Research Of Measurement Partition Algorithm For Multiple Extended Targets Tracking

In the classical target tracking cases, it is assumed that one target generates at most one measurement per time step. However, with the increase of the resolution of modern radars and other dection equipments, the echo signal of a target may be distributed in a different ra nge resolution cell, i.e., a single target may produce multiple measurements, such target is referred to as an extended target. In the conventional extended targets tracking, especially in clutter environment with unknown and varying number of targets, measurement partition is a matter of prime problem to firstly resolve. Whether measurement partition is correctly divided directly affect the subsequent targets state estimation. Based on probability hypothesis density, the dissertation mainly investigates measurement partition of multiple extended targets.1. For the problem of tracking an unknown and varying number of extended targets in clutter, in order to limit the number of partitions and computional cost, a novel partition method is proposed based on de nsity and spectral clustering technique. Different with traditional partition algorithm of multiple extended tracking, the proposed method does not directly partition measurement set, but firstly separates clutter measurements from target measurements by kernel density function, then partition targe tmeasurements. As to the high cost of time due to a large number of partitions for the general partition method, a novel measurement partition algorithm based on improved spectral clustering technique is introduced in this paper, and established a mathematical mode for the number of classes in extended target measurement set, which can greatly reduce the number of partitions and computational burden without losing tracking performance. The experiment results demonstrate the effectiveness of the algorithm.2. Measurements originated from multiple extended targets are interpreted just from one target resulting to underestimation of target s in situations where two or more extended targets are spatially close with the clutter condition. A novel partition method with unknown rate is presented. The algorithm selects mean shift cluster, and builds window width mode so that avoiding the defect of determining the number of classes in advance. In addition, the method refers to sub-partition, which is the process to cluster for error measurement cell. However, measurement rate determine whether a cell should be split or not to create an additional partition. The measurement rate were assumed to be known a priori, in this sect ion some scenarios where measurement rate is unknown are investigated. A new real-time estimate of unknown rate method is introduced, which make estimated rate convergence to actual rate. The results show that improvements of both partition accuracy and performance of filtering algorithms.3. We introduce a new measurement partition framework based on de nsity and cluster. In this paper, all the proposed algorithms is not limited to the algorithm, under the framework we change the sub-method, which is still valid. The effectiveness of the framework is verified by the simulations of the novel partition method based on AP and FCM clustering.