The design and deployment of traffic management solutions are expected to improve traffic flows, which will help detect and reduce traffic jams and accidents (Figure 2
). With flow optimization, it is also possible to reduce resource consumption and vehicle emissions [67
]. Usually, traffic management is achieved by taking into account the operation of routing protocols and how they manage data. Based on these two features, schemes to deal with traffic management can be classified in a taxonomy considering two main activities: Prediction and Optimization. To enable the prediction process, road traffic information must be collected using some kind of methodology. In [68
], the authors presented a taxonomy for road traffic information collection that is used to implement traffic prediction functionalities. According to this taxonomy, predictive methodologies may gather data using analytical models [69
], fixed sensors [71
], mobile phones [72
], or mobile devices (e.g., smartphones) combined with GPS receivers [74
]. After the prediction process, the collected information must be evaluated by a routing control [77
] or traffic signal control [70
] methodology to implement a traffic optimization strategy. Vehicles driven by humans can then use this information to reduce road congestion. In this section, the most important contributions for both activities are presented and discussed. Afterward, some open issues are also identified as a starting point for the proposal of new methodologies to deal with traffic management operations and problems.
4.1. Traffic Prediction in Vehicular Networks
Several ITS applications consider prediction methods to determine how flows (e.g., paths, links, and nodes) change with time and traffic intensity. The information collected by such methods can be used by traffic systems to provide real-time information. Commonly, traffic models consider three different measures to model traffic streams: speed, node density, and node flow. This subsection overviews the most relevant proposals related to traffic prediction for vehicular networks, grouping them according to the way that road traffic information is collected.
Analytical models have been proposed to study traffic models in different scenarios without the need for an infrastructure to collect information from the network. Castillo et al. [81
] designed a graph-theory-based methodology supported by two approaches (i.e., algebraic and topological) that solve the observability problem. This methodology determines which flow should be considered after the observation process, independently of route choice probabilities. The obtained results showed that a topological approach achieves better results because it consumes fewer network resources and involves no rounding errors. In [82
], a modeling framework for urban traffic system was proposed that allows the structure of urban traffic systems to be studied and node interactions to be understood, by systematically defining the components and their behavior that are necessary to enable driver simulations. To describe the structure and behavior of the components, a multi-level Petri-net-based formalism [83
] was considered. A cellular automata approach to building traffic mobility models was designed in [84
]. This approach provides rules for grouping vehicle motion based on a traffic signal coordination model of road intersections. An aggregation methodology to predict traffic flows in vehicular environments was proposed in [85
]. To accomplish the prediction task, this methodology bases its operation on four different parameters: the moving average (MA), exponential smoothing, autoregressive MA (ARIMA), and the neural network (NN) model. All parameters consider the traffic information collected over several days (24 h/day), which are generated taking into account three different time series (i.e., weekly, daily, and hourly). The NN parameter then collects the prediction of each model in the aggregation stage to give a final prediction. An adaptive hybrid fuzzy-rule-based system (FRBS) [86
] was proposed to model and forecast traffic flows in urban arterial networks. It considers powerful structures capable of handling faults and unpredictability. They can perform either online or offline, owing to the deployment of a genetic algorithm and an expert’s knowledge model. A traffic prediction model to control and coordinate urban traffic networks was designed in [87
]. This model takes advantage of a mixed-integer linear model to increase network optimization. Studies conducted through simulation have shown the accuracy of this model in reducing the computational complexity, thereby increasing its applicability in real environments. A theoretical framework to estimate traffic speed was proposed in [88
]. It combines low-resolution position data with other data sources to solve the problem some probe vehicles have in providing accurate positions. With this approach, this framework manages to estimate traffic speed more accurately.
Another way to predict traffic behaviors and create optimized solutions is to collect data from moving vehicles using fixed sensors placed at roadsides. By using these devices, it is possible to collect several metrics, such as the current vehicle speed, the number of passing vehicles, or vehicle flow. Following this data collection methodology, a framework based on non-conventional techniques to predict the speed at single-loop detectors is proposed in [71
]. From the framework predictions, it is possible to obtain a length-based vehicle classification that is further compared with concurrent measurements from video and dual-loop detectors. Tyagi et al. [89
] designed a traffic flow estimation system that gathers information about moving vehicles from a single roadside microphone. The microphone allows tire and engine noise and occasional honks, among others sounds, to be gathered. The collected sounds are optimized using algorithms to extract short-term spectral features. This optimization allows traffic flows to be classified according to three different density states: Jammed (0–10 km/h), Medium-Flow (10–40 km/h), and Free-Flow (40 km/h and above). To improve the optimization process, an approach is proposed in [90
] that considers the Cambridge Systematics from NGSIM [91
] to collect traffic information (e.g., flow, density, and speed) from vehicle trajectories.
