Flexible Path Selection and Obstacle Detection for AGV

Abstract

This paper desbcribes an improvement of a conventional AGV in the following two points: finding a path in complex junctions, and recognition of obstacles on the path. A TV camera is employed to look ahead of the vehicle to find lines in the input image. Centers of the lines are detected by combining dynamic thesholding and differentiation. Center points are detected at several vertical positions, and the arrangement of the lines in the image is classified into several categories: a branch, a crossing, and so on. The second problem is solved by a sort of motion stereo. The motion is obtained from the image of the white lie near the potential obstacle. The height of the obstacle is computed and the decision is made if it is really an obstacle. Some experimental results are shown to verify the proposed methods. The relation to other research themes for realizing AGV for more general environments are discussed.