Up to now we have shown the affine reconstruction of curved surfaces. In this section, we show that the affine reconstruction can be used for distinguishing apparent contours from fixed features.
Substituting (
13
) into (
1
) and using (
8
) and (
12
), we find that the reconstructed contour generator,
, at time
can be projected back to a point,
, in the original image at time
without any ambiguity. We now compute the differential component,
, between the projection,
, and the contour curve,
, in the image at time
along the epipolar line. If the image curve is of a fixed feature, the
projection,
must coincide with
, and
vanishes. If the image curve is an apparent contour,
does not coincide with
, and
is not equal to zero. Thus, the magnitude
is a good measure for distinguishing apparent contours from fixed
features. Even if the camera is uncalibrated, we can distinguish
apparent contours from fixed features by using the projection of
reconstructed contour generators back into the images.
J. Sato