Another possible solution to the problem of collecting traffic information uses mobile phones. To perform this task, they should send, from time to time, measurement reports to the core of the cellular network. This approach avoids the installation and preservation of fixed infrastructures and sensors. Following this approach, Valerio et al. [72
] considered the use of cellular network infrastructure as the base of a system designed to estimate road traffic. This is accomplished by collecting signaling messages from the mobile network. Another system to collect real-time urban traffic was proposed in [73
] by employing the LocHNES (Localizing and Handling Network Event Systems) framework to collect information about the instantaneous positions of buses and taxis, providing information about traffic on roads. This platform was deployed and evaluated in a real environment (Rome, Italy).
Mobile phones may be combined with vehicles equipped with Global Positioning System (GPS) devices and antennas. This approach allows vehicle positions to be gathered and data to be sent to dedicated servers with a global system for mobile communications (GSM) transmitters. Li et al. [76
] designed a method to determine recurrent congestion points using historical probe car data. The method was tested in Beijing for 57 days, considering 7000 taxis. Analysis of the proposed method resulted in several evaluation models for recurrent congestion points. In [92
], the authors proposed a sensing estimation algorithm that implements a monitoring center for traffic estimation. This monitoring center receives location and speed information updates from probe vehicles. Probe vehicles are equipped with GPS receivers, allowing them to know their current speed and location. Studies conducted in Shanghai, with 4000 taxis, have shown that hidden structures with traffic condition matrices underlie the probe data. Kang-Ching et al. [93
] also exploited probe vehicles equipped with GPS devices to optimize traffic status estimation. Data assimilation techniques and Newtonian relaxation methods are considered to integrate probe data into macroscopic traffic models. In [94
], a management system (PRTMS) for VANETs was designed to predict traffic flows considering a communications infrastructure based on the IEEE 802.11p standard. This approach implemented together with a modified linear prediction algorithm, allows a PRTMS to be deployed in scenarios with multiple road intersections. A central controller is also implemented to detect congestions and re-route vehicles in case of traffic congestion. DTMon [95
] was proposed as a dynamic traffic monitoring system to gather information about traffic conditions. To accomplish this task, it uses traffic information from moving vehicles collected by roadside infrastructures to generate reports for traffic management systems. Rong Du et al. [96
] designed two patrol algorithms and a sampling rule based on the relation between the disparity of samples and traffic estimation errors, which allows the paths of floating cars to be set, thus enabling them to participate in traffic monitoring functions. Studies showed that the sampling rule combined with a patrol algorithm reduces estimation errors. In [97
], the authors proposed a protocol specifically designed to send secure messages, the goal being to improve the network quality of service. These messages are sent assuming that all participating vehicles have GPS devices, which allows them to know their exact location when experiencing some failure or accident. When this type of situation occurs, vehicles broadcast their position to vehicles behind them. This approach allows the focus of traffic congestion to be determined, enabling a geocasting packet transmission approach. GeoWave [98
] is a network protocol that uses IEEE 802.11p to exchange useful messages that are used by an active/passive safety system. With GeoWave, vehicles exchange crucial information with their neighbors or with RSUs, enabling the network infrastructure to continuously gather information about traffic conditions. In [99
], the authors introduced a predictive system to calculate the best path between two points. This system is capable of dealing with the single characteristics of different types of obstacles and road structures while planning the optimal path to be followed by vehicles. It also implements a quadratic model with a predictive controller, allowing the solution of an important issue related to two-dimensional obstacle avoidance. To model the proposed system, the authors considered the CarSim vehicle model.
4.2. Traffic Optimization in Vehicular Networks
In the literature, the most efficient and commonly considered approaches to deal with traffic optimization are divided into two classes: traffic light control and routing algorithms for vehicles. Traffic light control can be described as a joint decision regarding the duration of the signal phases associated with each signalized intersection in the network. It can be implemented considering either a fixed-time or an adaptive-time method. When a fixed-time method is considered, the history of traffic statistics is used to set cycle time and the duration of one light, ignoring all the changes in traffic conditions. On the other hand, adaptive methods set the amount of time of a cycle and the duration of one light according to real-time information about traffic and road conditions. Routing algorithms for vehicles are used to inform drivers about optimal routes (e.g., routes free from traffic congestion), taking into account a set of constraints such as shortest path or time intervals.
