finding in [38] concludes that due to expected strong line-of-sight (high K-factor—7.46 dB in the case), long delay spread (of 63.8 ns), and frequency diversity, the urban canyon scenario could potentially lead to an improvement in performance of the system. This finding is also supported by [39], where an LTE link was tested. In the urban canyon scenario, LTE had stronger line-of-sight and less frequency dispersion in comparison to suburban and highway environments. However, as noted in [40], a testbed study of LTE V2V communication in urban canyon environments with varying building heights revealed that denser and taller canyons exhibit increased signal reflectivity. Consequently, signal attenuation and path loss are more pronounced, negatively impacting V2V performance metrics.

For the IEEE 802.11p WAVE standard simulation test, the researchers in [41] tested two scenarios, WAVE-CH07 (Urban Canyon, LOS) and WAVE-CH08 (Urban Canyon, NLOS), and proposed a new channel estimator (LSA-DD) that improves robustness by up to 3 dB SNR compared to prior methods. Reference [42] concluded research in three cities, mainly on the 60 GHz band; the study found that urban canyons yields high RMS delay spreads (up to 80 ns) due to strong, persistent reflections, where open streets with trees caused much less reverberation. In [43], the digital twin of the city is provided based on real-world measurements to test V2X systems based on the IEEE 802.11p WAVE standard. The 3D model incorporates building reflection and penetration losses using realistic ray-tracing techniques and empirically validated building attenuation models. A total of 60 GHz IEEE 802.11ad mmWave usage for V2I in urban canyon scenarios [44] has found that an RSU located at 5 m has an 8.5% blockage rate, but increasing the height by 10 m can eliminate over 99% of the blockages. For full coverage in an urban canyon, at least 2× more the current RSU density is needed at around 20 RSUs per square kilometer. As demonstrated in [45], mmWave performance in urban canyon environments can be further enhanced through machine learning-based beam alignment. Dynamic beam adjustments are particularly critical in scenarios where mmWave signals are obstructed by large vehicles or other transient obstacles. As a result, the beam search overhead is reduced by 70%. A framework for V2X combining several technologies, such as DSRC, LTE, 5G, NB-IoT, Bluetooth, GPRS, and mmWave, was proposed in [46], and in an urban canyon scenario, NLOS path loss can be extremely large after just one corner, especially for mmWave. The simulation performed in [47] states that RSUs significantly enhance object detection, reduce blind spots, and improve cooperative perception in ADAS systems. For transport modeling, RSUs can be a valuable source for travel data. Therefore, the need for RSUs in systems based on IEEE 802.11p shows promising insight for transport engineers.

In [48], Wi-Fi was selected over ZigBee as the V2X communication technology for autonomous golf carts designed for elderly passengers. The decision was driven by ZigBee’s insufficient bandwidth, which was inadequate for supporting video calls—an integral part of the system. In a similar scenario—using electric golf carts [49]—Wi-Fi was used; when the signal did not provide sufficient signal, 4G was successfully used as a backup. In [50], Wi-Fi is proposed as GNSS support along with IMU (Inertial Measurement Unit) in urban canyons with data fusion based on a Kalman filter, with a result of 1 to 10 m in the urban canyon scenario. State-of-the-art applications in five intersections in Michigan also used Wi-Fi. However, this was only for OBU-OBU communication [51]. In the NLOS scenario in the V2P usage proposed in [52], Wi-Fi performance is severely reduced with the maximum connection distance lowered to a maximum of 40 m due to long CET and signal attenuation. The beacon stuffing method was proposed to address connection issues, which allows reception of at least one BSM at 100 m. The presence of Wi-Fi, as a widely deployed technology in urban areas, can degrade the performance of C-V2X communication due to potential interference and spectrum congestion. A simulation performed in [53] indicates that the reliability range can be lowered up to 90% (20 m). Study [54] quantified the increased deployment costs resulting from Wi-Fi interference with ITS systems, which necessitate denser infrastructure. A similar issue may arise with the installation of Roadside Units (RSUs) in congested urban environments. However, from a transport modeling point of view, the denser the data sources, the more valuable detailed data is.

