走行位置に依存しない背景差分による走行路上の障害物検出

Abstract

An aged society is approaching in Japan. Electric-powered mobile scooter becomes widely used for the supports of livelihood and locomotion of aged. For support of the driving of the mobile scooter users, Obstacles on the road or pedestrians should be detected to avoid collision accidents. This paper shows the road map generation for detection of the obstacles. Birds-eye view images are generated from input images with perspective transformation. The perspective transformation matrix is varied according to the photography scene. Which is estimated with four pairs of points which show the corresponding pixels on the birds-eye view images and input images. We propose the decision method of a coordinate position of four points. First, the straight line of edges and the vanishing point are estimated then, Four points are picked up on radial lines from the vanishing point. Four points of the next frame are estimated according to the movement of a camera. The new perspective transformation matrix is provided using the estimation point. Road map is generated with concatenated birds-eye view images. Road map used the detection of the obstacle.