抄録
We previously reported on a new voting method of sight line intersections. To reduce computational load, the method partitions an image into subblocks and applies individually the conventional intersection voting technique to each set of subblocks which correspond to one another in image sequence. This paper proposes a new scheme which detects occlusion on the basis of the Epipolar constraint and MAD between subblocks, and then deals with occlusion and other subblocks separately, for the purpose of reconstructing 3-D structure more accurately.