fix: Fitting with ouliers

[Corentin-Allaire]

Fix an issue with the fitting when there are outliers.

Since the outliers use to not have filtered parameters and covariance, when going through the different states for the smoothing their covariance matrix was taken to be equal to 0 resulting in a bad smoothing. Instead we now put the filtered covariance for the outliers equal to the predicted one (like the material surface).

The material flag was also always on for the measurement state now we check that material is present to turn that flag on.

[codecov]

Codecov Report

Merging #804 (b11e085) into main (7b9f78c) will increase coverage by 0.00%.
The diff coverage is 33.33%.

@@           Coverage Diff           @@
##             main     #804   +/-   ##
=======================================
  Coverage   48.64%   48.64%           
=======================================
  Files         328      328           
  Lines       16898    16900    +2     
  Branches     7934     7935    +1     
=======================================
+ Hits         8220     8221    +1     
  Misses       3080     3080           
- Partials     5598     5599    +1     

Impacted Files	Coverage Δ
Core/include/Acts/TrackFitting/KalmanFitter.hpp	39.60% <33.33%> (+0.05%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7b9f78c...b11e085. Read the comment docs.

[asalzburger]

Makes totally sense, I will have a look if this is an issue with the CKF as well ...

[asalzburger]

Just to add: the CKF seems to have this already available:

      if (isOutlier) {
        ACTS_VERBOSE("Creating outlier track state with tip = " << currentTip);
        // Set the outlier flag
        typeFlags.set(TrackStateFlag::OutlierFlag);
        // Increment number of outliers
        tipState.nOutliers++;
        // No Kalman update for outlier
        // Set the filtered parameter index to be the same with predicted
        // parameter
        trackStateProxy.data().ifiltered = trackStateProxy.data().ipredicted;