Orientation correction based monocular SLAM for a mobile robot

Abstract

We develop a new algorithm for SLAM that uses a PTZ-camera for visual observation of natural landmarks. A new divided JCBB algorithm is presented to quickly address the visual data association problem without bounding the number of image features. Additionally, the presented orientation correction method is employed to well address the difficult SLAM orientation problem. The presented algorithm can greatly improve the performance of map building by reducing almost all the repeated landmarks, and can make the SLAM loop process to be simple as the orientation error is bounded. Experimental results in real environment showed that the proposed algorithm was very fast, stable and accurate.