It is more difficult than you may be assuming. How do you know the hits are false? These "hits" are collections of samples at points in time, not continuous tracks. The "tracks" are reconstructed by making inferences from the samples.

Determining whether any pair of sequential samples represents the same entity or two unrelated entities is an extremely difficult inference problem with no closed or general solution. If there is too much clutter, it becomes almost unresolvable. Aliasing will create a lot of false tracks.

History has shown that any heuristic you use to filter the clutter will be used by your adversary as an objective function to hide from your sensors once they know you are using it (e.g. doppler radar "notching").

For this reason the inference algorithms are classified but they will degrade rapidly with sufficient clutter no matter how clever. It is a limitation of the underlying mathematics.