Hybrid Approach for Enhancing Traffic Signal Control with Computer Vision and Reinforcement Learning

The surge in traffic density has exceeded the capacity of current traffic signal management techniques. The existing traffic management technique is to change the traffic signal on a fixed interval according to a predefined sequence. Such a predefined sequence doesn’t take into consideration the various parameters of real-world traffic intersections such as real-time traffic, lane capacity, and pedestrian density. If the aforementioned variables are considered when switching the traffic signal, the waiting time for vehicles at intersections can be greatly reduced. In recent research, technologies such as object identification and deep learning have been put to the test to develop a traffic management system that does not overlook the above-mentioned parameters of an intersection. However, most existing research and implementation either employ a single technology or partially combine many techniques. Furthermore, several studies attempt to reinvent object-detecting methods that have already proven themselves. In order to accurately represent real-world traffic intersections and optimize vehicle flow through them, we have suggested a method that combines Reinforcement Learning (RL) with already-developed object detection and tracking algorithms.