Using deep learning technology and multi-object tracking method to realize highway vehicle counting is a hot research topic in the field of intelligent transportation. This paper proposes a method of traffic flow analysis with fast speed. First, a vehicle dataset from the perspective of highway surveillance cameras is constructed, and the vehicle detection model is obtained by training using You Only Look Once (YOLO) vision 3 network. Second, an improved multi-scale and multi-feature tracking algorithm based on Kernel Correlation Filter (KCF) algorithm is proposed to avoid the KCF extracting single features and single-scale defects. Combined with the Intersection over Union (IOU) similarity measure and the row-column optimal association criterion proposed in this paper, using the matching strategy to process the case where the vehicle is not detected and wrong detected, thereby obtaining complete vehicle trajectories. Finally, according to the trajectory of the vehicle, the traveling direction of the vehicle is automatically determined, and the setting position of the detecting line is automatically updated to accurately obtain the vehicle counting result. This article conducted experiments in a variety of traffic scenes and compared them with published data. The experimental results show that our method achieves high vehicle detection accuracy and maintains high vehicle tracking Distance Precision (DP) and Overlap Precision (OP), and obtains accurate vehicle counting results which can meet real-time processing requirements. The algorithm of this paper has practical significance for the application of vehicle counting in complex highway scenes.

CICTP