Chapter Questions
Explain the difference between a bottom-up approach (object reconstruction) to 3D vision as opposed to top-down (model-based).
Explain the basic idea of active vision, and give some examples of how this approach eases vision tasks.
Give examples of perspective images from everyday life. Where do parallel lines in the world not correspond to parallel ones in images?
What are the intrinsic and extrinsic calibration parameters of a single-perspective camera? How are they estimated from known scenes?
Do zoom lenses typically have worse geometric distortion compared to fixed-focal-length lenses? Is the difference significant?
What is the main contribution of epipolar geometry in stereopsis?
Where are epipoles in the case of two cameras with parallel optical axes (the canonical configuration for stereopsis)?
What is the difference between the fundamental and essential matrices in stereopsis?
How are mismatches in correspondences treated in stereopsis?
What are the applications of epipolar geometry in computer vision?
Explain the principle, advantages, and applications of a trilinear relation among three cameras. What is epipolar transfer?
Why cannot we get depth if the camera does not translate?
Why cannot we get a correct panorama if the camera does translate?
Stereo correspondence algorithms are typically lost if the left and right images have large regions of uniform brightness. How can depth acquisition still be made possible?
Active range finders (e.g.4 with a laser plane) suffer from occlusions; some points are not visible by the camera and some are not lit. What are the ways of dealing with this problem?
What is Moiré interferometry? Does it give absolute depth?
Why is the relation between pixel intensity on one side and surface orientation, surface reflectance, illuminant types and position, and viewer position on the other side difficult?
Under which circumstances can the surface orientation be derived from intensity changes in an image?