Mobile robot loop closing using monocular vision SLAM

This paper presents a design and implementation of simultaneous localization and mapping (SLAM) of a mobile robot using a monocular camera. The vSLAM system is based on techniques of extended Kalman filter (EKF) and scaling invariant feature transform (SIFT) algorithms. We propose in this paper a new method to discard outliers and improve the feature matching rate by using the characteristic of monocular camera. This method helps for the stability of EKF algorithm and allows more accurate robot localization. In this work, reference images are saved into a database and the current image is matched with reference images to improve loop closing performance of a mobile robot. Experimental results show that the proposed method effectively improves the feature matching rate and therefore the loop closing accuracy. Multi-loop indoor navigation experiments reveal that the proposed localization algorithm can help robot to navigate in an indoor environment and build the features map simultaneously.