Using MATLAB as auxiliary tool in visual localization for mobile robots

Abstract

A feature-based visual robot localization algorithm is generally responsible for estimating the robot pose based on features from images using cameras. To address the issues of estimation accuracy and required computational time, an improved visual odometry (VO) system is proposed using direct approach of perspective-3-point (P3P) algorithm as well as map management method. Furthermore, a GUI platform is designed using MATLAB so that performances of the proposed VO system can be easily observed by users. In order to validate the reliability of the proposed system, various experiments are conducted using a binocular camera, where experimental results show that the proposed VO system outperforms conventional algorithms in terms of estimation accuracy and runtime efficiency.