hide forever  |   hide once

The objective of the Stereo lab is:

1. to examine area-based matching costs: Sum of Absolute Differences (SAD) (and Normalized Cross Correlation (NCC))
2. to build shiftable windows, and
3. to build coarse-to-fine strategy using your upsampling method

Software

• To access the necessary files for this lab, see directory /eecs/home/wildes/prism/4422/Lab4_Stereo
• You need to modify your code from Image Representation lab.


• To compile sampleOpenCV.c on CentOS,
  1. Link the library location in the binary
  % setenv LD_RUN_PATH /cs/local/lib64
  2. Compile the sampleOpenCV.c
  % g++ `pkg-config --cflags opencv` `pkg-config --libs opencv` sampleOpenCV.c
• To compile sampleOpenCV.c on Ubuntu,
% g++ `pkg-config opencv --cflags` sampleOpenCV.c `pkg-config opencv --libs`

Task

• Modify your code, sampleOpenCV.c from the Image Representation lab, such that it takes two images, LeftImageFileName and RightImageFileName, to visualize a disparity map:
  % ./a.out LeftImageFileName RightImageFileName [options]
  The disparity map can be computed by matching local regions between the left and right images using SAD (or NCC) on the Laplacian pyramid.
• When ./a.out LeftImageFileName RightImageFileName is executed,
  • 'ESC'-- exits the program
  • 'l' -- displays left pyramid
  • 'r' -- displays right pyramid
    
• Further modify your code such that it can accept the following optional arguments,
  • [Options]
  • -x [val] : local search range, by default "1"
  • -l [val] : pyramid levels, by default "1". NOTE: If you have more than one pyramid level, you should visualize all the levels.
  • -w [val] : aggregate window size, by default "5"
  • -maxDisp [val] : maximum disparity, by default "64"
  • -G : use the Gaussian pyramid instead of the Laplacian pyramid
  • -sw : use shiftable windows
• Example output using a Laplacian pyramid with 5 levels and a 20×20 window.

• Example output using a Laplacian pyramid with 5 levels and a 5×5 window.