feature match ff_v0.1
The feature-match-ff software applies a sieve on matches based on the fundamental matrix estimation. The sieve itself is based on a threshold condition depending on epipolar line distance. In order to compute the fundamental matrix estimation, a set of good matches have to be given.
This software is designed to be used when information on camera poses are completely unknown. The strategy is to obtain a first set of matches that are strongly filtered, typically using successive sieves implemented in feature-match-seive. The estimation is the made considering the remaining matches. As the estimation is known, a second large set of matches are injected through the fundamental matrix sieve.
The feature-match-ff software expects the following arguments and parameters. Starting with the stream arguments and parameters :
--strict -s
Path to the matches file containing matches that are assumed
to be good.
--input -i
Path to the matches file containing matches on which filtering
is applied.
--output -o
Path to the output filtered matches file.
Fundamental condition arguments and parameter :
--tolerence -t
Maximum distance, in pixels, to the epipolar line.
Considering the following couple of images taken using an Eyesis4Pi camera in Montpellier city center :
and after grayscale conversion and exposure correction, a first set of keypoints are generated using SIFT algorithm on both images. SIFT parameters are set to obtain a small number of keypoints. The following images show the generated keypoints :
Using Flann matcher algorithm, a small amount of matches is generated. Applying two successive statistical sieves on disparities length and three successive sieves on disparities direction, we obtain the small set of good matches needed for fundamental matrix estimation. The two images below show the raw and filtered matches obtain through this process :
Having good matches to perform fundamental matrix estimation, a second set of keypoints is generated on both images using SIFT with parameters ensuring a large collection of keypoints. The second set of keypoints is shown for both image on the following images :
Flann algorithm is then used to extract matches from the second set of keypoints. The sieve implemented in feature-match-ff is then applied on this second set of matches using the filtered first set of matches for fundamental matrix estimation. The two images below show the raw matches obtained through the second set of keypoints and the result of the fundamental matrix filter :
One can see how it is possible to extract reasonable amount of good matches using this process. Missmatches will of course always remains and lead to outliers. Complementary statistical sieves can also be applied after fundamental matrix filtering to remove the remaining outliers.
The images that show matches are obtained using feature-match-view software.
The software binary can be build using its specific makefile by simply typing in its base directory :
$ make clean && make
The binary is placed in the local bin directory and is not copied in the feature-suite main bin directory. To generate the documentation, use the command :
$ make documentation
When documentation is generated, the previous one is preliminary removed.