Evolutionary algorithm for data association and IMM-based target tracking in IR image sequences

Simultaneous tracking of multiple maneuvering and non-maneuvering targets in the presence of dense clutter and in the absence of any a priori information about target dynamics is a challenging problem. A successful solution to this problem is to assign an observation to track for state update known as data association. In this paper, we have investigated tracking algorithms based on interacting multiple model to track an arbitrary trajectory in the presence of dense clutter. The novelty of the proposed tracking algorithms is the use of genetic algorithm for data association, i.e., observation to track fusion. For data association, we examined two novel approaches: (i) first approach was based on nearest neighbor approach and (ii) second approach used all observations to update target state by calculating the assignment weights for each validated observation and for a given target. Munkres' optimal data association, most widely used algorithm, is based on nearest neighbor approach. First approach provides an alternative to Munkres' optimal data association method with much reduced computational complexity while second one overcomes the uncertainty about an observation's source. Extensive simulation results demonstrate the effectiveness of the proposed approaches for real-time tracking in infrared image sequences.