Research On Traffic Signal Detection Algorithm Based On Improved YOLOv4

In recent years,with the rapid development of emerging technologies such as 5G and artificial intelligence,intelligent driving technology has gradually been applied in practice.As one of the key technologies in the field of intelligent driving,traffic signal detection technology can assist drivers in judging traffic conditions and reduce the probability of traffic accidents.Therefore,the study of high-precision traffic signal detection algorithms is of great significance for the development of intelligent transportation technology and the improvement of road safety.Aiming at the problem of misdetection,missed detection and overlapping detection of traffic signal light detection in complex scenes by YOLOv4 algorithm,the method of bottom-layer feature enhancement is used to redesign the network structure of YOLOv4,and a scale feature output is added to improve the algorithm’s detection of traffic lights.Experiments show that the new network structure can effectively reduce the false detection,missed detection and overlap detection in complex scenes such as strong light,long distance and night,and improve the robustness of the algorithm;for YOLOv4 The algorithm has the problem of large bounding box positioning error for traffic signal detection.α-CIo U loss is used as the new bounding box loss function;the DBSCAN and K-means clustering algorithm are combined to re-cluster the data set.Selecting new anchor parameters,experiments show that the new loss function and anchor parameters reduce the positioning error of the bounding box,and improve the regression and positioning accuracy of the bounding box.In order to solve the problem of poor quality of current mainstream traffic signal data sets,traffic signal images were collected on the spot and manually labeled,and the data was expanded by data enhancement methods,and a high-quality traffic signal data set was made.The network improved by the proposed method is used to train the data set to obtain the optimal model,and the test data is input into the trained model to complete the detection of traffic lights.The results show that the improved algorithm effectively improves the detection effect of traffic lights in complex scenes,reduces the positioning error of the bounding box,and has higher detection accuracy among the compared algorithms.