Research On Tracking Before Detection Algorithms Of Dim Targets Based On Random Finite Set

With the increasing improvement of scouting aeroplane, stealth technique, etc. the technique of the infrared dim small target detecting and tracking plays a more and more important role in the modern military fields, such as the infrared surveillance, antimissile, defend and satellite remote sensing system, etc. For the dim small target detecting and tracking problems under the complex environment, due to the randomness of target number, and the influence of complex background, such as noise, clutter, etc., lead to SNR of targets is low so that it becomes difficult to detect and track the dim small target. The dissertation utilizes the RFS theory to carry out the research on the dim small target, and focuses on the key problems of the tracking before detection in the infrared dim small targets, unknown new target information under the dim small target detection and tracking, Gaussian mixture reduction, etc. The main contributions of the dissertation are as follows:1. For the problems of low accuracy in tracking and high computational complexity suffered by the current dim target TBD algorithms based PHD in the infrared imagery, a GPF-PHD based TBD algorithm is proposed in this dissertation, which is obtained via the recursive operation of the Gaussian components in PHD. The GPF-PHD-TBD can not only make the nonlinear estimation ability of the particle filter best, but also avoid the inaccuracy in tracking and high complexity. The simulation results show that the proposed algorithm can guarantee the performance of the dim target tracking, and effectively reduce the running time, implying a good prospect of engineering application.2. An improved Gaussian mixture multi-target multi-Bernoulli TBD algorithm is proposed, which considers the prior information of the unknown new targets and low SNR by a search mechanism using multi-Bernoulli components to search the newborn targets in the field of view. The purpose of doing so is that the accurate modeling of the target new process can be avoided. Moreover, the algorithm can also avoid the problem, such as resampling, clustering, etc., thus the performance of the filter can be improved and the complexity of that can be reduced. The simulation results show that the improved algorithm can achieve similar tracking performance with that of the idealcondition in the absence of any prior information, and can improve the tracking performance and reduce the complexity effectively.3. To reduce the number of mixture components in a GM PHD filter, a novel weighted KL divergence based hypotheses merging criterion is proposed. The GM-PHD is a multi-target tracking algorithm that reduces the growing number of mixture components through pruning and merging. However, when encountered with the challenging of tracking scenarios where closely spaced targets move side by side in parallel or approaching targets cross over their trajectories, existing merging criterion in GM-PHD often fails to resolve the ambiguity, causing the abrupt performance degradation. The implementation of the KL-based merging criterion is split into two separate parts: First, pruning and merging routine is formulated to reduce computation complexity; second, pruning and merging routine is to facilitate efficiently state extraction. Through extensive simulation, it is observed that the proposed KL divergence criterion not only significantly improves the accuracy of target tracking, but also reduces the computation cost.