Retinal Model based Plane Detection Method using Range Images for Unknown Object Recognition

摘要

This paper describes a robot perception of an unknown object for service robots using range camera. There are plane detection methods like RANSAC and Hough Transform to detect planes of unknown objects. However, the previous method has problems which are high computational costs and low-accuracy of small object detection. We propose the plane detection method based on retinal structure to solve aforementioned problems. The proposed method is constructed by simple plane detection based on retinal structure, and integrated object plane detection. The simple plane detection based on retinal structure is focused on small plane detection and reducing computational costs. The integrated object plane detection is focused on stability of detecting plane on the specific object. As experimental results, we show that the computational cost of proposed method is no more than 10% compared to previous plane detection method. And, the proposed method detects small planes of specific object. Furthermore we discuss the capability of proposed method which coordinate the ability of reducing computational costs and improving the plane detection accuracy.