1. Introduction
Nowadays, the automobile industry is focused on developing smart vehicles equipped with various sensors, computing power and communication functionalities. The Internet of Vehicles (IoV) [
1] is part of the Internet of Things (IoT) because in a broad sense smart vehicles are smart things and, in another way, smart vehicles are realized with things like various sensors. Smart vehicles can be aware of and act properly according to their surrounding situations as recognized by sensors. Vehicular communications (or vehicle-to-everything communications, V2X) are one of the necessary means for situation awareness and cooperative operations among vehicles and can be categorized into vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I) and vehicle-to-sensor (V2S) communications [
2].
The efficiency of road transportation depends heavily on the performance of traffic signal controllers. Traffic signal control systems have been rapidly evolved during the last several decades [
3]. Thanks to that, the traffic handling capacity of roads is significantly improved, and travel time and fuel consumption are reduced. In these days, traffic lights can be controlled in real-time by coordinated controllers at intersections which monitor traffic patterns with the assistance of devices like video cameras and sensors (e.g., loop detectors). Video camera-based monitoring [
4,
5,
6,
7,
8,
9,
10] requires high computing power for real-time image processing and sensor-based monitoring [
11,
12,
13,
14,
15,
16,
17] incurs high sensor installation and maintenance cost. Moreover, these technologies suffer from various environmental obstacles like weather, lighting and road condition, which cannot be completely overcome by any countermeasures.
Therefore, in this paper, we propose to use vehicular communications for traffic signal control because the vehicular communication functionality, one of the essential capabilities of IoV, has many advantages like no additional operational (installation and maintenance) cost, lightweight computing, resilience to lighting condition (i.e., can operate all day) and resilience to harsh road condition (e.g., can operate to a certain degree even in a non-line-of-sight environment). In controlling traffic signals, we can substitute vehicular communications for video cameras and sensors, which can be achieved by making vehicles notify traffic signal controllers of their existence via V2I communications so that traffic signal controllers can estimate how many vehicles are waiting for the green light (i.e., the vehicle queue length [
18,
19]).
For vehicular communications, the IEEE Wireless Access in Vehicular Environments (WAVE) [
20] and the IEEE 802.11p [
21] are standardized. From the perspective of vehicular communications, traffic signal controllers are road side units (RSUs). The channel access performance of the IEEE 802.11p significantly deteriorates as the access attempts to the channel increases because the MAC protocol of the IEEE 802.11p is based on the carrier sense multiple access (CSMA) mechanism. In the vehicle queue of a multiple lane road, vehicles tend to line up compactly and communicate various types of data traffic via V2X communications, so the message collision possibility of the vehicles in the vehicle queue (near the intersection area) is much higher than the other areas of the road. Therefore, it is desirable to alleviate the collision possibility of messages by reducing the messages generated by vehicles for traffic signal control (i.e., the V2I messages sent to the traffic signal controller). Particularly, with considering only the V2I communications for the vehicle queue length estimation, if each vehicle in the vehicle queue attempts to send a message to the traffic signal controller (we call this the Naïve mechanism), multiple message transmissions may co-exist in the air, causing collisions, because there can be more than one vehicle joined the vehicle queue almost at the same time due to multiple lanes of a road and the stopping speed. Therefore, we need a mechanism to limit the vehicles sending messages to the traffic signal controller in order to reduce the possibility of collisions and the message transmission overhead.
In our previous work [
22], we proposed a mechanism, called the distance-based mechanism, that considers the distance of a vehicle from an intersection as the criterion of controlling the message transmission to the traffic signal controller. To the best of our knowledge, [
22] has addressed this issue for the first time. In the distance-based mechanism, a timer is used to determine the time for a vehicle to send a message to the traffic signal controller according to the distance from the upcoming intersection. If a vehicle overhears a transmission from another vehicle behind itself, it gives up its transmission. Thus, the distance-based mechanism reduces the message transmission overhead to a half of the Naïve mechanism. However, on the red light, vehicles tend to line up one after another with slowing down their speeds with some time gap, so they may try to send messages to the traffic signal controller sequentially even if higher preferences are given to vehicles farther away from the upcoming intersection.
In this paper, to overcome this sequential transmission characteristic of the distance-based mechanism, we propose a new mechanism that can reduce the number of messages transmitted to the traffic signal controller. This newly proposed mechanism is called the sector-based mechanism. The sector-based mechanism further reduces the number of the vehicles sending messages to the traffic signal controller by adopting the concept of sectors. There can be a number of sectors in a road segment between two consecutive intersections. A sector of a road segment is a subarea of the road segment. Instead of all the vehicles waiting for the green light having rights to perform V2I communications, only the vehicles within the sectors are allowed to transmit messages to the approaching traffic signal controller. That is, the set of candidate vehicles for sending messages to the traffic signal controller of the sector-based mechanism is smaller than that of the distance-based mechanism, resulting in less transmissions to the traffic signal controller. For the performance evaluation, intensive simulations are carried out by utilizing the vehicle network simulation framework Veins [
23] based on SUMO [
24] and OMNet++ [
25] with considering various performance- affecting factors like sector length, inter-sector distance, vehicle density of the road segment, etc. In the performance evaluation section, we can observe that the sector-based mechanism, with the sector length 10 m and the inter-sector distance 10 m, performs almost the same as the distance-based mechanism in terms of the estimation accuracy of the vehicle queue length with significantly less V2I message transmissions, almost a third of the distance-based mechanism (i.e., a sixth of the Naïve mechanism). Because the parameters like sector length and inter-sector distance are easily adjustable, the sector-based mechanism can be a good candidate for estimating the vehicle queue length for intelligent traffic signal control in the IoV environment.
