A vision-based mobile robot localization method

A vision-based method is proposed in this paper for mobile robot localization. The proposed method mainly consists of two parts: mapping and localization. First, to build the map, robot moves randomly in the target enviroment and captures images using its camera. The SURF descriptor is used to extract the features of each image. Then map can be built with the SURF features and their positions. For the localization phase, a new image is obtained at each unknown position when the robot walks in the enviroment. SURF and FLANN are utilized to match the features of the new image with those in the map to infer the position of the robot. Besides, an improved RANSAC method is employed to reduce the outlier matches. Experimental results show that the proposed localization method performed well in both static and dynamic localization tasks.