Ask Your Question

Implementing SGBM block matcher in CUDA

asked 2017-03-16 07:50:23 -0500

mirnyy gravatar image

updated 2017-03-16 07:52:16 -0500


i just wanted to ask about your advise.

I need to use the SGBM (semi-global block matcher) to get a accurate disparity map. The SBM (stereo block matcher) is delivering quite bad and unuseful results.

The problem is just, that it seems that SGBM is not implemented in OpenCV CUDA yet (but the SBM).
But i need the SGBM to work faster (for example on the GPU).
Would you say it is possible to implement it in CUDA in a few days with not much experience in CUDA C programming?
And is it btw a good idea to implement it, or are the for example already good alternatives for it?

edit retag flag offensive close merge delete

2 answers

Sort by ยป oldest newest most voted

answered 2017-03-16 08:03:12 -0500

edit flag offensive delete link more

answered 2018-12-02 19:20:33 -0500

jasonzhang gravatar image

updated 2018-12-02 19:29:35 -0500

see sgm cuda

Opencv does not implement the gpu version of the sgm algorithm, this project is based on the opencv sgm algorithm. After my test, it is 3-4 times faster than the cpu version of the opencv sgm algorithm, and the effect is exactly the same as opencv sgm. The maximum disparity can be set to an integer multiple of 32, the maximum is 256.

edit flag offensive delete link more


Replace sgm with sgbm

jasonzhang gravatar imagejasonzhang ( 2018-12-03 00:47:40 -0500 )edit

Question Tools

1 follower


Asked: 2017-03-16 07:50:23 -0500

Seen: 1,801 times

Last updated: Dec 02 '18