Bayesian track initiation by time-reversion of trajectory models
R.A. Hogendoorn, H.A.P. Blom · 2005
The theory of Bayesian track initiation for multiple objects without a priori identification is well developed. To end up with a practically acceptable combinatorial complexity this theory requires the use of appropriate hypotheses reduction techniques. Unfortunately in cases of high false measurement densities the required hypothesis reduction is such drastic that initially weak correct tracks are deleted before they can grow strong while initially strong false tracks survive too long. The actual result is that probabilistic track initiation is too much occupied with past measurements and too little with more recent measurements. To remedy this situation, we exploit the theory of time-reversion of a Gauss-Markov process to the track initiation problem. In combination with drastic hypotheses reduction it yields a probabilistic track initiation which is mainly occupied with the most recent measurements. The effectiveness of this reverse-time approach has been verified by experiments.