[line_detection] compare RANSAC line fitting with Hough transform

chicagoedt / revo_robot

Code for EDT's IGVC entry, Revo.

http://www.igvc.org/

20 stars 23 forks source link

[line_detection] compare RANSAC line fitting with Hough transform #53

Closed bsubei closed 9 years ago

bsubei commented 9 years ago

Try implementing RANSAC using the opencv fitline method according to this SO answer.

bsubei commented 9 years ago

"If you convert your output as it is right now to a pointcloud, the PCL library has a RANSAC method already implemented so you wouldnt need to do anything by hand" - Krystian

l0g1x commented 9 years ago

[ ] Try implementing RANSAC using the opencv fitline method according to this SO answer.
[ ] try converting current output of line detection into pointcloud2
[ ] apply RANSAC to that pointcloud2

bsubei commented 9 years ago

I wonder whether not knowing how many lines we have (either one or two) will be a problem for RANSAC. Maybe we have to use line clustering instead like in #54 to figure out how many.

l0g1x commented 9 years ago

@basheersubei have you tried implementing a RANSAC model yet?

bsubei commented 9 years ago

yes, it didn't work well. The problem is mainly with the filtering stage that happens before this. Too much noise

https://github.com/basheersubei/Software_IGVC/commit/313e58e58c776417de79f7bb73ca785b5512383e

bsubei commented 9 years ago

no need for line fitting, we will just throw the output of image pixels into a pointcloud