Abstract:This paper proposes an effective pedestrian tracking algorithm based on multi-algorithm fusion to address potential issues in the task of tracking a specific pedestrian in complex backgrounds, particularly when there are multi-scale targets, deformed targets, occluded targets, and long-time tracking targets. The algorithm uses MobileNet to lightweight the YOLOv3 detection network, and integrates the detection information acquired at regular intervals with the KCF tracking information, so as to quickly update and fix the target pedestrian position. Additionally, the algorithm proposes a target selection strategy to determine the precise location of the occluded pedestrian, and establishes an occlusion update strategy to address any possible continuous occlusion. The contrast tracking experiments in multiple scenes show that the proposed tracking fusion algorithm has good real-time and robustness.

An Efficient Specific Pedestrian Tracking Fusion Algorithm