Mathworks motion based multiple object tracking

https://www.mathworks.com/help/vision/ug/motion-based-multiple-object-tracking.html

Lower down they cover information on how to determine whether new motion for a blob's centroid gets corresponded with which track.

track usually means "tracked record" like track and Object seem to be synonymous. Calculation works by calculating euclidean distance from previous track centroid to the new position for each thing of motion detected in the scene. also includes the confidence of the prediction

Becomes an MxN matrix with N detections and M tracks

Second step is the assignment. This is an experimentally tuned section. The value getting set too low results in fragmentation and the creation of new tracks for the same object's motion. In the reverse, if the value deciding the assignment is set too high then the same track may pick up separate detections not the same one the whole time through.

? Are there other methods for doing this type of track ? This seems pretty standard, but perhaps like just randomly deciding on the value set for assignment might cause issues.

parallel article question for how to do this in opencv https://stackoverflow.com/questions/26363257/tracking-multiple-moving-objects-with-kalmanfilter-in-opencv-c-how-to-assign

So it sounds like there are some options here for determining whether an object is new or the same one moved to new location.

euclidean distance of centroids comparison
bounding box overlap
hungarian algorithm

Bounding Box overlap can be thought of this way

DevinBayly / thermal_imaging

Mathworks motion based multiple object tracking #7