3-D Reconstruction Based on Projective Invariants

We present a new projective invariant-based technique for recovering depth information from stereo images. The basic idea of our method is to use intensity segments as primitives for stereo matching. To recover the depth information of intensity segments, two projective invariants are introduced. From the two invariants, we can derive a relation that should be satisfied between the matching pairs of intensity segments on the stereo images. By using the relation, the matching computation of the intensity segments can be performed easily. The proposed method can obtain a dense and accurate depth map directly from the input images. It is robust to both image distortions and occlusions.