Research on Vehicle Detection Based on YOLOv3

Abstract

In order to improve the recognition accuracy and detection speed of vehicle models, a vehicle model detection based on the improved YOLOv3 model is proposed. On the one hand, the SENet structure is introduced into the feature extraction network, and the weight optimization rules of features in the convolution process are added; On the other hand, the number of high-level repeated convolutional layers of the feature extraction network is improved to improve the time cost of the model in vehicle type detection. The test results on the BIT-Vehicle dataset show that the proposed method improves the recognition speed relative to YOLO v3 by about 24% on the premise of ensuring the vehicle model recognition accuracy, and verifies the effectiveness of the proposed target detection method.