Markov-chain Monte-Carlo approach for association probability evaluation

Data association is one of the essential parts of a multiple-target-tracking system. The paper introduces a report-track association-evaluation technique based on the well known Markov-chain Monte-Carlo (MCMC) method, which estimates the statistics of a random variable by way of efficiently sampling the data space. An important feature of this new association-evaluation algorithm is that it can approximate the marginal association probability with scalable accuracy as a function of computational resource available. The algorithm is tested within the framework of a joint probabilistic data association (JPDA). The result is compared with JPDA tracking with Fitzgerald's simple JPDA data-association algorithm. As expected, the performance of the new MCMC-based algorithm is superior to that of the old algorithm. In general, the new approach can also be applied to other tracking algorithms as well as other fields where association of evidence is involved.