ZigBee operates on a 2.4 GHz frequency band. The band is the same for Bluetooth and Wi-Fi described in previous paragraphs of this paper. Reference [55] conducted a comprehensive real-world test of the ZigBee XBee S2C module, concluding 3 dBm is enough antenna power for applications, as more power does not give much better results but consumes more energy. Also, the study states that multi-hop extends coverage/RSSI but at the cost of latency, similarly to Advanced Encryption Standard; however, it keeps energy consumption reasonable. The maximum reliable range and antenna type were set in [56], where the estimated reliable range is 600 m. However, ref. [57] indicates RSSI drops after a 300 m distance. However, the system based on Arduino and XBee successfully gave warnings for risks such as close stops (less than 3 m), intersection approach, and unfavorable environmental conditions. Ref. [58] proposed vehicle collision warning with speed prediction with less than 0.3 m/s error and location with less than 0.3 m error. However, for location data, GPS or other satellite-based systems are proposed, which could be interrupted in an urban canyon scenario, like another example using GPS and ZigBee for V2V in a low-speed scenario [59]. An algorithm was proposed to optimize speed to reduce stops and idling. At the same time, for crossing intersections at a green light [60], ZigBee was used for communication between OBU and RSU; however, this was done with GPS usage for speed and location estimation. Hybrid approaches are also possible; for example, using a combination of GNSS data, IMU, speedometer, and 3D maps with Kalman filter usage [61], achieving accuracy of 1.5 m in an urban canyon with a 40% error rate. The scenario presented in [62] is suitable for urban canyons. The study presents a system based on ZigBee for short-range and LoRa for long-range communications. The framework enables RSUs to track vehicle location without GNSS-based data. Another study proposed a ZigBee/Wi-Fi/BT RSSI-based location system, allowing for obtaining precise (less than 1m) location data without using a GPS node-to-node [63]. Therefore, the ZigBee-based system can be used in urban canyons. Another concern could be 2.4 GHz band usage in cities, widely used for Bluetooth and Wi-Fi. Study [64] proposed 900 MHz as an alternative with a range benefit over 2.4 GHz for V2V uses.

LoRa’s official range is 500 m with line-of-sight. However, in [65], researchers tested the LoRa obstacle environment, and the range dropped to 160 m during V2V and V2I communication tests. On the other side, ref. [66] states that LoRa provides excellent penetration through urban obstacles, maintaining connectivity with packet error rates below 10%. Also, the study highlights low cost can power consumption over DSRC and C-V2X. A system based on LoRa, achieving 100% packet delivery, was proposed in [67]. High delivery in real-world tests was achieved through device-to-device direct communication supported by a Kalman filter. A stable connection was provided in the urban canyon scenario, but with affected line-of-sight and RSSI stability. Worth noting is study [68], which proposes guidelines on how to operate with bandwidth and spreadfactor to resist Doppler and fast fading. In the NLOS canyon scenario, LoRa is viable for non-safety V2X applications up to 58 km/h [69]. Higher speeds do not correspond with the transport policy of modern cities focused on lowering accident casualties; therefore, SF-12, with lower RSSI at higher speeds, tested in the paper with comparison to SF-7, should be sufficient for the urban canyon scenario. Moreover, SF-12 performs better in NLOS scenarios, with better detectability of weaker signals. It is important to mention LoRa’s far superior capabilities in UAV applications in comparison to Bluetooth, Zigbee, SigFox, and 802.11p, managing to provide V2V communication over 10km range in urban areas [70].

Data from V2X systems will be possible to acquire from VANets. In [71], a Software-Defined Networking (SDN) approach to VANets is proposed. SDN improves programmability, scalability, and flexibility. The study features real-time and historical data collection for analysis. The data collected are location, speed, and connectivity status. Ref. [72] indicates that VANets can provide GPS coordinates. Ref. [71] adds that VANets can also collect trajectory, traffic density, and traffic flow patterns. VANets data is also applicable to urban planning and infrastructure planning usage. For this case, the Infrastructure Domain type of VANet architecture is more useful, as it communicates between OBUs and RSUs. RSUs will provide data from OBUs to the central servers in the case. An ad hoc domain connecting OBU to OBU will not provide sufficient information for traffic modeling. Traffic-oriented VANet architecture is also present in [73], where Congestion-Free Path (CFP), and Optimized CFP are added to SDN VANet. A real-world test of SDN is presented in [74], using Raspberry Pi and OpenFlow switches. OpenDaylight controller, in this case, allows for data collection via standardized OpenFlow messages. Another SDN-based solution, enhanced with a 3-stage fuzzy decision tree model, is proposed to aid traffic signal optimization in [75]. In this case, data is gathered from the vehicles via VANet to RSU. RSUs standardize and preprocess the data. Data is processed by central or fog-based fuzzy decision models. Machine learning approaches in VANet architecture can also result in a rich data source for traffic modeling, as it requires data acquisition for the learning process. In [76], vehicles and RSUs collect the data, the blockchain layer verifies the identity and reputation of the data source, then the Neuro-Fuzzy server aggregates data and performs fuzzy logic-based decision making, and a distributed deep neural network needs the data collected to train.

