Edge-based Extraction of a Grasped Object with Retina-like Sensor

This paper proposes a novel extraction method of a grasped object with a retina-like sensor. The grasped object extraction is necessary for the object perception in the interactive and practical scene. Because, the vision system of robots often faces with registering the object, which is displayed by humans hands or own hands. To realize this task, we use a retina-like sensor. Retina-like sensor has a space-variant resolution imitating the distribution of photoreceptors in the human retina. The proposed method employs the feature of a log-polar image of a grasped object displayed on retina-like sensor. The log-polar image obtained from the retina-like sensor changes significantly according to the object position. In the log-polar image, the exterior edge of the grasped object draws a continuous curve from left side of the image (0 degree), to the right side (360 degree), if the object is located in the central area of the retina-like sensor. The proposed method extracts the exterior edge, and extracts the object region from background. The validity of the proposed method is examined through the experiments using retina-like sensor.