stereo-matching

Dense disparity estimation via local stereo matching

Overview

The code implements the local color-weighted disparity estimation algorithm and evaluates its performance on a set of stereo image pairs. The algorithm includes the following steps:

1. Cost function calculation
2. Cost aggregation based on:
   1. Box filtering
   2. Gaussian filtering
   3. Local color-weighted filtering
3. Winner-takes-all (WTA) disparity estimation
4. Detection of occlusions
5. Computation confidence values for disparity estimates
6. Post-filtering to tackle occlusions and bad pixels (needs map from 4. and 5.)

Performance Evaluation of the algorithm

• Calculation of BAD quality metric given the true disparity map
• Comparison of the performance of block matching and Gaussian smoothing for different sizes of aggregation windows
• Comparison of local color-weighted filtering against the two others above
• Visual assessment of the effect of occlusion filling

Results:

• One slice of cost volume

• Left disparity estimation without aggregation

• Left disparity estimation with block aggregation

• Left disparity estimation with gaussian aggregation

• Comparison of block and gaussian filtering

• Left and right disparity estimation with gaussian aggregation

• Left occlusion map

• Left confidence map

• Left compansated (filled) occlusion map

• Comparison of color-weighted aggregation and the others

• Left disparity estimation with color-weighted aggregation