Research On Optimal Tracking Algorithm For Distributed Multi-Sensor Based On Multi-Bernoulli Filter

The main task of target tracking is to filter out the real target's state from the noisy observation data and output it.In the multi-target tracking task based on distributed sensor network,there are such things as new targets,changes in the number of targets,and nonlinearity and non-Gaussian features in target motion,the large and unknown observation errors of different sensors,and the limited field of view of the sensor,these problems make it very challenging to complete a stable multi-target tracking task.The multi-Bernoulli family filter based on the random finite set,such as cadinality-balanced multi-Bernoulli filter,generalized labeled multi-Bernoulli filter and labeled multi-Bernoulli filter,due to its direct transfer of the posterior probablity density of multi-target recursively,have certain advantages.After the label space is introduced to multi-Bernoulli family filter,the output of multi-target tracking can contain the unique label which can be used to distinguish different target,which makes the filter have the function of track management.However,in the complex environment,the traditional algorithm still has some defects,such as abnormal tracking,missing tracking and high time complexity.In the framework of multi-Bernoulli family filter,the distributed multi-sensor multi-target tracking algorithm is studied in this paper:1.In order to solve the problem that the decrease of fusion result and bad influences to the subsequent filtering process based on the GCI technique of MB filter due to the complex tracking environment and different sensing ability and location causing the huge difference between local sensors' filtering accuracy.This paper proposed an improved algorithm of distributed multi-sensor multi-target tracking based on multi-Bernoulli(MB)filter,including(1)A decision-level fusion strategy is proposed to improve the final output results which avoids the inaccurate estimated states caused by directly fusing the output results from different sensors in the complex scenarios.(2)A feature-level fusion feedback strategy is proposed to solve the problem caused by the direct feedback of the inaccurate fusion results and can reduce the false alarm.(3)An interactive feedback strategy is proposed to avoid the missing tracking of each single sensor.Through the fusion and feedback in different stages,the tracking accuracy of multi-target in complex environment is improved.Finally,the correctness and effectiveness of the proposed algorithm are verified by the comparative experiments.2.The traditional distributed fusion algorithm based on LMB filter with track management function has the problem of low real-time tracking performance due to high computational complexity.According to the characteristics of label filter,this paper introduces new variables for LMB parameter set to record the key fusion data,and improves the matching step in the process of distributed fusion,which can reduce the computational complexity and avoid the additional time consumption caused by the inconsistency of label space.The experimental results show that the proposed method has better tracking performance than the traditional algorithm in linear and nonlinear tracking scenes,and the running time of the proposed method is significantly lower than that of the traditional algorithm,which proves that it has better realtime performance and is more suitable for real tracking tasks.3.Aiming at the problem that the traditional distributed fusion tracking algorithm cannot effectively track the target under the condition of the limited field of view of each sensor,this paper proposes a robust distributed fusion algorithm based on the LMB filter.The extraction set division strategy solves the problem of target distribution loss caused by the traditional distributed fusion algorithm tracking under the limited field of view;the track maintenance strategy ensures that the target can maintain its label information before entering the sensor field of view and after leaving the field of view.Experimental results show that the proposed algorithm can complete effective multi-target tracking tasks,and has better robustness than traditional algorithms.