4. Traffic Modeling—How V2X Technologies Provide Data in Urban Canyons, Based on Actual Big Data Research

To anticipate how traffic modeling will evolve in future urban canyons, this section summarizes recent research on the use of big data sources in traffic modeling. Based on this review, a hypothesis is formulated in the following section.

Traffic modeling uses various types of data. This section is a review of data collection and usage by transport engineers. The technologies mentioned are already used/proposed as V2X, and advanced research consists of Bluetooth/Wi-Fi usage. This data source is independent of specific industries, as the information is collected using Bluetooth and Wi-Fi scanners. Spatiotemporal data enables the initial acquisition of origin–destination matrices for travel patterns. Based on the detected locations, it is possible to reconstruct the traveler’s path; by analyzing the time intervals between points, travel speed can be determined, which in turn allows for the identification of the mode of transportation. With census data and detection times, travel origins and destinations can be classified as home, work, or other to acquire travel motivation. With such data gained, transportation models and analysis based on the models are possible. BT/Wi-Fi scanners located on roads search for MAC addresses of Bluetooth devices in cars or phones with visibility mode on. Therefore, even when Wi-Fi or Bluetooth is used solely for OBU-to-OBU communication, it is still possible to obtain the necessary data. The least advanced BT scanner usage was traffic measurements with an average speed measure in the road section [77]. Alternatively, Wi-Fi scanners could be used to obtain MAC addresses. However, despite the high hopes in previous work, with usage of a Bluetooth scanner [78] has already demonstrated significantly lower data usability than address detection using Bluetooth. The effectiveness of Bluetooth technology used for this purpose increases with version 5.0 [79]. Experiments conducted in three cities [80] shows that in comparison with RADAR detection, Bluetooth scanners have only 62% detection accuracy. An interesting finding in the study was that Bluetooth detection performed better in dense urban areas. That finding is crucial, referring to urban canyon usage. State-of-the-art research [81] allows for the grouping of detected travelers, which may be crucial in using scanner data to study travelers’ transportation behaviors. The use of Bluetooth scanners in crowd research was investigated in [82]. In the case of determining travel matrices, which are still not precise, the problem was addressed in [83]. Paper [84] studied a dedicated group of students. The specific research group allowed the focus to be placed on detecting the means of transport using Bluetooth/Wi-Fi scanners. The study differentiated between five means of transport. In [85], it was shown how to use Bluetooth to determine the distribution of trips on a transport network. Report [86] showed that traditional methods are more economically advantageous for continuous measurements throughout the year. When congestion measurements are planned for a longer period, the use of Bluetooth scanners is justified, even in an indoor scenario [87], with a novel method for collecting and analyzing Bluetooth traffic during university exams using the Ubertooth One device. The method captured 263 unique Bluetooth MAC addresses across five sessions, classified traffic confidence levels, filtered out noise using RSSI thresholds, and localized devices using trilateration and heatmaps. In the urban area where the use of Bluetooth scanners was commissioned and a series of studies were being conducted is Brisbane, Australia, ref. [88] mentions a network of 400 Bluetooth scanners there. In [89], the Brisbane scanner network use was to combine scanner data with data from smartcards. The Brisbane scanners arrangement is shown in studies such as [90,91,92]. The possible bias of data based on socioeconomic factors is mentioned in [93]. A paper that used Bluetooth scanners and GPS data to verify travel distribution is [94]. Despite arguably low usage of GPS in urban canyons, multiple papers in the section above, including experimental ones, made attempts to use GPS data in urban canyons, with successful results. Therefore, studies combining both V2X urban canyon sources is promising for traffic modeling in urban canyons.

