Fusion of depth and vision information for human detection

In this paper, we present a novel approach to detect human base on the fusion of depth and vision information. Instead of conventional human detection method which is mostly done in color image, our approach not only detects human and annotates their position but also extracts whole body pixels from depth image. Our approach has huge advantage over traditional vision image based method, since depth feature is robust against illumination changes, discriminative to background clutter and stable to human pose variation. This system contains a model based human head detector, which detects head through fusion of depth and color haar feature based detector. Then we propose a graph based algorithm to extract whole human body from depth image based head pixels. The experimental results on a large real-world dataset demonstrate that our method has an outstanding performance. © 2013 Binary Information Press.