指差し動作認識及び風景認識による知的なカーナビゲーションシステムに関する研究

Abstract

We propose a car navigation system considering the passenger's knowledge of route guidance. Our approach is to recognize a hand signal of the passenger and a road scene by a CV (computer vision) technique. First, an inside camera on dashboard takes a video of the passenger. The hand signal is recognized by a CV technique, and a fingertip coordinate is calculated. The fingertip coordinate is transferred to the directed coordinate in the road scene taken by the outside camera on a front grille. The directed coordinate is utilized to calculate the directed region including a target object. A distance is calculated from the object, and the route is recalculated as a navigation system. In this paper, we experiment the fingertip coordinate recognition and the directed region calculation. Depth value from an inside camera to the passenger is used with Kinect. A road scene CG video is used for an ideal experiment. We ask seven examinees to indicate sign boards after camera calibration in advance. The successful ratio of the total is 47% because the view point of the user is not considered. The issue lists for practical system are concluded for future works.