Head detection and localization from sparse 3D data

Abstract

Head detection is an important, but difficult task, if no restrictions such as static illumination, frontal face appearance or uniform background can be assumed. We present a system that is able to perform head detection under very general conditions by employing a 3D measurement system namely a structured light distance measurement. An algorithm of head detection from sparse 3D data (19×19 data points) is developed that reconstructs a 3D surface over the image plane and detects head hypotheses of ellipsoidal shape. We demonstrate that detection and rough localization is possible in up to 90% of the images. © Springer-Verlag Berlin Heidelberg 2002.