Research On Extended Target Tracking Algorithm Based On Multi-sensor Information Fusion

Multi-sensor extended target tracking is more mutual cooperation through multiple high resolution sensor,the scene within the extended target centroid state,size of shape and number estimation effectively.In recent years,with the continuous development of Random Finite Set(RFS)theory,Bayesian multi-target tracking technology based on RFS theory has been widely concerned by scholars at domestic and overseas,and with the continuous improvement of sensor accuracy,target tracking gradually transients from the previous point target tracking to extended target tracking.Many multi-extended target tracking algorithms based on RFS have been developed to meet the ever-changing target tracking requirements.With the continuous development of communication technology,many scholars have proposed multi-sensor multi-target tracking algorithm based on multi-sensor information fusion,which improves the disadvantage of limited field of view of single sensor and further improves the target tracking accuracy.However,extended target tracking technology in multi-sensor networks has many challenges and problems,such as extended target estimation correlation problem.Based on multi-sensor information fusion,this paper mainly studies the distributed multi-sensor extended target tracking method,and makes three research achievements,as follows:1.The Arithmetic Average Extended Target Multi-Bernoulli tracking algorithm(AA-ET-MB)was proposed.The arithmetic average extended target multi-Bernoulli tracking algorithm was implemented based on Gaussian mixing.First,each sensor runs a local ET-MB filter without considering the shape of the extended target.Then,the Hungarian algorithm is used to realize the Bernoulli association of the same target between different sensors.Then,the AA fusion with consensus information is used to achieve the Bernoulli posterior density fusion of the same target and feed back to the current sensor to improve the accuracy of the next filtering iteration.The results show that the proposed algorithm has better tracking accuracy for the extended target’s position in multi-sensor networks.2.Based on Gaussian inverse Wishart-Probability Hypothesis Density filter,An Arithmetic Average Gaussian Inverse Wishart Probability Hypothesis Density(AA-GIW-PHD)tracking algorithm was proposed.Firstly,Gaussian inverse Wishart distribution is introduced to model the position and shape of the extended target,and thus to realize the extended target GIW-PHD filter with a single sensor.Each sensor runs a local GIW-PHD separately to obtain a posterior probability density.Secondly,flood communication is introduced to realize the sharing of posterior density information between adjacent sensors quickly and effectively.Subsequently,the AA fusion method is used for posterior probability density fusion,and finally feedback to each sensor for the next filtering iteration.The results show that the proposed algorithm can effectively estimate the location and contour size of the extended target centroid in distributed sensor networks.3.Based on Gamma GIW-Cardinality Balance Multi-target Multi-Bernoulli,A distributed multi-Bernoulli extended target tracking algorithm based on Arithmetic Average Gamma GIW MB(AA-GGIW-MB)was proposed.Firstly,on the basis of GIW,Gamma distribution is introduced to approximate target Poisson rate measurement.Thus,the extended target state was modeled as Gamma Gauss Inverse Wishart distribution(GGIW)and the GGIW-Cardinality Balance Multi-target Multi-Bernoulli(GGIW-CBMe MBer)filtering and tracking algorithm was implemented.A posteriori probability density is obtained by distributing local GGIW-CBMe MBer filters to each sensor,and then the posteriori density information is shared quickly and effectively by using flood communication.In order to effectively realize the posterior density correlation of the same extended target under different sensors,the elliptic distance is proposed by combining the Euclidean distance of the centroid and the non-Euclidean size-shape metric of the shape matrix.A posteriori probability density of the same extended target is divided into the same subset by elliptic distance,and then AA fusion rules are used to obtain the approximate spatial density of GGIW distribution after fusion.Simulation results show that the proposed algorithm can effectively track extended targets in distributed sensor networks.