Open 9a3eedi opened 8 years ago
It seems that the template tracking system from VISP cannot give probabilities, so we will need to add some filter to it. We're thinking of implementing a Kalman Filter + some kind of constant motion model.
I added a confidence level value in the TrackerData message. The TrackLandingMark class now publishes a confidence level, but it is done in a very simple manner. The way it works is that if the class is tracking, the confidence level is 100% (1.0f). If it's not tracking, then the data published is the same as previous data and the confidence level is halved every step, which means it becomes 50%, then 25%, then 12.5%, etc.
This is OK for now but we need a better way. Next step is to add a simple motion filter model
Currently the tracker is designed to only publish bounding box coordinates if it is tracking something. If it cannot track, it stops publishing. This is too simple and we need to also provide a confidence level to the tracker data.
This might mean we need to change the type of tracker we're using.