Hi Joost,
We do follow the approach that Blackman and Popoli suggest.
Essentially, each detection carries its own SensorIndex property with it. By default it is 1, for the single-sensor case. For multiple-sensors, if you define each detection with a different sensor index value, you let the tracker know that they have a different origin.
In the tracker, each set of detections from a different sensor is treated separately to create all the track branches based on their cost relative to the assignment thresholds. The trackerTOMHT still assumes the point target assumption, in other words at most one detection per object in a sesnor scan. However, the tracker does allow branches with multple sensor detections as long as they don't violate the one per sensor.
I hope this helps answer the question
Elad
