semi-global-matching

An implementation of Semi-Global Matching (SGM) on CPU with some optimizations

Description

• An implementation of Semi-Global Matching (SGM) based on [1].
• The 9x7 Census transform or Center-Symmetric Census transform is used as matching cost.
• Optimized with SSE4.1 and OpenMP.

References

• [1] Hirschmüller, H., Accurate and Efficient Stereo Processing by Semi-Global Matching and Mutual Information, IEEE Conference on Computer Vision and Pattern Recognition, June 2005

Requirement

• OpenCV
• SSE >= 4.1
• OpenMP (optional)

How to build

$ git clone https://github.com/gishi523/semi-global-matching.git
$ cd semi-global-matching
$ mkdir build
$ cd build
$ cmake ../
$ make

How to run

./sgm left-image-format right-image-format

• left-image-format
  • the left image sequence
• right-image-format
  • the right image sequence

Example

./sgm images/img_c0_%09d.pgm images/img_c1_%09d.pgm

Author

gishi523