Research On Detection And Recognition Method Of Traffic Light Light Based On Deep Learning

In recent years,with the rapid development of emerging technologies such as big data,5G,and artificial intelligence,intelligent driving technology has gradually been applied in practice.Intelligent driving technology can not only relieve the fatigue of drivers,but also reduce their energy dispersion and reduce the probability of traffic accidents.Traffic light detection and recognition technology is one of the core technologies in the field of intelligent driving,and its detection accuracy and detection speed are particularly critical.Therefore,the study of traffic light detection and recognition methods based on deep learning has important research significance for traffic road safety.Aiming at the problem of unbalanced samples in traffic light data and insufficient background of signal lights,a variety of data enhancement methods are used to expand and enhance traffic light data,simulate traffic lights in different weather and different backgrounds,and solve the problems of samples between categories The problem of imbalance.Use a variety of single-stage object detection algorithms and two-stage object detection algorithms to conduct experimental analysis on traffic light data,and select object detection algorithms that can meet high-precision real-time detection and recognition of traffic lights.The experimental results show that the YOLOv4 algorithm can meet the real-time detection of traffic lights with high accuracy.Therefore,this paper chooses the YOLOv4 algorithm for the detection and recognition of traffic lights.In order to strengthen the sensitivity of YOLOv4 algorithm to small object detection,a traffic light detection and recognition method based on YOLOv4 algorithm is proposed.According to the characteristics of the traffic light data set,cluster analysis is performed using the K-means++ clustering algorithm to calculate the anchor parameters suitable for the traffic light data,thereby modifying the number and aspect ratio of the prior candidate frames of the YOLOv4 network.Make it more targeted for the detection of traffic lights;improve the YOLOv4 algorithm,add a shallow feature enhancement mechanism and a bounding box uncertainty prediction mechanism,and improve the sensitivity of the YOLOv4 algorithm to small object detection;train traffic lights The data is input to the improved network,and the optimal model is obtained through multi-scale training of the GPU,and the signal light test data is input into the trained model to complete the detection and recognition of traffic lights.The LISA traffic light data set is used for comparison experiments.The experimental results show that the proposed method has higher detection accuracy in the detection and recognition of traffic lights than the compared method,and can realize the real-time detection and recognition of traffic lights.