Depending on the public or private RSU owners, adequate data can be similar as from the BTS (Base Transceiver Station). However, in urban canyons, the Transport Analysis Zones (TAZ), according to the review in previous sections, will be smaller, with a denser RSU grid (even around every 120 m) to ensure proper wave propagation and connection stability. With 120 m between RSUs, a transportation zone grid will consist of thousands of transport zones, where an example study using mobile data in Stokholm had only 32 zones of the entire city [95]; for example, much smaller Wrocław (Poland) in its Comprehensive Traffic Research [96] is divided into 475 transport zones, where mobile phone data was used as support to classical methods—surveys and traffic measurements. Wrocław’s Comprehensive Traffic Research was analyzed further in [97]. The establishment of TAZ borders using K-means clustering was described in [98]. Using the mobile phone network in [99], trips could be divided into classic types related to home, work, and other activities. Ref. [100] proposed software that combines data from mobile phone networks with traditional data in the form of surveys, censuses, and publicly available geolocation data. The frequency of updates for the area in which the traveler is located depends on the database used. In [101], attention was drawn to the existence of two types of databases. The first is a billing database, which is biased because it depends on the frequency of mobile phone calls. This will generate a bias correlated, for example, with the age of the traveler. Accordingly, a signal database was utilized to record changes in the base station coverage area in which the traveler is located. The location is verified every six hours, or every three hours in the case of newer systems. Analysis of this type of data allows the traveler’s route to be estimated. The ping-pong effect and the bias associated with the socio-demographic characteristics of mobile phone users are also discussed there. Data from mobile network operators were also compared with traditional survey data. In [102], the same research area as in [101] was taken up, namely the Paris agglomeration, with particular attention paid to the problem of detecting travelers moving by metro. For an urban canyon V2X scenario, blind spots similar to that would be solved just by using RSUs that are not GNSS-based; proposed solutions are in the Section above. The comprehensive review [103] summarizes eight publicly available databases from mobile phones with five uses, including transportation in the form of obtaining OD matrixes, traffic congestion detection, urban planning, passenger hub analysis, and transport mode detection by using rule-based heuristics, fuzzy logic, Partitioning Around Medoids, Random Forest, Pipe-based trip inference, and the classical Speed-based Classification. These methods can be used to detect the transport mode in case of partially available databases from RSU operators in urban canyons, as sometimes law regulations require anonymizing data (e.g., in the EU by RODO [104]). The usage of mobile phone databases requires including bias related to age, as with age fewer people use smartphones [105]. With the use of RSUs in urban canyons by the same method, the bias should be redefined for the use of autonomous vehicles.

The study in [106] addressed the issue of problems related to measuring passengers underground. The main purpose was to compare data from travelers’ mobile phones’ GPS with data from AVL (automatic vehicle location) systems located in motor vehicles. However, this was also based on GPS. GNSS-based data in an urban scenario, according to [107], has an up to 51% success rate in identifying travelers’ home locations, and according to the study, still has better coverage and accuracy than traditional methods, such as surveys. A Bayesian network can be used to estimate the O-D matrix, ensuring GPS-based trajectory data and traffic counts [108]. The study also states that connected AV and Floating Car Data in the future can provide a sufficient big data source for traffic modeling. The study in [109] focused on determining how accurate GPS data must be to estimate vehicle speeds. Vehicle speeds can also be determined at the level of entire transport networks [110]. Bluetooth scanners were used as a reference, with speed measurements being their first application in traffic engineering, as mentioned above in the review. In general, GPS data can be detailed enough to be used in micro-level modeling; in [111], this data was used to assess the variability of drivers’ driving styles. Even with a small amount of data, for example, in the absence of connections to different base stations, GIS-based methods have been developed that increase the accuracy of determining a traveler’s location [112]. Using a neural network, it is possible to accurately determine the type of vehicle [112]. Vehicle GPS data are already used in transportation engineering. In dense urban environments, ref. [113] proposes the Conditional Wasserstein Generative Adversarial Network with Gradient Penalty (CWGAN-GP), predicting ride-sourcing OD matrices using GPS data obtained from taxi fleets in Chengdu (China) and New York City (USA). Another example of an advanced method—Randomized K-Nearest Neighbor Regression with Spherical Distance—is in [114] using New York City, Porto (Portugal), and San Francisco (USA) databases for real-time analysis, predicting taxi travel time and detecting and scoring speeding behavior, enhancing passenger safety. Ref. [115] analyzed GPS-based data from the Chicago Data Portal, including trip duration, distance, fare, and pickup/drop-off locations aggregated to census tracts. The study proposed a variety of recommendations for improving taxi policy by sociodemographic analysis of the passengers. The study [116] found that by using taxi data, this transportation node plays a vital role in serving disadvantaged populations and enhancing multimodal connectivity, especially for short, first/last-mile trips. Over one billion trips were analyzed in [117] in Manhattan in New York City, which is comparable to an urban canyon, and GNSS data was suitable in a directed weighted network, along with identification of key zones with introducing OD rank. The study was supported with a Monte Carlo simulation for verification.

