| Target tracking is the process of estimating and predicting target position information and parameter information based on the observation data obtained by sensors.It has important applications in the fields of military,aerospace,radar navigation,security monitoring and intelligent transportation.This thesis studies the problem of multi-target tracking based on the probability hypothesis density filter.The main work is as follows:(1)In the GM-PHD filter,both noise of the system states and measurements are Gaussian distribution,which cannot completely reflect the distribution of real scene,so we supposed introduce Closed Skew Normal distribution which is a generalization of Gaussian distribution to PHD filter to derive the CSNM-PHD(Closed Skew Normal Mixture Probability Hypothesis Density)filter and conduct simulation experiments.The results show that CSNM-PHD filter is superior to GM-PHD filter in comprehensive performance,and it can track the target state and estimate the target number more correctly with a prior performance.(2)In order to provide accurate and reliable measurement data for target tracking,this thesis proposes the VibeImp algorithm to improve the performance of the ViBe algorithm which is prone to "ghosting" and has poor adaptability to process noise,illumination and dynamic environment.So the initial background used for initial background sample model is given by the multi-frame average method.And in the procedure of foreground detection,a self-adaptive calculation method of radius threshold is given by combining background difference,frame difference and OTSU algorithm.At the same time,the foreground detection and model update is conducted only in the motion region which is obtained according to the background difference method to reduce the computational complexity of the algorithm.Compared with other several improved algorithms in 25 different scenes video,the ViBeImp algorithm has superiority both in visual effects and objective evaluation indicators.(3)A video target tracking algorithm based on CSNM-PHD filter is proposed to solve the problem of tracking difficulty caused by changes in target number,states and video background environment.Firstly ViBeImp algorithm is used to obtain accurate and reliable target position and size information as measurement,then,the CSNM-PHD filter is initialized by the initial frame detection result and the CSNM-PHD filter is updated by the input measurement so that the tracking algorithm can more accurately track targets in complex scenarios.Testing the algorithm in different video scenarios and comparing with GM-PHD filter,experiment results show that the combination of ViBeImp algorithm and CSNM-PHD filter can track targets stably and reliably under the changes of targets number and state,which achieves the robust multi-target tracking in video. |
