An intelligent traffic management framework using yolo driven vehicle detection and Q-learning based reinforcement signal optimization

The proposed work is intended to improve the functioning of traffic control systems via the implementation of a distributed multi-agent reinforcement learning (RL) framework, designed to bring traffic flow to optimal levels in urban environments. This undertaking seeks to use existing research on intelligent transportation systems (ITS) by incorporating superior machine learning methods to adjust traffic lights in real-time in response to changing traffic conditions. The main aim is to mitigate traffic snarl-ups, reduce waiting times, and improve the efficiency with which traffic is managed. We seek to create a scalable, adaptive traffic management system via multi-agent reinforcement learning. The process aims to make decisions based on its perception of current traffic conditions and considers each traffic signal as a separate entity. It also coordinates with adjacent signals to facilitate the proper flow of traffic within transportation routes. The proposed model demonstrates superior performance with an mAP of 82.4% and significantly higher FPS of ~ 60. Balanced trade-off between accuracy and speed provides a promising object detection solution. The work ensures real-time optimization by managing complex traffic patterns efficiently. The performance evaluated by the 4 lane states starting from (0, 2, 45, 43) maintains Q value between 6.9 and 11. Performance was mostly recorded with a discount factor of 0.5–0.11 and a learning rate of 0.1–0.4. Real-time response recorded in between ~ 300 to 500 ms with a scalability index of ~ 97%.