Multiple data sources can reduce the impact of biases on results. For example, in [118], ANPR (Automatic Number Plate Recognition) data was used to reduce bias using data delivered from a taxi company. Ref. [119] focuses on the problem of bias in data from large databases. Although the publication primarily addresses traffic safety, it utilizes the same type of data, and the challenges encountered in traffic model development are comparable. The chapter also explores the application of specific data sources in modeling. The research methodology involved interviews with experts from academic, commercial, and governmental sectors.

In the case of bias in demographic data, the authors of [119] suggested that traditional research should be taken into account as a solution. The problem of data bias was also discussed in the book [120]. The largest source number found during the review was four in [121]—Bluetooth, loop detectors, navigation systems, and classical Census data. The study effectively addresses the urban canyon problem through multisource integration and quality weighting.

The review authors proposed the Personal Transporter [122] as a vehicle form in connected V2X environments in urban scenarios. Comparable vehicles would be e-scooters and bikes. Therefore, research including big data databases from e-scooters, such as [123], where the influence of the built environment and socioeconomic factors on e-scooter ridership was studied. The most interesting finding from an urban canyon research point of view is the correlation with higher ridership in a dense urban area. Therefore, that will be a valuable data source. Ref. [124] uses GPS data analysis and statistical models to uncover temporal patterns and policy implications for cities, comparing city bike fleets with e-scooter dockless fleets, by analyzing data by real-time API scraping from multiple vendors, gas prices, and weather data. This finding is interesting—a 1% gas price increase results in a 3.13% increase in e-scooter trips. To continue the bike fleets topic, ref. [125] used GPS data from three different bike fleets to analyze modal shift in Delft (Netherlands). User behavior was analyzed in [126] by a model combining local movements and long-range Lévy flight-like transitions, validated through Monte Carlo simulations. The study used two databases from Chicago and New York City (USA), containing over 40 million trips. While the study used end-to-end locations, a smaller study with bike usage in a university campus used GPS to retrieve point-to-point travel paths [127]. The paper identified areas for system expansion and suggested operational improvements, such as 6 h rebalancing, dynamic relocation, and incentive programs, as university bike sharing differs from city systems. With technological progress in a connected environment, dock stations for Personal Transporters can be virtual. The study in [128] analyzed travel patterns in Beijing (China) with a proposed clustering algorithm. The algorithm outperformed classical methods—K-means and MILP. The report in [129] indicates that shared bike fleets have “massive potential for expansion”, with electric bikes generating 30% of trips while taking up 21% of fleets; therefore, Personal Transporters connected to V2X systems will have the potential to research travel patterns with this micromobility transport mode. The study was possible to conduct by the use of data from Fluctuo, an aggregator of shared mobility data, data collected via open APIs, city partnerships, and operator agreements. In an urban canyon scenario, data would be collected by RSUs. The urban canyon will need sufficient transportation. Therefore, relying only on autonomous vehicles wouldn’t be effective as public transportation nodes. Big data GNSS data analysis [130] shows that bike fleets can support public transport, improving modal split in large cities. The more public transport connections, the more bike-shared fleet trips.

