Stereo Vision Based Robot Controller

Abstract

This paper presents the stereo vision component of the stereo vision based robot controller (SVBRC). The SVBRC is a multi-component project, integrating a stereo vision component (presented here) with a basic mapping and path-finding component on a mobile robot base. The stereo vision component is a winner-take-all (WTA) method. It implements various error-filters to explicitly deal with various types of stereo correspondence errors. However, overly stringent error filtering leads to very sparse depth map output; conversely, lenient filtering provides a dense output map with a high percentage of pixels in error. In this paper, error filtering techniques from various research papers are examined and implemented concurrently in two stages, a stringent stage and a lenient stage. A unique match-pixel-growing method is then utilized with the aim of using stringent error filters as a base from which to dasiagrowpsila a dense depth map. The SVBRC has been implemented and tested in the laboratories of Monash University Malaysia for stereo-vision based obstacle avoidance and target-finding. The results of the SVBRC's stereo vision component are presented here.