As urban traffic pressure continues to increase, existing traffic systems have achieved some success in improving vehicle throughput. However, they commonly face issues such as insufficient precision in visual recognition models, traffic signal control algorithms that neglect pedestrian needs, and excessive reliance on hardware infrastructure, leading to high deployment complexity. To address these challenges, an intelligent traffic system based on an improved YOLOv8 model and Deep Q-Network (DQN) algorithm is proposed. The system consists of a road dataization module and a traffic signal control module. In the road dataization module, the improved YOLOv8 model incorporates an attention mechanism to efficiently recognize vehicle information and uses OpenCV to process traffic images, enabling accurate data modeling. In the traffic signal control module, the enhanced DQN algorithm dynamically adjusts signal durations to balance the needs of pedestrians and vehicles, thereby improving traffic efficiency. Verified through the SUMO simulation platform, this approach demonstrates significant ad-vantages in optimizing traffic flow and throughput. Furthermore, by leveraging algorithmic data analysis of the environment, the system reduces dependency on hardware infrastructure, significantly lowering deployment complexity and showing strong potential for practical application. Additionally, the design of a visualized front-end interface and an embedded display system enhances the system's usability and integrity, validating its feasibility.
Intelligent Traffic Signal Control System Based on Improved YOLOv8 and Deep Reinforcement Learning
13.12.2024
1930506 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Deep Reinforcement Learning-Based Traffic Signal Control
ASCE | 2024
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Deep Reinforcement Q-Learning for Intelligent Traffic Signal Control with Partial Detection
Springer Verlag | 2023
|