Target Tracking in Complex Scenarios

Doktorsavhandling, 2010

This thesis is concerned with three important components in target track- ing, namely multiple-model filtering, data association and sensor resolution modeling. For multiple-model filtering, the preferred method has long been the Interacting Multiple Model (IMM) filter, which relies on the assumption that immediate model shifts have the highest probability. In this thesis, an alternative switching model is proposed, which forces the models to persist for at least a model-specific time, yielding a less complex problem in terms of model hypotheses. Further, a state estimation algorithm is derived, which is close to optimal under the model assumption. The proposed filter, called the Switch-Time Conditioned IMM (STC-IMM) filter, is shown to provide better performance than the IMM filter in benchmark scenarios. Traditional tracking algorithms are designed to estimate the states of the targets, while trying to maintain their identities. In this thesis, it is shown how these algorithms can be adjusted to problems where target identity is not relevant. More specifically, the Joint Probabilistic Data Association (JPDA) filter is considered, and two adjustments of it are presented, called the Set JPDA (SJPDA) and the Kullback-Leibler Set JPDA (KLSJPDA) filters. These filters both enable more accurate Gaussian approximations, and provide more accurate state estimates than the JPDA filter when evaluated with a metric that disregards identity. Another approach to the problem is to use Finite Set Statistics (FISST). In the thesis, the results of the first performance comparison of the most prominent FISST-based and traditional filters are presented and discussed. In the development of most tracking algorithms, it is assumed that the targets are always resolved by the sensor. However, when the targets are closely spaced in relation to the sensor resolution, this assumption is not valid, and may lead to decreased tracking performance. This thesis presents a multi-target sensor resolution model, for an arbitrary but known number of targets, which takes resolution effects into account. It is further shown how the model is incorporated into a Bayesian tracking framework, and two alternative JPDA-like filters are presented.