Abstract
Autonomous navigation in outdoor environments requires self-localization capability. This report describes an application of a view-based localization method to mobile robot navigation. Our localization method uses a two-stage SVM-based localization for increasing robustness to change of weather and seasons. It also adopts a Markov localization approach for an efficient and reliable localization. In applying the view-based method to mobile robot navigation, the change of robot heading during motion may degrade the localization performance. To cope with this problem, we use a panoramic image to search a range of orientations for a matched view. Additionally, a navigation course is divided into straight courses by corner. We also deal with a corner localization by examining views corresponding to the heading before and after turning the corner. We validated the effectiveness of the method using actual image sequences.