Diakiki et al. [100
] proposed a multivariable regulator approach (TUC) to deal with traffic signal control in urban scenarios. It employs the store-and-forward paradigm to formulate the urban traffic control problem. It also uses linear quadratic regulator theory to create a network-wide signal control suitable for roads with a high level of congestion. In [101
], a hybrid strategy for real-time traffic signal control for urban scenarios was designed. This approach adapts its functionality based on traffic conditions. If a road is under saturated signalized junctions, traffic signals are controlled using a real-time Webster-type demand-drive strategy. When the road is close to saturation, this strategy changes its behavior to control signalized junctions with TUC. A technique to control real-time signals was evaluated through an API proposed in [102
]. This technique allows traffic fluctuations to be detected in real-time because it considers a dynamic programming approach combined with adaptive signal control methodologies.
Zegeye et al. [103
] proposed a receding-horizon parameterized traffic control approach that combines prediction adaptation and nonlinear models with faster computation speed. This approach also implements a theoretical strategy to design control laws for variable speed limits and ramp metering. An architecture that bases its operation on Markovian properties combined with geometric fuzzy sets was designed in [104
]. In defines all agents to have equal decision capabilities, and their operation is based on five modules that operate concurrently. Agents are associated with an intersection and receive information from sensors, which allow them to calculate the duration of green phases. To perform this task, they consider the results gathered by two different modules. The data collection module determines the queue count, while the communication module regulates the status of each neighboring intersection. All this information is put together with an average flow rate. Oliveira et al. [105
] proposed a framework to deal with multi-agent control of linear dynamic systems. This approach allows breaking down a centralized model by employing distributed agents to solve small parts of the predictive control problem. To achieve this goal, each agent senses, and controls the variables for its part of the problem, communicating with its neighbors to obtain their variables to coordinate their actions.
] traffic is a platform that uses the standard GPS driving aid to effectively discover and disseminate traffic congestion routes to drivers. This platform was designed to exchange information only on areas with unexpected traffic conditions. To determine traffic patterns, this platform uses a combination of clustering and epidemic communication. It allows each vehicle to build a speed map containing other vehicles’ speeds to transmit it to its neighbors. Shinji et al. [107
] designed an automobile control method that attenuates the effects of traffic congestion by taking advantage of an ad hoc communication methodology among vehicles. Each vehicle shares information about traffic conditions, allowing other vehicles to calculate a suitable congestion-free route. BeeJamA [108
] is a distributed traffic control model that controls vehicle traffic based on the Bee Hive routing algorithm [109
]. In this model, vehicles are guided under a decentralized control system. Hussain et al. [110
] proposed a protocol to manage congestion situations and avoid deadlocks in urban environments. This is accomplished by guiding vehicles to routes free of congestion, using the infrastructure placed at traffic lights to collect information about nearby vehicles and neighboring infrastructures. The TraffCon routing algorithm [111
] was designed to deal with traffic congestion problems and seeks to improve road capacity and decrease fuel consumption and trip times. To deploy TraffCon, each vehicle must be equipped with a GPS receiver connected to some kind of wireless computing device with an interface that can show to the driver useful information about traffic conditions. By allowing direct communications with individual vehicles, this routing protocol aims to exert great influence on the transportation system. The i
CAR-II routing protocol [113
] was proposed to enable data communication between vehicular applications and Internet services. This protocol allows vehicles to forecast the existence of local connectivity and to update location servers with crucial and real-time information. i
CAR-II succeeds in building a global and scalable network topology.
Backfrieder et al. [114
] proposed a traffic optimization system based on a novel estimation algorithm called Predictive Congestion Minimization in Combination with an A* (PCMA*). This algorithm is designed to calculate routes for users that are far away from their final destination. This suggestion may bring about significant advantages because users far away from congestion points can early react to potential jams. It uses a wide area of timely notification approach that takes into consideration the current state of roads. To perform traffic optimization in VANETs, a protocol called Fuzzy Bacterial Foraging Optimization Zone-Based Routing was proposed in [115
]. This protocol tries to find, in a short period of time, the most stable short route to a target under uncertain conditions. To accomplish this, it uses three techniques: (i) Zone-Based Routing to provide stability; (ii) Bacterial Foraging Optimization to calculate the shortest routes; and (iii) fuzzy logic to deal with uncertain conditions. In [116
], the authors designed an algorithm capable of dealing with traffic-collision-free path-finding and route optimization between two points. With the aim of optimizing performance, this algorithm considers a system composed of three distinct models: prediction of destination points, region formation, and optimized route selection. Li et al. [117
] proposed the AQRV routing protocol, which has the ability to select the best intersection path for a vehicle to deliver data to its final destination. It considers three metrics: the connectivity probability, packet delivery ratio, and delivery delay. In other words, the best path for a vehicle to follow is the one with a higher connectivity probability and delivery ratio and a lower delay. Thus, the authors considered an ant colony optimization methodology to solve the problem of route selection.
