USE OF AFFINE CAMERA MODEL AND ALL STEREO PAIRS IN STEREO-MOTION
Pin-Hsuan Ho, R. Chung · 1998
Two vision cues for recovering three-dimensional #3D# information of a scene from multiple images are stereo vision and visual motion. They have complementary advantages: while feature correspondence is simpler in visual motion, stereo vision o#ers more accurate 3D reconstruction. This paper presents an approach of combining the two cues to retain their advantages and remove their disadvantages, when a mobile platform with two #xedcameraheads is available to capture stereopair of image streams. It is shown that if a#ne cameras can be assumed, the image data, if organizedinaparticular way in a matrix, can be factorized into four matrices which represent the 3D structureof the scene, the image projection parameters, the motion parameters, and the stereogeometry respectively. With the factorization, there exists a linear method of inferring stereocorrespondences from motion correspondences, and both the 3D structure of the scene and the platform motion can berecoveredaccurately even with...