A robust approach for multi-human detection and tracking

Multiple human detection and tracking is a complicated task in real scenarios. We present in this article a robust approach for multiple human detection, tracking and identification. For human detection, an improved HOG human detector is used. In multiple human tracking, serving to inter-frame accordance, it is essential to generate distinctive signatures for each person. As global appearance description can be sensitive to illumination and intra-class variation, we therefore incorporate a local invariant feature (SIFT features) to increase robustness of matching. Both color-histogram and SIFT features are used in our approach. Experimental works demonstrate its effectiveness and robustness.