Research On CPHD Filtering Algorithm In Multi-target Tracking

As an important branch in the field of information processing,multi-target tracking technology has broad application prospects in national defense and civil applications.Using Random Finite Set(RFS)theory to solve multi-target tracking problem can avoid complex data association,not only can ensure the accuracy of target estimation,but also has good real-time performance.Among them,the most representative are Probability Hypothesis Density(PHD)filter and Cardinalized Probability Hypothesis Density(CPHD)filter,however,in low signal-to-noise ratio environments,PHD filter has a great error in estimating the number of targets.CPHD filter corrects the problem that the PHD filter estimates the target number inaccurately,and improves the accuracy of the target estimation.This dissertation primarily researches the CPHD filter in multi-target tracking,The specific contents of this dissertation include:(1)Aiming at the problem of fixed interval smoothing in multi-target tracking,the cardinalized probability hypothesis density filter and the Rauch-Tung-Striebel(RTS)smoother are combined,and a cardinalized probability hypothesis density smoothing algorithm for RTS is given.Considering the problem of large output delay in the smoothing process,a piecewise RTS cardinalized probability hypothesized density smoother is proposed using the idea of piecewise smoothing.Firstly,estimation values are segmented using a fixed interval.Secondly,track-estimate is associated using Hungarian algorithm.Finally,the RTS smoothing is performed on the associated tracks.The experimental results show,the cardinalized probability hypothesis density smoother using piecewise RTS can estimate the target state more accurately comparing with the cardinalized probability hypothesis density filter,and can effectively avoid the problem of poor real-time performance when used RTS smoother directly.(2)In multi-targets tracking,it is usually assumed that the clutter information and the target state are independent of each other and subject to uniform distribution in the monitoring area,While in the actual environment,a large number of clutter is more easily generated around the target.Considering the correlation between clutter and state,a smoothing algorithm is introduced,and the Gaussian Mixture PHD smoothing filter with state-dependent Clutter is proposed.Firstly,the clutter intensity in the whole surveillance area is recalculated.Secondly,targets are divided into two categories,i.e.,surviving targets and new targets.The adaptive ellipsoid threshold is used to preprocess measurements.The measurements within the threshold are used to update the surviving targets.The measurements outside the threshold are used to update the new targets.Finally,the RTS smoother is used to reverse smoothing.Experimental results show that the proposedalgorithm has better tracking performance under this condition,and is superior to Gaussian Mixture PHD filter in unsmooth state and clutter related environment.Figure 14,Table 3,87 references.