Ask Your Question

Entering >5 points into "findEssentialMat"?

asked 2016-03-07 08:36:42 -0600

gandalfsaxe gravatar image


I'm trying to figure out the relative pose (R, t matrices) of two cameras using two synchronized videos of the same scene viewed from different positions. As input I've selected points of the same object in corresponding frames of both videos, and findEssentialMat needs at least 5.

My question is: what happens if I feed in more than 5 problems? Ideally I'd like to feed in many more points than 5 points and get a more accurate estimate than just 5 points would give. That's what I hope. Otherwise I'd consider finding R,t from many pairs of 5 points of points, then taking the median or mean as the best estimate.


edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2016-03-07 18:55:03 -0600

Tetragramm gravatar image

More points is good, although it will likely take longer to run. Just add them to your input vector and call the function as normal.

The default parameters for the findEssentialMat function is to use RANSAC to remove outliers from the estimation. So if you have one point that is far off a correct match, it will have little to no effect on the results, but only if you have more than 5 points so that another can be used.

edit flag offensive delete link more


That's what I was hoping for, thanks. Do you know how MEDS is different and pros/cons compared to RANSAC?

gandalfsaxe gravatar imagegandalfsaxe ( 2016-03-08 13:26:33 -0600 )edit

Here's a good explanation of LMedS.

As for which is better? I don't really know. I suspect the differences are unnoticeable in most situations. If you have some test data, give both a try and see which gives you better error characteristics.

Tetragramm gravatar imageTetragramm ( 2016-03-08 16:17:29 -0600 )edit

Question Tools

1 follower


Asked: 2016-03-07 08:36:42 -0600

Seen: 514 times

Last updated: Mar 07 '16