Open ulassbin opened 2 years ago
It is clearly visible from image above, only the 0'th element started from a position different than 0. This means initial message from the sensor hold only 1 object, the other ones were empty but I started 9 of them in 0,0.
Randomly initializing 10 filters issue is handled here: 7f5a8c0441cae87e128fcf899f1de34d3e5f0468
Also, some issues related with gating are fixed as well.
There still seems to be an issue about covariance calculations, covariance value at one direction does not grow as quick as the other one. Need to check matrix dimensions. Sometimes, tracks do crazy stuff, like moving towards a random target. Visualizing connections can be helpful to track that issue.
Crazy stuff like random moving target is caused by gating several obstacles far away. Now the node visualizes correspondences as wel, here is a screenshot of a very large correspondence acceptence. The lines show correspondence vectors.
Still covariance increases in prediction step seems messed up.
Currently the algorithm is facing several issues with filter handling. Those are: