A factor graph approach for efficient JPDA implementation in multi-target tracking

This paper develops a factor graph (FG) based efficient implementation scheme for joint probabilistic data association (JPDA). The association hypotheses probabilities in JPDA are computed using the sum-product algorithm in a factor graph framework. The multi-target tracking (MTT) data association constraint of not having more than one track assigned to a measurement is incorporated in the proposed approach by modifying the sum-product algorithm. Using the modified sum-product algorithm, this paper shows in detail the message passing in a tree structured factor graph. Compared to other fast JPDA implementation techniques, such as suboptimal JPDA and near optimal JPDA, the proposed method obtains the exact JPDA association probabilities in data association scenarios having tree structured graphs with significant reduction in computational cost. The message passing scheme developed for tree structured graphs are used for scenarios consisting of graphs with loops, and the advantage is shown in this paper using simulations. The computational advantage of the proposed FG based approach is also analyzed for the implementation of the recently proposed Iter-JPDA algorithm for avoiding track coalescence when targets move close. Monte Carlo simulations, comparing the root mean square (RMS) positional error and the computational reductions obtained by the proposed approach, standard JPDA and Iter-JPDA, are presented.