Indigenous design of a Traffic Light Control system responsive to the local traffic dynamics and priority vehicles
Introduction
The traffic light schedules significantly affect traffic movement through a junction. It has been observed that the traffic flow in the road network of urban areas is highly dependent on the coherence of traffic lights at signalized road junctions. The controlled optimization of traffic light phases at junctions can alleviate congestion and its resulting issues (Sen & Head, 1997).
The Traffic Light Control (TLC) systems can be classified majorly as static or adaptive. The static TLC uses manually-set fixed phase duration. Green phase duration in this case is configured based on the traffic profile of a junction at the time of deployment. Most of the time these pre-configured schedules are not responsive to the real-time changes in the traffic condition at a junction and lead to congestion in the region. At present, inefficient traffic light scheduling remains to be a prevalent problem in many urban areas. Most of the traffic light control systems are static. An additional concern in the static TLCs is the lack of any special traffic scheduling for priority vehicles such as transportation vehicles such as buses, or emergency vehicles such as ambulances, fire trucks, etc. Since the reduction in the response time of these high-priority vehicles has a significant advantage in safeguarding lives, an efficient adaptive TLC catering to this requirement is the need of the hour.
The adaptive TLC adjusts the green phase duration and cycle time at a junction, based on historical and real-time traffic information in order to optimize the throughput of a junction. The adaptive TLC is further classified as globally and local synchronous. The global synchronization-based adaptive TLC utilizes the historical as well as real-time traffic information gathered at all the junctions in a road network to adapt the phase and cycle time of the TLC at a junction. The proposals in literature use statistical processing (Mirchandani and Head, 2001, Yu and Recker, 2006), graph analysis (Chen et al., 2007), controlled optimization (Xie, Smith, & Barlow, 2012) and machine learning (Baluja, 2018), reinforcement learning (Abdulhai et al., 2003, Arel et al., 2010) based approaches. The global synchronization-based TLC implementation requires communication among all the junctions and is relatively costly to deploy. The phase and cycle time for all the junctions can be computed centrally (SCOOT— Split Cycle Offset Optimization Technique Hunt, Robertson, Bretherton, & Winton, 1981) or in a distributed manner (SCATS—Sydney Coordinated Adaptive Traffic System Lowrie, 1990). In such networks, the traffic lights are interdependent i.e., the phase of one traffic light affects the scheduling of neighboring traffic lights, and they fall under the category of global synchronization-based adaptive TLC systems.
The local adaptive TLC uses the traffic information gathered locally at a junction for computing phase and cycle time at the junction. We encountered a few works (Park et al., 2000, Smith et al., 2013) which process traffic data of a local junction using genetic algorithms to implement local adaptive TLC. Qi, Zhou, and Luan (2016) is an example of a study which proposes such a solution for isolated traffic light scheduling algorithm for an arterial street scenario.
The majority of adaptive TLC systems employ roadside sensors, such as loop detectors and traffic monitoring cameras. The loop detectors are widely deployed in the western world to fetch vehicle count and speed information. However, they have been observed to be highly erroneous in monitoring less-lane-heterogeneous traffic. Ali et al. (2013) designed a prototype of a strip-based multi-loop detector system for monitoring heterogeneous traffic. However, the large-scale deployment of this system is very costly due to the requirement of multiple loops per lane. Further, the deployment of loop detectors is intrusive and hence, difficult to maintain.
At present, in many developing countries, CCTV cameras are being widely deployed at junctions for traffic monitoring and surveillance. The camera feed from these surveillance videos, can further be processed, to count and classify the vehicles using the currently available state-of-the-art image processing algorithms. We encountered several works in literature that use FPGA (Adam, Garani, & Ventzas, 2014) and Raspberry Pi (Sorwar, Azad, Hussain, & Mahmood, 2017) based implementations for processing the video feeds.
Priority vehicles such as mass transportation vehicles and emergency vehicles such as ambulances, fire trucks contribute a minor part in the overall traffic dynamics of any city. Although being less in number, these vehicles render many crucial services, and therefore optimizing or reducing their wait times at junctions is a must. However, we observed that many of the most efficient adaptive TLCs (Covell et al., 2015, Park et al., 2000) in literature do not focus on this key issue. We came across several works in literature that focus on the need for a special algorithm or system for scheduling priority vehicles (Deepa et al., 2021, Djahel et al., 2013). Some of these are IoT based (Joshi, Jain, & Pandey, 2020), reinforcement learning-based (Su et al., 2020) and a few others (Humagain and Sinha, 2020, Obrusník, 2019, Tomar et al., 2020). The work accomplished in Kumar, Rahman, and Dhakad (2020) draws many parallels to the solution proposed in this manuscript. In Kumar et al. (2020), Neetesh et al. proposes an interesting RL based approach to address the issue of traffic light scheduling while taking into consideration priority and emergency modes as well. We noticed two key issues that have been overlooked in some of these approaches. First, several of these approaches need an extensive vehicle to infrastructure (V2I) and infrastructure to infrastructure (I2I) communication. Second, while prioritizing the emergency vehicles, some of these approaches do not focus on the impact that the algorithm has on other general classes of traffic.
To address these issues, this work proposes the design of a local adaptive TLC, which caters to the need of serving priority vehicles at a preference, while ensuring minimal disturbance to the other general traffic. In essence, this work is an extension to our previous work (Bisht, Ravani, Chaturvedi, & Kumar, 2019) where we established the feasibility of upgrading the existing static TLC to a local adaptive TLC with minimal infrastructure requirement.
