Research On Improved Method Of LMB Filter Based On Dynamic Multipath Noise Model With Millimeter Wave Radar

In the event of emergencies such as fires in closed spaces like large building complexes or underground roads,millimeter wave radar is capable of transmitting electromagnetic waves that can penetrate smoke,flame,and high temperature fields and can be used to detect and estimate number of pedestrians as reference information for victims counting and trajectory estimation in disaster relief operations.Since the electromagnetic waves transmitted by millimeter wave radar have a stronger multipath reflection phenomenon during propagation in closed space than in open space scenes.This causes the radar to generate noise sequences similar to moving target trajectories in closed space observation scenarios.This multipath effect can bias the target number estimated by the Labeled Multi-Bernoulli Filter(LMB Filter).To address this problem,this paper analyzes the characteristics of millimeter wave radar point cloud data affected by the multipath effect and further carries out several major research efforts as follows :(1)In this paper,we analyze and construct a model of the spatio-temporal distribution characteristics of millimeter wave radar for two-dimensional point cloud data of moving pedestrians in closed space scenes.The model classifies data noise into three categories: static noise,dynamic outlier noise and dynamic multipath noise.Among them,the dynamic multipath noise sequences are highly similar to the moving target sequences,which is the main cause of multi-target tracking accuracy decrease and target number estimation bias.In this paper,based on the formation mechanism of dynamic multipath noise,the spatio-temporal characteristics of its relative moving target trajectory are modeled.Using the actual pedestrian data collected in the closed space,this paper verifies the consistency of the dynamic multipath noise sequence feature model with the real data.(2)A multi-target tracking algorithm based on Random Finite Sets(RFS)theory,the LMB filter,and its main principles are introduced.The impacts of the dynamic multipath noise produced in the closed space on the prediction,update,and state estimation stages of the LMB filter are analyzed,as well as the causes of the target number estimate’s bias.(3)The Hybrid Clutter Model-based LMB(HCM-LMB)Filter is proposed to address the problem that dynamic multipath noise tends to bias the target number estimation of the LMB filter.Based on the characteristics model of dynamic multipath clutter introduced above,HCM-LMB filter’s clutter model is improved,and a false track label management mechanism is implemented at the same time.By comparing the pedestrian target number estimation experiments with LMB filter and HDBSCAN(Hierarchical Density-based Spatial Clustering of Applications with Noise)clustering algorithm,it is demonstrated that the HCM-LMB filter reduces the target number estimation in closed scenes relative to LMB filter and HDBSCAN reduces the bias of target number estimation by 22.5% and 26.9%,respectively,in closed scenes.