Gupte et al. [118
] designed a model called VANET-based Autonomous Management that makes routing decisions based on the travel paths of nearby vehicles. This model allows vehicles to exchange information about route selection, congestion, and traffic alerts, which allows them to make decisions about the best path to select. It also considers traffic light controllers to solve congestion situations, which dynamically shifts the traffic patterns considering not only vehicle destinations but also road conditions. Two delay-tolerant routing algorithms to deal with traffic in vehicular networks were proposed in [119
]. The first one, delay-bounded minimum-cost forwarding (D-Greedy), handles local changes in traffic conditions by focusing on the number of vehicles and their speed. The second algorithm, D-MinCost, focuses on global traffic conditions, calculating vehicle speed and density in the entire network. Both algorithms reduce the number of transmissions needed to deliver messages to their final destination. Prakash et al. [120
] proposed a path selection algorithm that aims to reduce not only the trip time but also fuel consumption. To accomplish this goal, this algorithm ensures the best path selection by combining information about the origin and destination of each vehicle with an approach where vehicles could choose a new direction at each road intersection. An opportunistic traffic management system was proposed in [121
] for traffic optimization in vehicular networks. It uses vehicles to exchange individually crowd-sourced traffic information to dynamically recalculate new routes. This work also quantified the effects of deploying a decentralized traffic-based navigation system by performing real experiments in Portland (USA) downtown.
], a traffic optimization data-driven approach was proposed to deal with the control of vehicles equipped with wireless communication devices. This approach tries to overcome several V2V communication issues (e.g., limited communication range or input saturation) by exploiting recent advances in approximate/adaptive dynamic programming. Won et al. [123
] designed a jam control protocol that considers vehicle-to-vehicle communications and a three-phase traffic theory to detect and avoid traffic jams. To calculate traffic jam dynamics, a fuzzy inference model is deployed together with a V2V-based jam detection algorithm. By treating traffic jams as a dynamic phenomenon instead of a binary event, this protocol aims to overcome some issues regarding the mitigation of phantom jams. In [124
], a congestion control strategy was proposed to deal with packet loss in high-congestion road intersections. This approach uses RSUs placed at road intersections to collect, filter, and cluster warning messages, in an attempt to decrease their delivery delay and drop ratio. This strategy was able to reduce the number of traffic collisions, increasing significantly the number of warning messages delivered to passing vehicles.
A driver monitoring system to improve road safety was proposed in [125
]. This system uses different parameters to make traffic decisions. For example, it considers the vehicle’s temperature, noise level, driver heart, and respiratory rates. Based on the data collected by sensors, the system can analyze the ability of the driver to safely operate a vehicle and decide which is the best action to take to avoid traffic collisions. Zheng et al. [126
] studied how RSUs should be placed to mitigate their placement problem in traffic flow monitoring systems to improve the security of passing by vehicles. They proposed three algorithms to place RSUs on roadside based on specific characteristics of traffic flows. To improve the use of resources and decrease the number of deployed RSUs, the authors proposed an extension to a credential propagation mechanism by using vehicle-to-vehicle communications to increase the coverage of each RSU. An intelligent smart vehicle monitoring and assistant system, supported by cloud computing and mobile applications, was proposed in [127
]. This system uses a set of novel techniques to improve collision avoidance, accident detection, and video/photo surveillance mechanisms. Most of these techniques consider the speed-based lane change model to reduce traffic congestion which will result in fewer road accidents due to improper lane change. A vehicle traffic management system (dEASY) is proposed in [128
] to deal with traffic events that may have an impact on the overall network performance. This system considers three different layers to implement an architecture that is capable to deal with the selection of the best vehicle to forward data by implementing an altruistic approach to choose alternative routes. Guidoni et al. [129
] proposed Re-Route, which is a traffic management service that uses nodes density to implement a traffic model capable to detect congested routes. By detecting congested routes, Re-Route aims to reduce traffic jams and improve traffic flows. The idea behind this scheme is to reduce jams instead of moving them to other roads.