Research On Multiple Extended Targets Tracking With Shape Estimation Based On Random Finite Set

Because of the improvement of the radar and sensors resolution with the progress of science and technology constantly, target may generate more than one measurement in a moment, and an extended target is available. If still using the traditional method to associated the measurement and the target has been unable to meet the requirements. In recent years, the multiple target tracking method based on random finite set attracts the more attention of scholars due to its effective compared with the traditional algorithm. This thesis mainly focuses on the extended target tracking algorithm with shape estimation based on random finite set. The main research contents are as follows:Firstly, Aiming at the problem that the K-means clustering method is excessively dependent on the initial center point and the low efficiency, a new method based on mean shift clustering is proposed. This method uses gaussian function as the kernel function, uses the mean shift method to obtain the attractive basin, then merges the attractive basin, and removes the nosie. This method does not need to give the number of the specific number of clusters and the center point, and it is effective.Secondly, Aiming at the problem that the distance division method cannot accurately divide the closed targets, a new improvement method based on distance partition is proposed. This method uses another division based on the distance division, then uses maximum likelihood estimation to estimate the number of targets in each partition set, and then partitions the set of target number greater than 1, so that the measurement set of each extended target is separated. This method can effectively deal with the situation that the targets are closed or crossing.Finally, Aiming at the problem that the existing extended target modeling as an ellipse that cannot solve the star shape, a method is proposed based on star convex random hypersurface model, which can adaptively estimate the extended target shape using Gamma Gaussian mixture Cardinalized probability hypothesis density(SRHM-GGM-CPHD) algorithm. The shape of extended target can be modeled as a Star convex Random Hypersurface Model(SRHM) through this algorithm. Then it is embedded into the Gamma Gaussian mixture CPHD filter framework. Thus the tracking of multiple extended targets is completed. The proposed algorithm is verified in the centroid position and the performance of the extended shape is superior to the traditional gamma Gaussian inverse Wishart CPHD filter based on random matrix.