Single camera vision pointing recognition for natural HCI

A novel method is proposed to analyze pointing recognition based on a single un-calibrated camera. Face skin information is explicitly adopted as view quality in selecting the best view from multiple cameras. View measurement has been developed to overcome the effects from camera view directions and the distances between the user and the camera. No explicit constraints are set on the camera placement. Pointing users can move freely inside a wider interaction environment while pointing at some targets. The mentioned model does not constrain whether the pointing surface is flat or not. Experiments have highlighted that the proposed method is efficient and the performance of pointing recognition from the optimal viewpoint camera outperforms other cameras.