Re: explain in Bayesian Filter

Posted by matlabbe on Oct 17, 2015; 4:32pm
URL: http://official-rtab-map-forum.206.s1.nabble.com/explain-in-Bayesian-Filter-tp779p780.html

Hi,

I will refer you to the section III-C of this paper: "Appearance-Based Loop Closure Detection for Online Large-Scale and Long-Term Operation".

Some values come from this paper: "Fast and incremental method for loop-closure detection using bags of visual words".

FAB-MAP uses 0.9 too for the new place prior:
"the prior that a new topological link leads to a new place was 0.9"

For the PredictionLC, this is a gaussian of sigma=1.6 (as described in the first paper above). The choice of 16 is dependent on the neighborhood used to retrieve the locations from the Long-Term Memory. 16 was satisfying for every datasets tested: enough large to retrieve more future nodes in case the environment is dynamically changed, and not too large to avoid filling the WM with only neighbor localizations.

cheers