Robot localization method based on visual features and their geometric relationship

摘要

This paper presents a novel method to recognize current location of a mobile robot from input stereo images with visual and geometric features. We extract structural planes from 3D depth data by using SLIC (Simple Linear Iterative Clustering) based superpixel and RANSAC algorithm. The experimental results show that the proposed method using visual features and their geometric relationship provides better performance in robot localization.