Motor vehicle obstacle monitoring system using optical flow processing

Abstract

A monitoring method for moving vehicles allows the presence of danger to be determined from the magnitude of an optical flow vector. The optical flow vector is detected as a movement of a single point on an object taken over two images, with one image taken at a preceding time and the other image taken at succeeding time in a series of images. A long and narrow window set in a radial direction from a focus of expansion (FOE) of the earlier image is moved in the same direction on the latter image. An optical flow vector of a target point is defined by a vector connecting the midpoint of a location of a subsequent window and the midpoint of a location at which the long and narrow window is set in the earlier image. The location of the subsequent window is determined by minimizing the sum of absolute values of differences in luminance between the long and narrow window and an area of the latter image overlapping such long and narrow window. The optical flow calculation area is limited to areas in which a difference in luminance between the earlier and latter images exceeds a predetermined threshold, and any influence from scenes outside the road boundaries and from lane-dividing lines or other symbols drawn on the road surface is removed in calculating the optical flow.