Abstract:

A YOLOv3-based multi-target detection method is proposed to address the problem of missed and false detection caused by the small percentage of target pixels and mutual occlusion in traffic scenes. The method implants a spatial pyramid pooling module in the YOLOv3 network structure to enhance feature representation, and a multi-scale feature fusion mechanism is proposed to obtain both spatial information and semantic information. The semantic information of the target to be detected is refined by extending the prediction branch of the prediction layer. In addition, the improved K-means + + clustering algorithm is used to extract the initial center of the priori box and improve the matching degree between the prediction anchor box and the target to be detected. Meanwhile, a flexible non-maximum suppression algorithm is applied to adjust the confidence score flexibly. The experimental results based on the hybrid data set show that the proposed method improves the detection accuracy effectively.

Key words: traffic multi-target detection,  feature extraction,  multiscale feature fusion,  predictive anchor box matching