The rest of the paper is organized as follows: in
Section 2, we will describe the related work on traffic pattern monitoring mechanisms.
Section 3 describes the detailed operation of our V2I communication-based traffic pattern monitoring and vehicle queue estimation mechanisms. In
Section 4, we evaluate the performance of our mechanisms from the intensive simulation results. Finally,
Section 5 concludes this paper.
2. Related Work
In this section, we first go over the definition of the vehicle queue. The vehicle queue is defined as a line of the vehicles stopping at the red light and the vehicles approaching to the stopping vehicles at speeds slower than the given stopping speed in the Highway Capacity Manual [
19]. In [
20], the vehicle queue is composed of the standing queue and the moving queue. The standing queue is with the vehicles stopping at the red signal and the moving queue with the vehicles slower than the stopping speed because of the standing queue. The equivalent standing queue is defined as the vehicle queue including both the standing queue and the moving queue. In this paper, we adopt the equivalent standing queue of [
20] as the vehicle queue.
For the estimation of the vehicle queue length, first of all, the vehicles waiting for the green signal have to be recognized, which can be accomplished by utilizing devices like video cameras mounted on fixed roadside structures such as traffic signal controllers or like sensors installed under the pavement. The time-stringent control of traffic signals requires real-time processing of video frames and the accurate measurement of vehicle queue length requires sophisticated deployment of sensors.
In the video-based approach, the first thing to do for the vehicle queue length estimation is detecting vehicles from video frames in real time. After the vehicle detection process, vehicles are tracked and counted in real time. Thanks to various computer vision techniques and hardware capabilities, real-time processing of vehicle detection, tracking and counting becomes possible [
4,
5,
6,
7,
8]. The mechanisms that can be used for real-time vehicle detection from video images are background subtraction method, blob analysis, thresholding, hole filling, morphological operations, etc. Once vehicles are detected, vehicle tracking and counting are performed with using various schemes like similarity measurement, patch analysis, virtual detection line, virtual detection zone, shadow detection, removal, etc. A sequence of complex processing of video images induces very high computing overhead and requires specialized hardware to expedite the processing. In [
7], ARM/FPGA processor-based vehicle counting system is proposed to expedite video processing. As an example, the video processing procedure of vehicle detection and counting proposed in [
4] consists of preprocessing, background update, background subtraction, image segmentation, lamplight or shadow suppression, contour extraction and filling, vehicle detection and vehicle counting using virtual coil or detecting line depending on traffic congestion situation. Even with various video processing techniques, the adversary road surrounding environment, like bad weather (e.g., rain drops and snowflakes), dim lights, curved roads, etc., may significantly downgrade the quality of video images. The authors of [
4] aimed to provide robustness to video processing for vehicle detection, tracking and counting in various weather and light conditions. [
9] and [
10] improve robustness and accuracy even under bad road situations by adopting a feature-based detection method and a machine learning-based method, respectively, but consume abundant resources and may not guarantee real-time processing of video frames due to processing complexity. Recently, the mechanisms based on video images from unmanned aerial vehicles (UAVs) for traffic monitoring have been studied and this UAV-based approach is appropriate for large area monitoring with overcoming obstacles from wider top-view video images. For instance, in [
8], a framework based on UAVs is proposed for moving-vehicle detection, multi-vehicle tracking and vehicle counting. As we have described, most of the work on the video-based approach tackles previously-mentioned environmental hurdles which may not be completely overcome by the means of various video processing methods.
In the sensor-based approach, various types of sensors, like inductive loop detectors, ultrasonic sensors, magnetometers, radar/lidar based sensors, etc., are installed near to intersections for vehicle detection, tracking and counting [
11,
12,
13,
14,
15,
16,
17]. Each sensor is equipped with devices like a microphone to collect acoustic, seismic or any signals to classify vehicles. From the collected sensing signals, sensors and base stations detect, track and count vehicles. However, in the harsh road environment, sensing signals are affected by ambient noise, resulting in resource-intensive signal processing. Typical road sensors are deployed under the road surface at specific points and monitor the presence of vehicles at fixed locations, separately in each lane. Each sensor transmits a sequence of binary values indicating the presence of vehicles which is used for estimating vehicle flow, vehicle speed, vehicle classification, etc. For instance, inductive loop detectors are deployed at pre-specified points for traffic signal control as illustrated in [
17]. In [
17], we can find various deployment strategies of inductive loop detectors for various applications. For the accurate estimation of vehicle queue length, road sensors are to be deployed at sophisticatedly arranged points, which requires high installation and maintenance cost. Also, in order to supply power and allow communications, long cables are required to be installed along with sensors. Even with excluding the cabling cost, the high sensor installation and maintenance cost makes sensor deployment in all intersection areas infeasible. Wireless sensors can avoid cabling, but they have the drawback of short lifetime due to their power-constrained batteries. The lifetime of wireless sensors can be lengthened by implementing energy harvesting capability in wireless sensors which converts the vibrations induced by vehicles into energy.
