A Vision-based Taillight Intention Detection Method for Intelligent and Connected Vehicles
Luyao Du, Wei Chen, Changzhen Li, Bingming Tong, Donghua Zhang, Bo Liu · 2022
Vehicle taillight intention detection is crucial to the perception of intelligent and connected vehicle (ICV). In this paper, a vision-based method of vehicle taillight intention detection for ICVs is proposed. Based on the YOLOv4 network structure, an improved method is conducted for the problem that the vehicle taillight object is susceptible to background effects such as traffic lights. The proposed method firstly detects the vehicle object, and then classifies the taillight’s intention information within the coordinate range of the detected vehicle object, thereby reducing the interference of the taillight information caused by the traffic background. The experimental results on real scene dataset show that the proposed detection method has a 4.16% higher accuracy than the original YOLOv4 model, proving that the improved method can detect vehicle taillight intention information more effectively. In addition, the potential of the method for practical application is tested on the NVIDIA TX2 mobile terminal.