Group Target Tracking Algorithm In Complex Environment

Different from traditional point targets,group targets are composed of a group of multiple targets whose spacing meets certain constraints and have similar movement patterns.These targets can split,merge,get close to each other,or move independently of each other to a large extent.Group targets involve video surveillance,military applications,situational awareness and many other fields.Its tracking problem has been concerned by scholars.Because the random finite set has the advantages of small computation and high tracking accuracy in target tracking,the combination of group target contour modeling,internal relationship modeling and random finite set filtering can achieve good tracking.Therefore,based on the random finite set filter,this paper studies the group target tracking method in complex environment.As follows:1.In order to solve the problem of group target tracking in spawning scenarios,a new Poisson multi-Bernoulli mixture filter algorithm was proposed by combining the group target splitting model with the standard Poisson multi-Bernoulli mixture filter.This filtering algorithm uses random matrix method to model group target shape.In the prediction step,the standard Poisson multi-Bernoulli mixture filtering algorithm is first used to predict the state of the target at the next moment,and then the group target splitting criterion is used to predict the group target splitting event that meets the conditions.There are many kinds of splitting modes of the target,and we predict the splitting events of multiple populations with multi-hypothesis structure.Finally,the prediction of spawned target and target did not spawn are put together into the update step.Simulation results show that the algorithm improves the tracking performance of spawned target effectively.2.In order to solve the problem of unknown newborn target parameters,an adaptive newborn group target multi-Bernoulli filter algorithm is proposed to predict newborn target.The algorithm firstly selects the measurement used for newborn target through likelihood,and then uses the selected measurement information to obtain the position of newborn target at the current moment.Finally,the newborn target replaces the traditional fixed newborn and puts it into the prediction results for recursive operation.Simulation results show that the proposed algorithm can achieve the same tracking accuracy as traditional algorithms,and has strong robustness in different clutter rates and detection probabilities.