A Robust Long-Term Pedestrian Tracking-by-Detection Algorithm Based on Three-Way Decision

Abstract

Pedestrian Detection Technology has become a hot research topic in target detection field recent years. But how to track the pedestrian target accurately in real time is still a challenge problem. Recently deep learning has got the extensive research and application in both target tracking and target detection. However, the tracking effect based on deep learning needs to be improved in the motion blur and occlusion cases. In this paper, we propose a new model that combines the target tracking and target detection and introduce the idea of granular computing to realize high-precision long-term robust pedestrian tracking. In this model, we use a pre-trained tracking model to track the specified object and use the three-way decision theory to judge the color histogram feature and correct the results by the detector. Compared with the separated tracker, our model invokes the target detector to detect the current frame when the tracking result is wrong and the detection result which is the most similar to the target is selected as the tracking result. Experimental results show that our model can significantly improve the tracking accuracy especially in the complex situations, compared with the separated tracker and the detector.