Research On Sequence-sensitive Multi-sensor GM-PHD Tracking Algorithm

The problem of multi-sensor multi-target tracking has always been one of the key concerns of scholars in the field of information fusion.For some more complex environments,the complementary and redundant characteristics of data are used to achieve data fusion according to certain strategies to improve the accuracy of target tracking.However,when data are fused,there exists some problems: 1)the sequence of fusion may affect the result,and there may be data loss during the fusion process;2)although the centralized fusion structure can greatly ensure the data integrity,the communication and computation burden of the fusion center is large;3)the computation of the filter is too large under dense clutter and the real-time performance cannot be guaranteed.To this end,this paper studies the multi-sensor Gaussian Mixture Probability Hypothesis Density(GM-PHD)tracking algorithm which is sensitive to fusion order,the main work of this paper is listed as follows.1)Aiming at the problem that the sequential fusion method is sensitive to the fusion order,a multi-sensor adaptive observation iteratively updating GM-PHD tracking algorithm is proposed based on the centralized sequential fusion method and Optimal Subpattern Assignment(OSPA)method.Firstly,a multi-sensor observation iteratively updating framework is built using GM-PHD filtering.Secondly,two methods are proposed to optimize the fusion order based on the OSPA metric,one is the fusion consistency metric based on the measurement information and the other is the fusion consistency metric based on the posterior Gaussian mixture set.Finally,the fusion consistency metric of each sensor is calculated and ranked from largest to smallest by the optimization method,and the adaptive sequential fusion is realized by combining the multi-sensor measurement iterative update fusion framework.The simulation shows that the two optimization algorithms can effectively optimize the fusion order and improve the estimation accuracy of the fusion result.2)Aiming at the problem of centralized communication and large computational burden,a distributed multi-sensor measurement iterative updating GM-PHD tracking algorithm based on the weighted pseudo-measurement is proposed to reduce the task load of fusion center and ensure the data integrity as much as possible.Firstly,each sensor is filtered locally to obtain the posterior Gaussian mixture set.Secondly,the target state position and weight information are extracted from the posterior Gaussian mixture set and sent to the fusion center as pseudo-measurements,which are more accurate than the original measurements and the valid data are kept more complete.Finally,the pseudo-measurements are used to iteratively update the priori Gaussian set to improve the accuracy of estimation results.The simulation results show that this algorithm has higher tracking accuracy than the multi-sensor measurement iterative update GM-PHD tracking algorithm.3)A distributed adaptive multi-sensor sequential fusion GM-PHD tracking algorithm is proposed based on the cumulative amplitude likelihood ratio for the multisensor sequential fusion order optimization problem under dense clutter.Firstly,to mitigate the interference of dense clutter,the amplitude is introduced into the GM-PHD filter,and the effective measurement set is filtered by the amplitude feature.Secondly,a fusion order optimization algorithm with cumulative likelihood ratio of amplitudes is proposed.Finally,the fusion is realized by combining the distributed sequential fusion GM-PHD tracking algorithm.Simulation experiments show that for the single-sensor GM-PHD tracking algorithm,adding amplitude is better than the tracking without adding amplitude;for the order optimization problem of multi-sensor GM-PHD tracking algorithm,the fusion order optimization algorithm based on the cumulative likelihood ratio of amplitude can effectively optimize the sequential fusion order to a certain extent.