PURPOSE: To obtain a distance image which has no unevenness by finding the equation of a surface from inflection points of a right and a left images and obtaining the three-dimensional coordinates of the inflection points from the equation of the surface.

CONSTITUTION: An area division part 22 divides images picked up by cameras 20 and 21 into areas which have the same gradation and pattern (structure). A left label image 23 and a right label image 24 are supplied to an inflection point extraction part 25, the contour of each of surfaces of the label images is tracked to detect inflection points on the contour, and a correspondence part 26 makes inflection points on contours of respective surfaces correspond to each other between the left and right images. A surface equation calculation part 27 uses a least squares error calculation part 28 to calculate the surface equation by a least squares error method. Thus, the equation of the surface is found from the inflection points of the left and right images and used to obtain three-dimensional coordinates of the inflection points, so an error in distance due to a quantization error caused by parallax is not generated, distance information on the distance image is accurate, and the distance image having no unevenness is obtained.