This paper uses a method based on the YOLOv5s network to study the real-time traffic light detection task in the vehicle-mounted intelligent system. Due to the limitation of computing resources on the actual vehicle and the particularity of the traffic light detection task, this paper optimizes the YOLOv5s network and proposes an explicit multi-scale fusion module and a cross-long-order spatial attention module to improve the model’s detection ability for small targets such as traffic lights. At the same time, it reduces the number of channels and scales of the network, and improves the detection efficiency while ensuring accuracy. On the optimized network, by using public datasets and self-collected datasets for training, the loss of computing power of the on-board computing unit is reduced while ensuring high accuracy. Finally, on the intelligent connected vehicle platform, efficient and accurate real-time traffic light detection function is realized.
Research on Efficient Traffic Light Detection System in Intelligent Connected Vehicles
20.09.2024
46093125 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
British Library Conference Proceedings | 2022
|Estimating Freeway Lane-Level Traffic State with Intelligent Connected Vehicles
Transportation Research Record | 2022
|