Multitarget Multisensor Trackingin the Presence of Wakes

TARGET tracking is an essential requirement for surveillance and control systems to interpret the environment. This environment may contain multiple targets, and the environmental information may be obtained by multiple sensors in a multitarget multisensor tracking system. In this thesis we focus on targets which, in addition to reflecting signals themselves, also have a trailing path behind them, called a wake. This wake causes additional measurements to those originating from the target. When the measurements are processed, the estimated track can be misled and sometimes lose the real target because of the wake. This problem becomes even more severe in multitarget environments where targets are operating close to each other in the presence of wakes. In this thesis a probabilistic model is developed which reflects the probability that a false measurement originates from the wake behind a target. This wake model is integrated in the probabilistic data association filter (PDAF) to improve the track continuity for tracking single targets. The modified PDAF is further extended to handle multiple targets in the presence of wakes by using a probabilistic wake model for each of the targets in the multitarget environment that has a wake behind it. These single wake models are combined to form a joint wake model which augments the joint probabilistic data association filter (JPDAF) for both coupled and decoupled filtering. The wake-originated measurements may also cause confusion in the track initiation. To prevent this problem, a clustering method is proposed based on morphological operators which allows tracks to be initialized based on two-point differencing of the cluster centroids from succeeding scans. The modified PDAF is tested on data of a real scuba diver with an open breathing system. In this case the air bubbles produced by the diver form a wake which extends far behind the diver. The experiment showed that the above modifications of the PDAF improved the track continuity significantly. Finally, a relatively extensive simulation, based on real scuba diver data, is presented. Four different multitarget multisensor tracking scenarios are simulated, considering two targets with wakes that are: 1. Crossing each other. 2. Moving in parallel to each other. 3. One following after another. 4. Meeting and then passing each other. The results of these simulation scenarios show that the presented modifications improve the tracking performance, and the probability of lost tracks is significantly reduced. The targets are observed by two sensors, and it is shown that tracks estimated in a centralized fusion configuration are better than the local tracks estimated using data from individual sensors only. It is also shown that applying the wake model to targets that do not generate a wake, yields almost no deterioration of the tracking performance