Counting people in crowded scenes by video analyzing

Abstract

People counting has many important applications in practice. The two key techniques in video based people counting system are people detection and people tracking. Most of the current people counting methods use body detection or motion detection to detect the people, which can produce some good results in sparse situations but fail in crowded scenes. In the crowded scenes, we know that the head is the most probable target that can be full visual. In this paper, we propose a people counting method in crowded scenes by detection the head information from the video taken from a camera installed straight down on the ceiling. A head classifier based on boosted cascade of Statistically Effective Multi-scale Block Local Binary Pattern (SEMB-LBP) features is proposed for people detection. The detected head is then tracked by a model matching method using harr feature. Combining the head detection and tracking together, a people counting strategy is presented to count the number of the people in the video frames. Experiment results show that the proposed method work well and robust in crowded scenes.