MULTI-TARGET TRACKING ALGORITHMS FOR CLUTTERED ENVIRONMENTS

Commercial and military surveillance systems must provide continuous and reliable position information for all targets in the surveillance space in the presence of false returns. These false detections may be produced by unwanted terrain clutter returns and noise sources. To accomplish this task successfully, detection and continuous track of all targets in the surveillance space is required. This dissertation presents two new algorithms that allow the tracking of multiple maneuvering and non-maneuvering targets in cluttered environments.; The first algorithm describes a general approach for initiating, maintaining, and terminating maneuvering and non-maneuvering targets in cluttered environments. This method accounts for missing target detections and resolves conflicts such as crossing target trajectories. In addition, the algorithm eliminates false target trajectories using an m out of n detector, and redundant target tracks are reduced by combining tracks with similar state estimates. To accomplish this however, this method splits the target trajectory when more than one measurement is found in the track validation region and requires multiple scans to confirm true target trajectories and eliminate unlikely tracks. Three methods are proposed for target state estimation and prediction. The algorithm can be implemented in a real time computer system.; The second tracking approach does not require multiple scans, but instead uses all the available measurement data at each sampling time to update the target track. This method is based on a probabilistic data association filter which is exponentially weighted. The scheme exhibits the ability to track multiple targets which are rapidly maneuvering in cluttered environments. The algorithm has fixed memory and computational requirements and can be implemented in real time tracking systems. Additional processing is required for track initiation.; Simulation results for both algorithms are presented. The results demonstrate good tracking performance for various maneuvering and non-maneuvering target scenarios in cluttered and clear environments. In addition, the exponentially weighted probabilistic data association (EPDA) method is compared to the standard probabilistic data association (PDA) algorithm. The EPDA was shown to have superior tracking performance over the standard PDA method in various maneuvering target scenarios.