トポロジカルマップを用いた自律走行のための画像に基づく地点認識

抄録

Autonomous mobile robots have been an active research recently. Although laser range finder based navigation is known to work well in limited environment, expanding 3D geometric maps used in those methods to worldwide requires great effort. The same is true of maintaining them up-to-date. For this reason, we develop a visual navigation method using a topological map to reduce prior information and enhance scalability of the map. Topological maps only contain information about impressive places (i.e. intersections) and their linkage with roads. In the topological approach, the robot conducts three primal tasks: 1) road following between impressive places, 2) recognizing impressive places, and 3) taking action at impressive places (e. g., road selection). This paper proposes a visual place recognition method based on GIST descriptor and Support Vector Machine. The proposed method is evaluated at 13 point in outdoor walkways under several situations. As the result, the proposed method could identify all of 13 recognition points.