5. Traffic Modeling—A Hypothesis Based on the Review of Traffic Research in the Age of Connected Environment in Urban Canyons

5.1. RSU-Based Big Data Traffic Modeling

The review highlights technologies and systems that show promise for enabling safe and efficient V2X communication, even in complex urban canyon environments. This section integrates insights from both research domains to formulate a hypothesis regarding traffic modeling in the era of VANETs. From a traffic modeling point of view, urban canyons cause problems in efficiently planning and managing traffic flows, generated by the high-rise density grid of buildings, causing a high population density that must obtain nodes of transport with sufficient capacity. With limited space between buildings, it would not be possible to ensure the capacity of the road for private autonomous vehicles. Therefore, it is needed to support AVs by correctly planned and cost-efficient public transportation nodes. Transport models are necessary for planning transportation in urban areas. The review showed that big data used for transportation models results in precise and cost-efficient models. With connected V2X environments, more data sources will be available. However, as the review on V2X communication systems shows, urban canyon data sources can provide much more valuable and less biased data. An example can be a comparison of reviewed cellular data and data from RSUs. The methodology would be similar. However, RSUs will provide a more precise location. Figure 3 shows the TAZ Voronoi Diagram based on BTS locations in the Powstańców Śląskich district in Wrocław (Poland). An example case study street—also called Powstańców Śląskich (the district is named after the main street)—is shown as green lines in Figure 3. The street is a four-lane road with a tram track in the middle, with bike paths and pavements on both sides of the road, and trees. The street was chosen because of the high-rise buildings, up to over 200 m. The historical street buildings were destroyed during World War II, then rebuilt in Soviet residential style in the 1960s. In the recent 30 years, this street has been intensively densified with offices and high-rise apartment buildings. The BTSs are shown as red triangles. This representation varies from actual transmission ranges, as high-rise buildings interrupt signal. However, big data research uses zones according to the Voronoi Diagrams. Therefore, the example and figure are adequate.

In Wrocław’s Comprehensive Traffic Research, zones are not comparable to the BTS transmission range. The classical TAZs are set according to the natural barriers (rivers, lakes, hills), streets, railroads, and historical district borders. Figure 4 shows classical zones in Wrocław’s Comprehensive Traffic Research 2018. In Figure 4, the red rectangle shows the borders of Figure 3; the green line represents Powstańców Śląskich street’s section shown in Figure 3.

RSUs in a connected environment can provide the same type of data as BTSs. According to the review, the displacement of RSUs has to be a few times denser. Based on the review in the hypothetical case study, 120 m is assumed. RSUs are placed in two rows every 240 m. Therefore, from one to another is around 120 m. In Figure 5, the layout of the RSU is proposed on the street of the case study, with the Voronoi Diagram range for each RSU. The coverage in such a scenario is much more precise. Consequently, the data exhibits reduced bias, improved accuracy, and greater usability. TAZs derived from RSUs can be aligned with traditional zoning schemes, as illustrated in Figure 4, due to their significantly smaller size—typically a 120 m trapezoid, compared to conventional polygons that may span several dozen meters to over a kilometer. Furthermore, the increased spatial resolution of these zones enables the classification of trip origins and destinations based on land use, even within TAZs that serve mixed functions. For example, this approach allows for differentiation in areas such as the Powstańców Śląskich district, which includes both residential and commercial zones, as demonstrated in the case study. In the much smaller zones, it is possible to identify, e.g., the 8 am origin of the trip as work due to the location of the last connected RSU by the commuter, and the destination as home, due to a zone with only residential buildings.

5.2. 2.4 GHz Technologies Scanner-Based Big Data Traffic Modeling

The second example based on the review is the usage of Bluetooth/Wi-Fi scanners. During the review, numerous papers using at least one of these two technologies were found, mostly non-related to safety messages, V2P applications, etc. Therefore, vehicle or traveler detection can occur more frequently than nowadays. When fewer phone users have visibility mode on, and only cars with Bluetooth car kits are detected, detection of devices without visibility mode is possible. However, due to legal constraints, this aspect falls outside the scope of the present study. One key advantage of the approach based on Roadside Units (RSUs), as opposed to Base Transceiver Stations (BTSs), is its independence from private data owners. Consequently, research can be conducted without the need to purchase datasets. However, it comes with the cost of acquiring scanners. Therefore, the scanner layout will be much more spread than BTSs shown in Figure 5.

According to the literature examples from the review, an example BT/Wi-Fi scanner placement in the case scenario street is proposed in Figure 6. The scanner placement is shown as a red hexagon. The trips are recorded by scanning for identical MAC addresses of devices. For example, a trip along Powstańców Śląskich street would result in consecutive detections from scanners C-D-F-H, and a trip with an origin in the south-east of Figure 6 up to the northern end of the case study’s street would be an I-E-C-D-F-H trip. The trip with the destination in the middle of the street section would be C-D. Depending on the connected environment architecture, other technologies can be scanned. ZigBee is a promising source, as it operates similarly to BT and Wi-Fi at 2.4 GHz frequency.

