Computer Science 4442A/B Lecture Notes - Lecture 17: Camera Resectioning, Depth Perception, 3D Reconstruction
10 Mar 2016
School
Department
Course
Professor
Document Summary
Stereo: cues for 3d reconstruction, stereo cues, stereo reconstruction, camera calibration and rectification, an easier, mostly solved problem, stereo correspondance, a harder problem. 2d images: in 2d images, depth is largely lost (the third coordinate, includes the human retina, depth is ambiguous from a single view, however, we still perceive the world in 3d. What is needed for stereopsis: monocular cues, object cues, an image with none of the above is simple a bunch of random dots. 3d shape from stereo: we use two cameras, and stereo cameras are not on the same plane, better numerical stability, we encounter triangulation. Formula: depth from disparity: the top down view on geometry (a slice through the xz plane, from camera calibration, we know the distance between camera optical centres (baseline b) and the camera focal length f. X left optical center baseline b right optical center: the height to base ratio of triangle angle c.
