Active vision inspired by mammalian fixation mechanism.

Publisher Summary Focusing on the mammalian fixation mechanism, this chapter proposes a localization technique for object-detection in a scene and implements it into an active vision system. In the technique, large and small camera movements are coordinated to detect an object precisely at the optical axis of the camera. Each type of movement is performed via two separate localization methods: approximate localization and adjustment localization. In the former localization, a preliminary fixation point is detected using multiresolution retina. After the optical axis of the camera is pointed in the direction of this point, the latter localization method determines a precise fixation point by means of visual feedback control, which minimizes the angle between the object's center and the optical axis of the camera. Experimental results show that the active vision system using our proposed technique succeeded in fixating objects one after another in a scene, even if several objects are placed within it. According to these results, the chapter confirms that bottom-up processing, using primitive visual features, is an effective technique for localizing an object.