2D地図中の対象物検出に向けたオクルージョンに強いテンプレートマッチング手法の提案

抄録

We proposed and evaluated a robust method for occlusion to detect the reference object from a two-dimensional (2D) map acquired by SLAM (Simultaneous Localization and Mapping) so that a mobile robot running along a route can correctly stop in front of the reference object even if the room arrangement is changed. The stopping position of the conventional mobile robot had to be configured through trial and error, considering factors such as repetition accuracy, camera angle of view, and manipulator's motion range, which required man-hours at the work sites with frequently changing layouts. In this study, we proposed a system that recognizes an object that serves as a reference for the stopping position from a 2D environmental map acquired by SLAM and can take over the setting of the stopping position before and after room arrangement changes. The robustness of the proposed detection algorithm to occlusion was also evaluated.