An End-to-End Traffic Light Detection Algorithm Based on Deep Learning

Traffic light detection algorithm, as an important algorithm in automatic driving and assisted driving tasks, is the focus of research. Because the traffic light is in the complex environment, it brings many difficulties to the detection algorithm. Traditional algorithms are based on color and geometric features and have many barriers. Therefore, a traffic light detection algorithm based on deep learning is proposed in paper. With YOLOv5 as the algorithm core and K-means to generate anchor, 63.3%AP and 143FPS detection speed are obtained on traffic lights in BDD100K data set respectively. The method in this paper achieves good results in speed and accuracy in an end-to-end manner.