1P1-M07 見え情報に基づくパーソナルナビゲーションシステム(移動ロボットのための視覚)

概要

論文の詳細を見る
This paper discribes a vision-based personal navigation in outdoor environments. Since localization is indispensable for navigation, we use a view-based localization method. This method compares the current image with an image sequences taken while moving along a route to follow, and determines the current location as the one where the best-matched image was taken. For the image comparison, we choose a set of locations on the route and generate a classifier for each location using SVM (support vector machine). The orientation of a camera continuously changes when the person moves with attaching the camera to his/her head or body, and this may cause a matching failure. We cope with this problem by using an omnidirectional camera. By extracting and testing perspective images in several directions in the omnidirectional image, we search for the direction for which the best matching is obtained. We also deal with the navigation for routes with corners by extracting images for the direction to turn. We developed a personal navigation system consisting of a notebook PC and an omnidirectional camera, and validated the on-line navigation capability of the system in an actual outdoor scene. It is also shown that the system can cope with the orientation changes of the camera and the navigation at corners.
2011-05-26