Simultaneous Localization and Traversable Region Mapping Based on Pedestrian Behavior Learning

Abstract

In human-robot coexist environment, autonomous mobile robots must solve the problem of traversable region perception and self-positioning. Hence proposes an algorithm of simultaneous localization and traversable region mapping based on pedestrian behavior learning. The algorithm samples the traversable region in the unstructured road environment based on pedestrian behavior learning, and uses region growing algorithm to obtain connected traversable region. On simultaneously localization and mapping, visual SLAM technology is combined with the detected pedestrian information and traversable region information. The feature points on dynamic pedestrians are excluded in the matching process, and the constant speed model + DLT algorithm and greedy search + DLT are used to speed up the matching process and enhance the robustness. In the Bundle Adjustment optimization process, different weights are given to different feature points to further improve the accuracy. The detected traversable region information is then incrementally added into the point cloud map. Experiments show that in a complex unstructured road environment, the algorithm can achieve accurately self localization and obtain traversable region mapping on the mobile robot platform.