In this paper, image filtering method is proposed for the situation that the interframe displacement between the images is large. When the optical flow is measured from the images obstained by the cameras in different positions and with different orientations, the larger the interframe displacement of a point between the images is, the more optical flow measurement error increases because of the deviation from the assumption that the interframe displacement in an image is small when deriving the motion constraint equation.
By the proposed filtering method, first, the maximum interframe phase difference that minimizes the effect of large interframe displacment on optical flow measurement error is determined, and the real interframe phase difference is compared with the maximum interframe phase difference as frequency is increased, thus the cutoff frequency of Butterworth lowpass filter is determined.
In the experiment, the effectiveness of the proposed filtering method is investigated for the 1-dimensional synthetic images. And when this filtering method is applied to real images, the result of recovering the three-dimensional range is presented. As a result, three-dimensional range data is sufficiently well estimated at the area of objects in an image.