Research On Multi-sensor Information Fusion Algorithm Based On Random Finite Set

Multi-sensor target tracking is to track and estimate the state and number of the target effectively through the cooperation of multiple sensors.Information fusion is a key algorithm in multi-sensor network,which can extract more accurate data information by communication fusion of the data obtained by sensors in different directions in space.Through the cooperation of multiple sensors,the measurement accuracy of the system is improved,the perception dimension of the system is increased,the speed of information processing is accelerated,and the ability of environmental adaptation is enhanced.In recent years,the proposed random finite set provides a new theoretical basis for the research of multi-target tracking.This paper studies the information fusion algorithm and target tracking method based on random finite set in distributed multi-sensor networks,and obtains the following three research results.1.Aiming at the problem of multi-target tracking in distributed sensor network with limited perception range,a distributed field-of-view complementary multi-Bernoulli correlation arithmetic average fusion tracking method is proposed based on the tracking theory of multi-Bernoulli filter.Firstly,the field of view complementary method is used to make the measurement information of each sensor include the measurement information of the whole scene.On this basis,each sensor runs a local multi-Bernoulli filter respectively,and a posterior information is shared between among sensors through flooding communications.Then,the multi-Bernoulli posterior obtained by communication sharing is correlated by distance division in each fusion center,and the fusion estimation of the Bernoulli posterior of the same target is performed by arithmetic average fusion method.Finally,the simulation results of sequential Monte Carlo show that the proposed method can effectively track multiple targets in the distributed sensor network with limited sensing range.2.Aiming at the problem of labeled multi-Bernoulli target tracking in distributed sensor networks,we propose a new distributed labels Bayesian posterior arithmetic average fusion method based on label consistency.Due to the label inconsistency of the posterior labeled densities of different sensors,the performance of communication fusion is reduced.Therefore,an efficient tag matching method is proposed to solve this problem.We introduce the label reference space and match label with each sensor to unify labels.Secondly,the information sharing among adjacent sensors is carried out by the distributed flooding communication method,and the arithmetic average fusion of the registered label posterior is carried out according to the way of label.Simulation results show that the proposed method can effectively solve the problem of label inconsistency and realize efficient and robust AA fusion,which is superior to other methods.3.In order to solve the problem of model mismatch in the process of maneuvering target tracking,we propose a method of fusion tracking of maneuvering target by using interactive multi-Bernoulli filter in distributed finite field of view sensor network.In the interactive multi-Bernoulli prediction stage,the same Bernoulli component is predicted through different model state transitions,and the prediction results under different models are obtained.In the update process,the results of different model predictions are updated by means of the field of view complementation measurements,and the model weights are updated by the likelihood function,and the multi-Bernoulli posterior of local updates are obtained by the model weights and sums.Finally,the information sharing among adjacent sensors is realized through the flooding communication,and the Bernoulli component of the same target are estimated by the arithmetic average fusion method.The effectiveness of the proposed method is verified by simulation experiments.