An active vision system for a location task using an inhomogeneously blurred image

This paper describes a simple method to detect a location of each object using an inhomogeneously blurred image, and the implementation of the method to an active vision system in order to execute a location task. In the proposed method, which is based on our previous theoretical study on the early vision, a camera-taken image is processed with high resolution near the optical axis of the camera, and with low resolution in the periphery. Using such an inhomogeneously blurred image, the objects are classified by neural networks with respect to their location in view . This method is implemented to our proposed mobile camera as the active vision system. The experimental results show that every object in view is easily foveated one after another. >