Obtain Stereo-vision disparity map using feature desriptors-SIFT, ORB
Matching algorithm for SIFT: Flann based kd-tree
Matching algorithm for ORB: Flann based Locality sensitive hashing
The disparity, disparity map, RMS, Bad pixel percent calculations are implemented.
We compute disparity map for leftimage only.
Example : Input data: Cones
Output Data: cones_sparse or cones_dense
Each Input folder contains 4 files:
1. Left stereo image - left.png
2)Right stereo image - right.png
3)disparity Ground truth for Left Image - left_gt.png
3)disparity Ground truth for Right Image - right_gt.png
Each output folder contains 4 images:
1)disparity map obtained by SIFT - disparity_SIFT.png
2)disparity map obtained by ORB - disparity_ORB.png
3)Good matches between keypoints using SIFT - GoodMatches_SIFT.png
4)Good matches between keypoints using ORB - GoodMatches_ORB.png
Performed the experiment on 3 datasets of middlebury stereovision----> Cones, Teddy, Art
Default: Sparse
To do dense disparities: Uncomment the lines 161,162 and comment the lines 158,159 in main()
Do not run dense disparity on "Art" dataset. Due to its higher resolution, takes a lot of time(in hours)
and is not recordable in the lower-medium PC's.
Running the code by all defaults produces cones_out_sparse, Teddy_out_sparse, Art_out_sparse
1)code folder : contains the project files and source code.
2)Input Folder : Contains the three datasets which are used for input
3)Output Folder : Contains the sample outputs(already generated by student) for the input datasets
When the code is executed new output folders are created within the Output folder.