作者: Neil Robertson , Ian Reid
DOI: 10.1007/11744047_31
关键词:
摘要: In this paper we describe a new method for automatically estimating where person is looking in images the head typically range 20 to 40 pixels high. We use feature vector based on skin detection estimate orientation of head, which discretised into 8 different orientations, relative camera. A fast sampling returns distribution over previously-seen head-poses. The overall body pose camera frame approximated using velocity body, obtained via automatically-initiated colour-based tracking image sequence. show that, by combining direction and head-pose information gaze determined more robustly than each alone. demonstrate technique surveillance sports footage.