Instead of using video cameras or sensors, the mechanisms utilizing GPS-mounted probe vehicles have been proposed for the estimation of the vehicle queue length [
26,
27,
28,
29,
30,
31,
32,
33]. Probe vehicles are special purpose vehicles designed for monitoring road traffic situations and collecting trajectory data. The performance of the probe vehicle-based approach is affected by the number of probe vehicles deployed on the road. The ratio of the number of probe vehicles to the total number of vehicles is called the penetration ratio of probe vehicles. Larger penetration ratio is better for achieving higher accuracy in terms of the vehicle queue length estimation. In the probe vehicle-based mechanisms, due to low penetration ratio of probe vehicles, one of the major issues is to enhance the accuracy based on the insufficient information from probe vehicles. Another issue is how to efficiently estimate vehicle queue length or traffic volume from the substantial data collected by probe vehicles. The main purpose of using probe vehicles is to collect traffic-related data throughout their journey and, then, to do the off-line analysis or estimation of traffic situations based on the collected data. Therefore, the probe vehicle-based approach is not for the real-time control of traffic signals.
The aim of our mechanisms differs from that of the probe vehicle-based mechanism in that our mechanisms use V2I communications for the real-time traffic signal control. That is, we consider the environment where the traffic signal controller detects vehicles through V2I communications and estimates the length of the vehicle queue and, then, controls the traffic signals. As the age of IoV is approaching [
34,
35], all the vehicles performing V2I communications (i.e., the penetration ratio of probe vehicles is 1) will be realized in the near future. In this case, V2I communication attempts from all the vehicles may cause collisions, so our objective is to limit the number of vehicles sending messages to traffic signal controllers without deteriorating the accuracy of the estimated vehicle queue length.
5. Conclusions
The Internet of Vehicles (IoV) allows vehicles to communicate with anything in the Internet, including vehicle themselves and traffic signal controllers on the road. In this paper, we focused on a cost-effective ITS application of controlling traffic signals in real-time in the IoV environment. Vehicular communication capabilities enable vehicles to communicate directly with traffic signal controllers (i.e., RSUs) located at intersections so that traffic signal controllers can estimate how many vehicles are waiting for the green light (i.e., vehicle queue length). Thus, with V2I communications, we can avoid using conventional computing-intensive video cameras or high operational-cost sensors for the real-time estimation of vehicle queue length. Furthermore, V2I communications are robust compared with video-taking and sensing in the harsh road environment with full of obstacles.
In the V2X communication-based approach, both the V2I communication overhead and the accuracy of the vehicle queue length estimation must be considered. For the reduction of V2I communication overhead, we proposed the distance-based mechanism in [
17] and newly proposed the sector-based mechanisms in this paper. In the distance-based mechanism, vehicles farther from the traffic signal controller have higher possibility of sending messages to the traffic signal controller. In the sector-based mechanism, sectors are specified in a road segment and only the vehicle first stopped in each sector is allowed to perform V2I communications.
For the performance comparison of our mechanisms, we carried out simulations based on the Veins vehicle network simulation framework for various performance determining factors and analyzed the performance in terms of the message transmission overhead and the accuracy of the vehicle queue length estimation. The message transmission overhead indicates the utilization of constrained wireless link resource, so less message transmissions are preferred. From the simulation results, we showed that our mechanisms significantly reduce the number of message transmissions without losing the accuracy of the estimated vehicle queue length, compared with the Naïve mechanism. The sector-based mechanism decreases the message transmission overhead about a sixth of the Naïve mechanism and a third of the distance-based mechanism. This indicates that our V2X communication-based mechanisms are especially good for the road situation with many vehicles. Also, we verified that the proper selection of sector length and inter-sector distance is important for achieving acceptable performance of the sector-based mechanism. According to the simulation results, the sector length 10~20 m and the inter-sector distance 10~20 m are good enough for the case of the vehicle length 5 m and the inter-vehicle distance 2.5 m. Because sector length and inter-sector distance are easily tunable parameters, our sector-based mechanism can be applied to any road environments at no extra cost. Also, based on the discussion in
Section 4.3, we come to the conclusion that our vehicular communication-based approach can be the ultimate enabler of intelligent traffic signal control in the age of IoV.