Correspondence by Tracking Edges in a Dense Image Sequence


Correspondence of two or more views of a rigid 3-D object is usually found by stereo. If the capturing conditions allow it, it is easier and more robust to do it by tracking in a dense view sequence.

Dense sequence of a rigid 3-D object (plaster cast of human teeth arc) is captured. Simplified epipolar geometry: epipolar lines = scanlines.

Epipolar image = intensities in a single scanline in different frames.

Edges detected and tracked in each row separately. The maximum interframe disparity is so small that the matching is unique --> the tracking is very simple.

Resulting correspondence between two selected frames.


Get research report and software (described in the report).
Page maintainer is Tom Werner.
<~ Go to the CMP Home Page