Research On Multi-target Tracking Based On Random Finite Set

With the continuous update of sensor technology,multi-target tracking technology develops rapidly.In the complex multiple target tracking scenario,the time-varying number of targets and a large amount of clutter all lead to the performance degradation of traditional multitarget tracking based on data association.With the development of multi-target tracking technology based on random finite set,the target state and measurement are constructed as sets,so as to solve the problem of increasing computation due to correlation.This paper mainly studies multi-target tracking based on random finite set.Research on the basic algorithm of multi-target tracking based on stochastic finite.This part mainly introduces Probability Hypothesis Density filter(PHD),gaussian mixture and particle.Aiming at the problem that PHD needs the strength of new targets and can't realize the track identification,a measurement driven GM-PHD filter is proposed,which realizes the track identification by adding labels to the Gaussian components,and generates the strength of new targets by using the measured values obtained by each sensor,avoiding the prior information,Simulation results show the effectiveness of the algorithm.GM-PHD algorithm with low detection probability.Because of the missed target,the tracking performance of GM-PHD declines sharply.An improved GM-PHD filtering algorithm is proposed to solve this problem.By setting decision criteria to distinguish the missed targets,and using multi frame information to modify the weight of the missed targets,the impact of the missed targets on the GM-PHD tracking performance is reduced.Simulation results show that the missed targets have good tracking performance under low detection probability.Research on the improved Gaussian Mixture PHD algorithm for multiple extended maneuvering targets.An effective measurement set partition can reduce the computational complexity of the Extended Target Gaussian Mixture Probability Hypothesis Density(ET-GMPHD).To solve this problem,an improved dynamic grid is proposed to preprocess the measurement set.Simulation results show that this method can effectively filter out part of the clutter,reduce the number of observation subsets,and reduce the running time.In view of the low tracking accuracy of ET-GM-PHD for maneuvering targets,the input estimation model is introduced,and the acceleration component is added into the original target state matrix.The fading factor is introduced to improve the tracking performance of the algorithm for strong maneuvering targets.The simulation results show that the improved ET-GM-PHD can effectively deal with the rapid change of acceleration and has good tracking performance.