Vehicle recognition algorithm based on improved YOLOV3

摘要

In view of the problems of traditional vehicle recognition technology in detection speed, accuracy and stability, an improved YOLOV3 target detection algorithm is proposed. First, improve the feature extraction network in YOLOV3’s network structure; then improve the detection ability of small vehicle regions by improving the loss function; then through the improvement of multi-scale detection, enhance the detection effect of small vehicle region images in real-time vehicle detection; finally Through the nonmaximum suppression algorithm, the local maximum search is performed to complete the precise identification of the vehicle. Through experimental verification, the results show that the method not only has better real-time performance and accuracy, but also has strong robustness in dealing with complex environments.