People counting base on head and shoulder information

This paper presents an application for counting the people who pass through the supervised area. Instead of traditional camera, this study used Kinect 2 to get the depth information of image. The processes of our approach includes preprocessing, candidate detection, tracking, identification and people counting. In the preprocessing stage, the foreground object was sliced by depth information to make detection result more robust and to reduce the computation time. In the candidate detection stage, Hough Circle Transform was applied on color image to find candidates and depth image. Calculating pixels by a circle can decide whether candidate is people or not. Finally, the results of secondary stage provide the candidate's center coordinates that was used by nearest point tracking method to track path in 30 fps.