Abstract
We do research on an object recognition system for a service robot that can help humans by getting objects by a user's demand. In this paper we propose a scheme to segment and detect individual objects. At first, we divide foreground including objects from backgrounds by filtering based on hue variances in a HSV color space. This filtering also has the effect to remove redundant edges like ones of printed figures on the object surface. After edge detection operation is executed, the image is divided into some regions separated by some edges. Yet, segmentation is not perfect. The ambiguity of regions on the edge still remains. Thus, we propose the robot with an arm touches or taps objects. The visual information changes as a result of physical contacts between the robot and objects. This change makes region separation clear.
