Abstract
The human objects detection is an important basis to many applications, such as the people counting and the loitering detection. We focus on the topic of the people counting. First of all, we extract the human objects depending on our proposed modified codebook model. Besides, we add a modified shadow removal method to overcome the illumination effect. In opposition to the wide tracking algorithms, we find the best correspondence in the history to solve the matching problem. We do not need to assume that people entering the scene are individual. In reality, the surveillance system is usually set in the crowded area, and we should concentrate on the occlusion problem. We classify the object detection into different steps and accumulate total number from the estimate number of each frame. In our experiments, we test our system to different conditions and it is effective to the people counting.