An important property of a local adaptive TLC is that it can be deployed incrementally in a road network based on the availability of funds, making it suitable for deployment in developing countries. The proposed adaptive TLC algorithm, even while prioritizing certain vehicles, tries to follow round-robin scheduling of phase transitions to ensure fairness among competing for traffic flows of different approaches. Following are the major contributions of our work:
- •
The design of a local adaptive TLC algorithm catering to the need of special handling for priority vehicles.
- •
The simulation-based performance evaluation of the proposed adaptive TLC for different traffic classes, namely general traffic, priority vehicles and emergency vehicles, for mainly two metrics – average travel time and average waiting time – under the less-lane-disciplined heterogeneous traffic conditions (for developing countries), capturing the majority of the traffic dynamics.
- •
Inclusion of an additional class of vehicles namely, auto-rickshaws in the vehicle detection model.
The rest of this paper is organized as follows: Section 2 discusses the system architecture of the proposed local adaptive TLC system. Performance analysis of the customized deep learning models for detection of various classes of vehicles namely, two-three-four wheeler vehicles—from video frames under mixed traffic and varying congestion conditions is discussed in Section 3. Section 4 elaborates on the simulation set up for experiments and the performance analysis of the proposed adaptive TLC system. The feasibility of deployment of the proposed adaptive TLC is discussed in Section 5, and finally, Section 6 presents conclusions of this study and future directions.
Section snippets
System architecture
In this section, we elaborate on the system architecture (see Fig. 1) of the proposed adaptive TLC. The Sensing and communication infrastructure mainly captures the raw data of traffic dynamics of heterogeneous traffic and priority vehicles, which are further processed by the Data processing layer to compute the PCU (Passenger Car Unit) counts of every approach of a junction. The PCU count along with the data of the priority vehicles in the road network is then provided as input into the
Vehicle detection
Lately, deep learning-based methods have demonstrated immense success in detecting and classifying objects from images. These techniques remarkably outperformed the existing state-of-the-art conventional algorithms for object detection (Han, Zhang, Cheng, Liu, & Xu, 2018). The rapid development of Convolutional Neural Networks (CNN) (Krizhevsky, Sutskever, & Hinton, 2012) accelerated the use of deep learning-based techniques in object detection. Since CNN’s can learn high-level feature
Simulation
In this section, details of the simulations performed to evaluate the performance of our proposed TLC system are presented. Section 4.1 elaborates the simulation setup used for performance evaluation, Section 4.3 presents the effect of the proposed system on travel time and wait time of different types of vehicles (general traffic, priority vehicles, etc.), Section 4.4 assesses the robustness of the proposed system to the error in PCU count, and Section 4.5 presents the discussion of our
Deployment feasibility
In this section, we analyze the feasibility of deploying the proposed one-stop solution of traffic scheduling for all classes of vehicles, including priority vehicles, in real-world scenarios. The proposed algorithm for an adaptive TLC system can be executed in constant time and can be deployed on most of the available hardware devices. The majority of the computational complexity lies in the part of detecting vehicles using images from traffic surveillance videos. Although the deep learning
Conclusions and future work
This work proposes a one-stop solution for an adaptive TLC system, which not only is responsive to prevailing traffic conditions at junctions but also serves priority vehicles at a preference than the regular traffic. As a part of this system, a novel adaptive TLC algorithm is proposed in this work. To strengthen the applicability of the proposed TLC algorithm, a comprehensive simulation environment reflective of the typical vehicle composition in a city is set up in SUMO. Further, the
CRediT authorship contribution statement
Abhyudai Bisht: Conceptualization, Methodology, Implementation. Khilan Ravani: Data curation, Writing – original draft. Manish Chaturvedi: Visualization, Investigation, Reviewing and editing. Naveen Kumar: Supervision and editing. Shailesh Tiwari: Software, Validation.
References (50)
- et al.
A real-time traffic signal control system: architecture, algorithms, and analysis
Transportation Research Part C (Emerging Technologies)
(2001) - et al.
Stochastic adaptive control model for traffic signal systems
Transportation Research Part C (Emerging Technologies)
(2006) - et al.
Reinforcement learning for true adaptive traffic signal control
Journal of Transportation Engineering
(2003) Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
(2017)- Adam, G. K., Garani, G., & Ventzas, D. (2014). FPGA design of a camera control system for road traffic monitoring. In...
An efficient multiple-loop sensor configuration applicable for undisciplined traffic
IEEE Transactions on Intelligent Transportation Systems
(2013)- et al.
Reinforcement learning-based multi-agent system for network traffic signal control
IET Intelligent Transport Systems
(2010) Continuous selection of optimized traffic light schedules: A machine learning approach
- et al.
Edge-computing video analytics for real-time traffic monitoring in a smart city
Sensors
(2019) - et al.
A feasibility study on upgrading the static TLC infrastructure to adaptive TLC
Micro-auction-based traffic-light control: Responsive, local decision making
Traffic controller for ambulance and VIP vehicles
Annals of the Romanian Society for Cell Biology
Adaptive traffic management for secure and efficient emergency services in smart cities
Greater ahmedabad integrated mobility plan - Data collection and analysis
Greater ahmedabad integrated mobility plan - situation analysis interim report, vol. 1
Advanced deep-learning techniques for salient and category-specific object detection: a survey
IEEE Signal Processing Magazine
Routing emergency vehicles in arterial road networks using real-time mixed criticality systems
IoT-based traffic management system including emergency vehicle priority
Moving object detection on a vehicle mounted back-up camera
Sensors
Cited by (1)
Review on environmental aspects in smart city concept: Water, waste, air pollution and transportation smart applications using IoT techniques
2023, Sustainable Cities and Society