Research On Distributed Information Fusion Algorithm

Information fusion technology processes data and information obtained from single and multiple information sources through association,correlation and synthesize,and can obtain position and identity state of observed target accurately.The distributed fusion structure for information fusion processes measurement on local sensor,and improves the reliability and survivability of the system compared with centralized fusion structure.Distributed target tracking can reduce computational load of fusion center and improves the stability of the system.In order to improve the tracking accuracy and reduce usage of the sensor,researches of distributed target tracking and sensor management algorithms are performed in this paper.The main work are listed as follows:Firstly,current progress of distributed target tracking and sensor resource management technology at home and abroad are reviewed.The basic theory of distributed target tracking is introduced,and the structure of distributed target tracking and basic algorithms are presented.Secondly,to solve the problem of distributed target tracking with linear systems and gaussian observation noise,a distributed kalman filtering algorithm based on weighted information consistency algorithm is proposed.It firstly combines the consistency algorithm with kalman filter algorithm,and performs the uniform processing on the local estimation value of the target.Local state prediction value of the current time target is then corrected using the target state estimation value of the neighbor node,where local prediction values are also weighted.The consistency of local prediction value and estimation value is improved,so the consistency of global information is also improved.The simulation results demonstrate that,the proposed algorithm reduces the error of average consistency estimation and improves the target tracking accuracy.Thirdly,to solve the problem of distributed target tracking with linear systems and observation noise as non-Gaussian noise,a distributed particle filter algorithm based on Gaussian mixture model(DGMMPF)is experimentally compared.Gaussian mixture models are used to approximate posterior probability distribution of weighted particles,and consistency algorithm is used to exchange parameters ofgaussian mixture model around neighborhood,where posterior distribution represented by gaussian mixture model is obtained for each node.Since the model parameters are Only exchanged,the algorithm reduces the computational load.In order to reduce the sensor usage of the distributed structure,a sensor management algorithm based on GMMDPF and distributed posterior Cramer-Rao lower bound(d PCRLB)is presented.It adpots d PCRLB as sensor selection criterion,and selects a subset of sensors which minimize d PCRLB to perform object estimation with DGMMPF.Simuation results show that,the proposed algorithm reduces usage of sensors compared with DGMMPF,and improves target tracking accuracy at the same time.Finally,the main work and further research of the thesis is summarized.