Automatic Optimal View Selection for Natural HCI

Multiple cameras can be used to detect a certain object in a wide area so that users can freely move inside a wider interaction environment. One key problem is how to select the most appropriate camera with better view to perform a natural human-machine-interaction (HCI). An unsupervised scheme is developed for a best view selection. Face skin information is explicitly employed as view quality. The proposed view measurement has been defined which facilitates the system to select the most appropriate camera automatically. The proposal overcomes the effects of camera view directions and the distances between the user and the camera in selecting the best view. The approach is tested with different face databases and an actual interaction environment which shows that it is flexible and computational cost-effective. Experiments have highlighted that the mentioned method is robust and efficient when selecting the best view in a multi-camera system.