This code compares three different disparity map calculation approaches, i.e., naive approach, dynamic programming approach, and the built-in OpenCV StereoSGBM approach, to a ground truth disparity map. Using the disparity maps, the point clouds are reconstructed including an oriented point cloud and a rudimentary triangulated surface mesh.
The algorithms are tested on the public Middlebury dataset.
- Working C++ compiler (C++ 17 or greater)
- CMake (version >= 2.8)
- OpenCV (version >= 4.0.0)
- PCL
./stereo <path to left image> <path to right image> <path to ground truth disparity image> <output file prefix> <window size> <lambda> <dmin>
 |
|---|
| Disparity map calculated using Naive approach |
 |
|---|
| Disparity map calculated using dynamic programming approach |
 |
|---|
| Disparity map calculated using OpenCV SGBM approach |
 |
|---|
| Point cloud from disparity map calculated using dynamic programming approach |