Association, Fusion and Adaptation for Multisensor Target Tracking Systems

Advances in sensor technology constantly bring new challenges to the fields of estimation and target tracking. Development of new types of sensors may necessitate new models for the measurements they provide. Common assumptions that made suitable performance of previously developed tracking systems possible may no longer hold. Utilization of multiple sensors in both heterogeneous and homogeneous cases leads to new challenges for the association and fusion of multiple measurements and targets. Additionally, it is also necessary and desirable to ensure that the systems designed are able to extract the maximum possible information from the sensors which are utilized.;This work aims to address the challenges of sensors with poorly understood measurement noise characteristics, track association with indirectly related measurements, and the evaluation of the statistical efficiency of sensor measurement fusion. Firstly, the problem of poorly understood measurement noise characteristics will be examined through the design for consistency of a passive collision warning system for unmanned aerial vehicles. In order to provide a collision warning, the system will perform a statistical test on the targets tracked in the vehicle's surrounding airspace. Due to the nature of this test, consistency of the estimator is of the utmost importance. To this end, the measurement noise characteristics, which were found to vary over time, must be adapted to. Application of a recently developed method of measurement noise variance estimation allows for the design of an adaptive target tracker which will exhibit the necessary consistency.;Secondly, the challenge of track to track association (T2TA) with nonlinearly related feature measurements will be presented. A T2TA scheme is developed, which will take advantage of traditional kinematic state information as well as additional state information in the form of state augmentation. The main contribution is the use of two nonlinearly related state augmentations at the two sensors, as well as accounting for their uncertainties. The results of T2TA when using the full augmented state are compared to the results of T2TA with either kinematic or state augmentation information alone. The full augmented state is shown to provide the best association results, both in terms of accuracy and the number of samples needed to provide that accuracy.;Finally, the statistical efficiency is examined for two cases of line-of-sight (LOS) measurement fusion. In the first, LOS measurements from passive sensors, assumed to be synchronized, are combined into a single composite Cartesian measurement (full position in 3D) via maximum likelihood (ML) estimation. The use of composite measurements can circumvent the need for nonlinear filtering — which involves, by necessity, approximations. This ML estimate is shown to be statistically efficient, even for small sample sizes (as few as one LOS measurement from each of two sensors), and as such, the covariance matrix obtainable from the Cramer-Rao lower bound provides the correct measurement noise covariance matrix for use in a target tracking filter. In the second case, an acoustic target and sensor localization system with position dependent measurement noise is examined. The system itself has been previously examined, but without deriving the CRLB and showing the statistical efficiency of the estimator used. Two different versions of the CRLB are derived, one in which direction of arrival (DOA) and range measurements are available ("full-position CRLB"), and one in which only DOA measurements are available ("bearing-only CRLB"). In both cases, the estimator is found to be statistically efficient; however, depending on the sensor-target geometry, the range measurements may or may not significantly contribute to the accuracy of target localization.