Aggregate surveillance: A cardinality tracking approach

This paper introduces cardinality tracking, a special case of the more general multi-target tracking problem for which measurements do not provide any target state information. That is, each scan only provides information as to how many targets are present. We address the problem with a modified form of the multiple-hypothesis tracking formalism using equivalence classes. Structural results exist which enable optimal track extraction to be achieved. We introduce as well some variations, approximate approaches that introduce further hypothesis aggregation. We show that we are able to improve significantly over a straightforward MHT approach to the problem. Similar results can be obtained by considering the problem as one of Kalman filtering over the aggregation of targets.