A method for detecting an obstacle by using two cameras, the obstacle
being on a reference plane in a field of view common to the two cameras,
the method inputting a first image and a second image from the two
cameras, respectively, transforming the first image to a transformed
image, obtaining a degree of similarity D indicating a similarity between
an image in a processing region established in the second image and an
image in a corresponding processing region established in the first
image, obtaining a degree of similarity P indicating a similarity between
an image in the processing region established in the second image and an
image in a corresponding processing region established in the transformed
image, detecting the obstacle based on of the degree of similarity D and
the degree of similarity P on a reference plane region, obtaining a
difference K between the degree of similarity D and the degree of
similarity P both obtained as being on the reference plane region,
judging that the obstacle is at a position where the difference K is
large, and determining a width and a height of the processing region on
the basis of a position in the vertical direction on the second image.