Mobile Robot Global Localization Based on Model Matching in Hough Space

This paper presents a global localization method based on model matching in Hough space. The classical Hough transform is introduced to solve this problem. To implement global localization with known environment models, a local map is firstly built via the vision system. Then the matching between known map of the environment and a local map is performed in the Hough space. By exploiting the decomposability of Hough transform and the environment model correlation, a set of possible poses represented by Gaussians is computed. By considering their covariance matrices and probability distribution as well as the information in the reference map, some inaccurate poses are discarded. The technique is especially suitable for structured environments. Experimental results validate the favorable performance of this approach.