Disparity Map Generator

Stereo Matching

When our eyes view a scene, the left and right eyes see slightly different things. The left and right images are known as a stereo pair. By matching the corresponding points in the two images, we are able to infer the depth of the scene.

The paper below proposes using a dynamic programming algorithm to perform stereo matching, which I have implemented in C++.

Cox, I.J., Hingorani, S., Maggs, B.M. and Rao, S.B. (1992). Stereo Without Disparity Gradient Smoothing: A Bayesian Sensor Fusion Solution. Procedings of the British Machine Vision Conference 1992. [online] Available at: http://www.bmva.org/bmvc/1992/bmvc-92-035.pdf.

Usage

Both files can be compiled using the shell script which can be ran with ./compile.sh.

To generate a disparity map for two images (same size)

Execute ./disparity <LEFT_IMG_FILENAME> <RIGHT_IMG_FILENAME>. Pairs of images to test with can be found in the Stereo Pairs directory

To generate a random dot test case

Execute ./stereogram to generate two random dot stereograms, left.png and right.png in the same directory as the program.

These can be passed as arguments like so ./disparity left.png right.png to generate a disparity map for the random dot stereograms.

Prerequisites

• clang with C++11 support

• opencv4