Human detection for an interactive art in a dark environment using the GPU hardware

Abstract

Human detection is the initial step of human recognition for an interactive system. A depth sensor is used to detect audiences in a dark environment, such as a media art exhibition or a theater. The depth data are converted to three-dimensional points using camera parameter. The detected human and the background are separated using the position of converted points. As these processes require much computation, in this study, we reduced the amount of computation time for real-time human detection systems through the GPU multi-threaded programming. The head and body are separated using the geometric characteristics of a human shape. The facial components are classified by analyzing projected curves of x- and y-coordination and by applying them to the geometrical characteristics of human face. © 2014 SERSC.