| Description |
3D reconstruction from image pairs relies on finding corresponding points between images and using the corresponding points to estimate a dense disparity map. Today's correspondence-finding algorithms primarily use image features or pixel intensities common between image pairs. Some 3D computer vision applications, however, don't produce the desired results using correspondences derived from image features or pixel intensities. Two examples are the multimodal camera rig and the center region of a coaxial camera rig. Additionally, traditional stereo correspondence-finding techniques which use image features or pixel intensities sometimes produce inaccurate results. This thesis presents a novel image correspondence-finding technique that aligns pairs of image sequences using the optical flow fields. The optical flow fields provide information about the structure and motion of the scene which is not available in still images, but which can be used to align images taken from different camera positions. The method applies to applications where there is inherent motion between the camera rig and the scene and where the scene has enough visual texture to produce optical flow. We apply the technique to a traditional binocular stereo rig consisting of an RGB/IR camera pair and to a coaxial camera rig. We present results for synthetic flow fields and for real images sequences with accuracy metrics and reconstructed depth maps. |
