Human Body Volume Recovery from Single Depth Image

We propose on-line human body volume recovery framework using only single depth image. Depth image contains partial 3d geometry information of human body surface seen from the sensor viewpoint. Previous volume reconstruction methods require multiple images from different viewpoints to reconstruct complete closed body volume. They have limitation in real-time application with dynamic objects or require multiple sensors. In this paper, we propose a generic model based human body volume recovery. First, we register the pose of the generic model to partial body surface observation from single depth image. And then remaining 3d points on unseen surface is optimized by propagating confidence of the partial observation. Experimental result shows our method captures reasonable human body volume from single depth image on-line.