Font Size: a A A

Research On Algorithms Of Extended Target Tracking Based On Random Finite Set

Posted on:2015-02-24Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y Q ZhangFull Text:PDF
GTID:1108330482453168Subject:Intelligent information processing
Abstract/Summary:PDF Full Text Request
With the increasing improvement of radar, infrared, etc. sensor resolution, the technique of extended target tracking(ETT) plays a more and more important role in the military fields like the antimissile, warning system, guidance and battlefield surveillance. For the ETT problems under complex environments, due to increase in the amount of information obtained by sensors(including the target observation, size, shape and orientation, etc.), we require adding the extension state estimate on the basis of the kinematical state estimate to improve the traditional point target filter model. However, since the model and computational complexity of the traditional point target tracking approaches considering only target kinematical state and based on data associated does not match and is too high, which is difficult to take full advantage of high-resolution sensors. The dissertation utilizes random finite set(Random finite set, RFS) and random matrix theory to carry out the research on the ETT methods, and focuses on solving the key problems of the measurement set partitioning, mixture reduction and target shape modeling. The main contributions of the dissertation are as follows:1. In an extended target Gaussian mixture PHD(ET-GM-PHD) filter, the exact filter requires all possible partitions of the current measurement set for updating, which is computationally intractable. For this reason, a fast Fuzzy ART partition method is proposed, which substitutes Distance partition with a fuzzy ART model. In Fuzzy ART partition, the subset of partitions of the current measurement set is generated by the different vigilance parameters. However, when the values of vigilance parameters are higher, the ET-GM-PHD filter using Fuzzy ART partition occurs the cardinality overestimation problem. Additionally, when the targets are spatially close, this filter appears the cardinality underestimation problem. To reasonably handle the cardinality overestimation and underestimation problems, cardinality overestimation mechanism and sub-partition method are proposed, respectively. The simulation results show that Fuzzy ART partition method can well handle the close-spaced extended targets and obviously reduce computational burden without losing tracking performance.2. Compared to the ET-GM-PHD filter, the extended target Gaussian inverse Wishart PHD(ET-GIW-PHD) filter considers the kinematical state and extension state of the target. Similar to the ET-PHD and ET-GM-PHD filters, this filter also exists the partitioning problem. However, for the separating tracks, when the two or more different sized extended targets are spatially close, this filter occurs the cardinality underestimation problem. For this reason, combining the fuzzy ART with Bayesian theorem, a novel robust Bayesian partition method is proposed. In Bayesian partition, with the increase of the measurement input, the true distribution’ shape of the category is iteratively depicted by updating. Since it takes into account the extended targets’ extensions, Beyesian partition has the potential advantage for the different sized and spatially close extended targets. Simulation results show that the proposed partitioning method has better tracking performance than the existing partitioning methods, implying good application prospects.3. In the ET-GM-PHD filter, the number of Gaussian mixture(GM) components N grows exponentially with time. To keep N at a computationally tractable level, GM reduction becomes necessary. For this purpose, a GM reduction based on the fuzzy ART(GMR-FART) is proposed. The architecture of GMR-FART is similar to that of the fuzzy ART, however, its choice function, match function and learning update equations are characterized by features of GM. Its performance is evaluated by the normalized integrated squared distance(NISD) measure. The results show that the reduced mixture formed by the proposed algorithm can well approximate the original mixture and it requires less computational burden.4. To simultaneously estimate the kinematic and extension states of the targets, Granstr?m et al. in the ET-GIW-PHD filter adopted GIW mixture approximating the posterior density of the target state. Similar to GM, GIW mixture also requires the mixture reduction. For this reason, a weighted Kullback-Leibler(KL) difference is proposed. This difference is derived by considering the weights of GIW components, which is more reasonable and effective than the existing KL difference. Additionally, an evaluation criterion is developed, called a global difference measure. It is a deviation between the original and reduced GIW mixture, which is obtained by solving the NISD. Finally, both the proposed evaluation criterion and algorithm are tested on simulation examples, and the results show that the proposed evaluation criterion can depict correctly the result of the curve analysis.
Keywords/Search Tags:Random finite set(RFS), Extended target tracking(ETT), Gaussian mixture PHD(GM-PHD) filter, Gaussian inverse Wishart PHD(GIW-PHD) filter, Gaussian mixture reduction, Gaussian inverse Wishart mixture reduction, Fuzzy ART partition, Bayesian partition
PDF Full Text Request
Related items