Data association in multi-target tracking: a solution using a layered Boltzmann machine

Data association is the problem of determining the origin of measurements in a multitarget tracking algorithm, and assigning probabilities β_i^t to the event that the i -th measurement originated from the t-th target. A parallel computational method is presented for solving the data association problem using a layered Boltzmann machine. The association probabilities can be computed with arbitrarily small errors if a sufficient number of layers of binary neurons are available. Specifically, the probability β_i^j is shown to be equal to the relative frequency of activation of neuron v( i,j) in a layered two-dimensional network. The authors present some simple tracking examples comparing the performance of the Boltzmann algorithm to the exact data association solution, and with the performance of an alternative parallel method using the Hopfield neural network