A New Fast and Robust Method Based on Head Detection for People-Flow Counting System

This paper presents an implementation method for the people counting system which detects and tracks moving people using a fixed single camera. The main contribution of this paper is the novel head detection method based on body’s geometry. A novel body descriptor is proposed for finding people’s head which is defined as Body Feature Rectangle (BFR). First, a vertical projection method is used to get the line which divides touching persons into individuals. Second, a special inscribed rectangle is found to locate the neck position which describes the torso area. Third, locations of people’s heads can be got according to its neck-positions. Last, a robust counting method named MEA is proposed to get the real counts of walking people flows. The proposed method can divide the multiple-people image into individuals whatever people merge with each other or not. Moreover, the passing people can be counted accurately under the influence of wearing hats. Experimental results show that our proposed method can nearly reach to an accuracy of 100% if the number of a people-merging pattern is less than six. Keywords-People Counting; Head Detection; BFR; People-flow Tracking