PHD Smooth Multi-target Tracking Algorithm Based On Box Particle Label

Multi-target tracking(MTT)technology has always been an important subject in the tracking field.The development of random finite set(Random Finite Set,RFS)has promoted the research based on RFS tracking algorithm.Among them,Probability Hypothesis Density(PHD)filtering is one of the most important algorithms.After smoothing the filtering results,a more accurate multi-target state estimation can be obtained.The particle filter algorithm with label box is a new algorithm proposed in recent years,which is used to improve the tracking efficiency such as operation efficiency.Based on the PHD algorithm,this paper focuses on the box particle PHD filtering,labeled PHD smoothing and labeled box particle PHD smoothing algorithms.The main work is as follows:(1)The box particle PHD filtering algorithm is studied.First,the Gaussian implementation of PHD—Gaussian Mixture Probability Hypothesis Density(GM-PHD)filtering algorithm is given.Aiming at the problems of measurement uncertainty and high calculation intensity.Using the principle of interval analysis,when the measurement is updated,thebox particles are used to replace the traditional point measurement to fit the posterior probability density of the target,so as to perform the filtering process.It is verified by MATLAB simulation experiments that the algorithm can effectively solve the problem of large calculation burden caused by high-dimensional integration operations.(2)The PHD smoothing algorithm with labels is studied.The basic principle and algorithm steps of GM-PHD smoothing algorithm are mainly studied.To solve the problem of erroneous estimation caused by the disappearance of targets in multi-target tracking,the intermediate time is corrected by using the number of target estimates at the two times before and after through the smoothing algorithm;secondly,label information is added for each Gaussian item on the basis of GM-PHD smoothing It evolves together with the Gaussian term,and then distinguishes different tracks by tags.It is verified by MATLAB simulation experiments that the labeled GM-PHD smoothing algorithm has significantly improved tracking accuracy when the number of targets changes suddenly and the track crosses.(3)The PHD smoothing algorithm of box particle with label is studied.This paper focuses on the implementation of the combined algorithm of labeled PHD smoothing and box particle PHD filtering.For the problem of complex calculation steps due to backward smoothing suppression in labeled PHD smoothing,the box particle labeled method can reduce the amount of calculation.It is verified by MATLAB simulation experiments that when the target track crosses,the computational efficiency of the GM-PHD smoothing algorithm based on box particle labeling is improved,and it has good tracking performance.