Mirrobot: Fast Detection of Body-Parts in the Scale-Space of Range Images with an Application to Human Motion Copying

In this paper we present a novel approach to fast detection of human body-parts that relates the idea of scale-space to range images. The resulting method is a pre-detector that extracts end-points at different scales that may correspond to the human's head or hands. The method is aimed to initialize more sophisticated human kinematics estimators and can be enhanced with known face or hand detectors based on color or grey-level imaging sensors. We apply our approach to the human motion copying problem (mimicking) and report on a set-up shown on two fairs in Germany in 2006. In the set-up an android robot is used to mimic observed human upper torso movements. We conclude by discussing results and point to further work