ECE 564 Computer Vision Assignment 5

Submit HTML documentation and MATLAB code on Isidore.

  1. Use the images taken in class 13 Feb 2017 and the results of Q2 from Asgn 4, to generate figures similar to those for Room KL351G. Note our measured room dimensions are 22.25 ft wide, 37.36 ft deep, and 10 ft high. Try to use three images for the front wall (left center and right) and one image each for the side walls and rear wall.

  2. Use the images takein in class 6 Feb 2017, particularly those in CV2017_0206subset.zip, find the distance of the camera from each of the walls the the location of the corner with respect to the camera.

  3. For thes same images, find the associated vanishing points. The line connecting the vanishing points is the horizon line for the corresponding plane.

  4. Use fitgeotrans and imwarp to generate an ortho view of one of the checkerboard images. You may crop the results to show just the region of the target board. You can use pat2s.jpg to obtain reference points.

  5. Use the results of the previous problem to find the projective transform. Then read Hartley and Zisserman, Section 4.1, and write a MATLAB function for the direct linear transformation (DLT) algorithm. The result should match the matrix generated by fitgeotrans or cp2tform within a multiplicative factor.

  6. Study the Bouguet code visualize_distortions and apply_distortion and document the analysis performed. Modify the scripts so they may be used with MATLAB vision toolbox distortion parameters. Show how a large square in the center of the image would be distorted. Muliply the distortion by 10 or more so that the form of distortion, barrel or pincushion is evident. Write a script to plot radial distortion vs. radius. How would you calculate the distortion in percent?

  7. Some of the images taken in class 13 Feb 2017 are "duplicates" except for camera jitter. Show difference images between nominal duplicates. Do the differences appear to generated by translational motion of the camera or by rotational motion?

Reference

Richard Hartley and Andrew Zisserman, Multiple View Geometry in Computer Vision, Cambridge, 2000. ISBN 0-521-54051-8.


Maintained by John Loomis, last updated 1 March 2016