Image Analysis of Human Staying based on Background Alternation

Abstract

A new method for analysis of human behavier in public spaces like an atrium is proposed. There have been some difficulties in analysis of human staying, because that it is difficult to discriminate human staying from stationary background scene. In this paper, general human behavier in staying can be analyzed and modeled by short term brightness fluctuations caused by arm or leg motions or pitching around the stationary bodies. In the background subtraction scheme, background alternation can be detected robustly by using a LMedS estimation scheme and utilized for detection of the short term brightness fluctuations. An image processing technique is developped for detecting them from observation images and estimating stationary.bodies by extending the fluctuations into neighboring regions. Detected regions can be filtered by a robust texture descrimination shceme as human motions or other ones such like illumination change or shading. After this, human staying regions can be accumulated to obtain stable estimation of the staying characteristics. Some fundamental experiments utilizing real data from an atrium are performed and the effectiveness of the proposed method is verified.