6. Conclusions

In the comprehensive literature review, 131 sources are listed, mostly including research papers. The review methodology and structure of the papers are analyzed in Section 2. The purpose of the review was to introduce traffic research possibilities in urban canyon connected environment scenarios. For this purpose, a two-part review has been performed. The first review in Section 3 was to obtain a state-of-the-art point of view of V2X systems, VANet architecture, and data acquisition. The Section 4 review was performed to acquire state-of-the-art big data modeling knowledge. Findings from both of the sections are summarized in Table 1 and

Table 2. Summary of Section 4 findings about each technology.

Technology/Data Source	What It Measures	Spatial Resolution	Temporal Resolution	Advantages	Disadvantages	Typical Applications
Bluetooth/Wi-Fi Scanners	MAC addresses, travel paths, speed, transport mode	Medium (scanner coverage area)	High (real-time or near real-time)	Low-cost, passive data collection, useful in urban canyons	Lower detection rate (~62%), privacy concerns, device visibility required	OD matrix estimation, transport mode detection, congestion analysis
Mobile Phone Data (Billing and Signal DBs)	Location changes, trip purpose (home/work/other)	Low to Medium (depends on BTS density)	Low to Medium (3–6 h)	Wide coverage, passive data	Biased by age and phone usage, ping-pong effect, privacy regulations	OD matrix, urban planning, transport mode detection
GPS Data (From Phones, Taxis, AVs)	Precise location, speed, trip paths	High (point-level)	High (real-time)	High accuracy, suitable for micro-level modeling	Signal loss in urban canyons, privacy, cost	Speed estimation, driving behavior, OD matrix, AV modeling
Smartcards (Public Transport)	Boarding/alighting data, trip frequency	High (station-level)	~High (per transaction, check-in, check-out)	Accurate for PT users, integrates with other data	Limited to PT users, no full trip path	Public transport planning, multimodal analysis
ANPR (Automatic Number Plate Recognition)	Vehicle counts, travel times, license plates	High (camera coverage)	High (real-time)	Accurate vehicle tracking, useful for bias correction	Infrastructure cost, privacy concerns	Traffic flow analysis, bias reduction
AVL (Automatic Vehicle Location)	Vehicle GPS location	High (vehicle-level)	High (real-time)	Accurate fleet tracking	Limited to equipped vehicles	Fleet management, public transport monitoring
Loop Detectors	Vehicle counts, speed	Point-based (road section)	High (real-time)	Proven technology, reliable	Fixed location, no OD info	Traffic volume, speed monitoring
Micromobility Data (E-scooters, Bikes)	Trip paths, usage patterns	High (GPS-level)	High (real-time)	Supports modal shift analysis, urban canyon-friendly	Vendor-dependent, data access issues	First/last-mile analysis, urban mobility planning
Census and Survey Data	Trip purpose, socio-demographics	Low (zone-level)	Very Low (annual or less)	Rich contextual data	Outdated quickly, expensive	Model calibration, bias correction

. Section 5 summarizes the obtained knowledge based on the literature review and, based on this, proposes two hypotheses on how big data research will develop in an urban canyon connected environment. The first hypothesis proposes using RSUs as data sources, with their placement determined by insights from the literature review and proposed in the case study. The second hypothesis proposes how Bluetooth/Wi-Fi scanners can be used in the urban canyon scenario, with the proposition of scanner locations in the case study street. Both hypotheses and the review prove that V2X systems in urban canyons will provide rich data sources for traffic modeling. The hypothesis from Section 5.1 shows that with a dense, around 120 m RSU grid, the data will be at a high level of accuracy and usability in traffic modeling. Research gaps and trends found in Section 3 review are as follows: BT integration with smartphones, GNSS support with Wi-Fi, GNSS-free localization in case of ZigBee usage, NLOS applications for 5G/NR-V2X and LoRa, RSU density optimization for C-V2X systems, and mmWave integration, beamforming, and digital twin modeling for DSRC. The research gap for traffic modeling is to obtain and base a case study on data delivered by existing connected vehicle environments and testbeds.