Detection of obstacle from monocular vision based on histogram matching method

This paper proposes a detection method of an unknown obstacle trespassing upon the field of view from a monocular vision installed in an autonomous land vehicle which moves in the unknown static environment. Taking account of the motion restriction of the vehicle, we investigate the vanishing point of motion and the direction of each optical flow as the motion vector observed on a 2D-image. Then, in the case of no obstacles, it is shown that the vector field forms a uniform radial configuration centered at the vanishing point of motion. In this paper, the obstacle detection is performed by extracting a certain disturbed region in the vector field caused by the trespassing obstacle. The normalized histogram matching method suitable for obtaining the vector field is introduced. The experimental result for the real images in the case of an obstacle trespassing upon the field of view is shown