Auto target detection for robot hand manipulation using an anthropomorphic active vision system

In this paper, we propose a very simple method to detect an object from a camera-taken 2-D image when many objects are presented in a scene. The advantage of this method is its simplicity for use. In this method, a camera-taken image is blurred with high spatial resolution near the optical axis of the camera, and with lower resolution in the periphery. Such inhomogeneous image processing, which resembles to a primate's vision, is very effective to determine the position of each object in a scene one after another. This method is implemented into our proposed active vision system which send the position data of each target object to a robot hand controller in order to assist hand manipulation. In experiments, we show that the position of each object is detected quite easily and precisely enough for robot hand manipulation. >