RANSAC - Control over sampled data points

Hi,

I'm running into a problem using the Matlab RANSAC (is it actually MLESAC as cited on the documentation page?) implementation.

For the data that I'm working with, my model is uniquely determined for 2 data points (sampleSize = 2) such that the model parameters are estimated via matrix inversion as (or using Matlab's ). Where and . This looks like:

The problem is that the data I'm working with is the output of a sensor that bins the values into a set of 64 angular bins. And so for some samples, I get and thus A is singular and the matrix inversion cannot be done.

However, this does not indicate that either data point 1 or data point 2 are outliers, just that these two points are not a good sample.

Is there any way in which I can reject a sample generated by RANSAC based on user-determined criteria (in this case if )?

Thank you,

Carl

Answer 1

[Star Strider]

Try using the lsqr (link) function. It’s intended for sparse matrices, however it gave a finite result when I tested it with your model and . Another option is the pinv (link) function. Both gave the same result.