Research On Maximum Likelihood Recursive Updating LMB Multi-sensor Fusion Tracking Method

In the field of multi-target tracking,methods based on Random Finite Set Theory have been continuously receiving attention and flourishing development.Typical finite set trackers include: Probability Hypothesis Density(PHD),Cardinalized Probability Hypothesis Density(CPHD),Multi-Bernoulli(Me MBer),Cardinalized Balanced Multi-Bernoulli(CBMe MBer),and Labeled Multi-Bernoulli(LMB).Among them,the LMB tracker is a relatively new research result,with high tracking accuracy and strong robustness.Nevertheless,LMB needs to consider a large number of combinations of labeled bernoulli items and measurements,resulting in a large computational burden.Therefore,improving the real-time performance of traditional LMB in complex tracking environments is a challenge.Furthermore,due to the high computational complexity of single-sensor LMB,the fusion tracking design problem of multi-sensor LMB has always been a research difficulty,especially how to achieve global consistency estimation of multiple targets’ states and identities(labels)under controllable computational conditions.For these,this paper will conduct a maximum likelihood recursive update LMB multi-sensor fusion tracking method research on the above difficulties,focusing on solving the following two problems:(1)How to achieve fast,robust and accurate tracking of multiple targets based on single sensor information;(2)How to achieve global consistency estimation of multiple targets’ states and identities under multi-sensor networking.The main tasks of this paper are as follows:(1)In order to address the problem of low real-time performance of the Labeled Multi-Bernoulli tracker in complex tracking scenarios with high target density and strong clutter,a Labeled Multi-Bernoulli tracking method based on maximum likelihood recursive update(MLRU-LMB)is proposed.The core steps are as follows:first,the maximum likelihood element is found from all combinations of measurements and Labeled Multi-Bernoulli terms;then,all measurements are used to update the Labeled Multi-Bernoulli terms corresponding to the maximum likelihood;then,the Labeled Multi-Bernoulli terms and measurements corresponding to the current maximum likelihood element are removed;finally,the above steps are recursively executed until the update is completed.In order to fully demonstrate the effectiveness of the proposed method,its performance is verified from the aspects of detection probability,clutter rate and target number.The experimental results show that the proposed MLRU-LMB tracking method has high tracking accuracy,strong robustness and excellent real-time performance,which can be summarized as follows: 1)When the clutter density is low,the target number is small and the detection probability is high,MLRU-LMB and Gibbs-Labeled Multi-Bernoulli(Gibbs-LMB)have similar performance;2)When the clutter density is high,the target number is large or the detection probability is low,MLRU-LMB can still maintain good tracking quality.Gibbs-LMB can improve the tracking effect by increasing the number of samples,but at the expense of increasing the computational complexity;3)MLRU-LMB has achieved a good balance between tracking effect and computational efficiency.(2)Building on the work(1),an iterative correction fusion tracking algorithm is proposed to address the poor single sensor tracking performance in complex tracking scenarios with low detection probability and strong clutter.Firstly,MLRU-LMB is implemented using measurements of the first sensor;then,the posterior trajectory estimation result is used as the prior information of the second sensor to continue executing MLRU-LMB;finally,the process is repeated until the update of the last sensor is completed,and the fusion trajectory estimation result is acquired.Through four different tracking scenarios,it is verified that the multi-sensor MLRU-LMB fusion tracking method based on iterative correction has good tracking performance and robustness,as summarized below: 1)The fusion tracking method achieves global consistency estimation of the targets’ states and labels;2)The fusion tracking method can detect the targets faster and the tracking accuracy is improved compared to the single sensor.(3)Based on the work(1)and(2),a new fusion strategy,the Matching Fusion Tracking Algorithm,is proposed based on the single sensor MLRU-LMB tracking algorithm.Firstly,each sensor uses its own measurement to obtain its posterior trajectory estimation by MLRU-LMB;then,by correlating the posterior state estimation of the two sensors,the state global consistency estimation is realized by covariance intersection fusion,and the reference label space is the previously fused label set,and the strategy of continuously expanding the label set is used to realize the label global consistency estimation;finally,the fusion result is repeatedly matched and fused with the other sensor,and the trajectory estimation result is acquired.Through four different tracking scenarios,it is verified that the multi-sensor MLRU-LMB fusion tracking method based on matching fusion has good tracking performance and robustness,which can be summarized as follows: 1)The fusion tracking method realizes the global consistency estimation of the targets’ states and labels;2)In the complex environment with strong clutter and low detection probability,the proposed matching fusion method effectively improves the tracking quality of the target.This method has higher tracking accuracy,compared with the iterative correction method,which is a fusion tracking method with better tracking quality fusion and easy implementation.