Research Of Information Fusion For Distributed Networked Systems With Communication Constrains

The networked systems introduce the distributed strategy to suit the growing information computation and awareness requirement.Due to the capacity constraints of the communication bandwidth,networked systems inevitably exist network-induced phenomena,which usually deteriorate the system performance.This thesis is concerned with the issues of spatial localization and target tracking,and investigates the information fusion for the distributed structure with communication constraints.While the objective of the distributed fusion strategy is to improve the reliability of systems and the precision of localization.The main work is summarized as follows:Firstly,the precise image recognition obtained by the vision sensor is an essential fundament for accurately localizing the targeting.Based on the active contour model,a novel multiphase double curve(MDC)approach is presented for image segmentation.It adopts the maximum-likelihood(ML)estimation and the expectation maximization(EM)algorithm to establish N level set functions;while,the inhomogeneous image is segmented into 2N sub-regions.On each target sub-region,MDC develops the active double curve evolution function to be embedded into the level set equations,so that the image contour evolutions are able to be bilaterally extended.The proposed MDC has the robustness for suppressing noise disturbance,and improves the accuracy of image segmentation.Secondly,the complex dataset classification achieved from different sensors is investigated,and a novel robust dataset classification approach(NSKFCM)is proposed to traverse and classify dataset,based on neighbor searching and kernel fuzzy C-means methods subsequently.Some optimized strategies,including neighbor searching,controlling clustering shape and adaptive distance kernel function,are employed to solve the number of clustering,the stability and consistency of classification,respectively.As the theoretical analysis,NSKFCM method possesses the advantage of improving the robustness for suppressing noises,and alleviating the impact of parameters uncertainties for dataset classification.Moreover,the measurement dataset is better applied to distributed systems for information fusion.Thirdly,the state estimation issue for uncertain networked systems considering data transmission time-delay and cross-correlated noises is presented.A distributed robust Kalman filtering-based perception and centralized fusion method is proposed to improve the estimation accuracy from perturbed measurement.To describe the transmission time-delay and noise disturbances for distributed systems in the exchange measurement among neighbors,a weighted fusion reorganized innovation strategy is used to reduce the computational burden and suppress noise disturbance.Moreover,to obtain the optimal linear estimate,a fusion estimation approach is used for information collaboration by weighting the error cross-covariance matrices,which is the again optimal for each subsystem.This method fuses the local state estimation and measurement data to acquire higher estimation accuracy than each local counterpart.Fourthly,for a class of discrete-time stochastic uncertain systems,the modeling and filtering methods are presented.Data transmit from plant to filter are influenced by the presence of random delays,out-of-order packets and correlated noises.To determine the packet disorder of holding or dropping,the system model is established by utilizing two signal choosing schemes of zero-order-holder(ZOH)and logic ZOH,respectively.Based on the established model,a robust finite horizon Kalman-type filtering is proposed by augmenting state-space and minimizing error covariance methods,in which the upper boundary of estimation state covariance could be obtained from estimation covariance constraints.To improve the filter performance,a linear estimation-based delay compensation strategy is presented by employing the time-stamped reorganized measurement.Moreover,for solving the missing measurement and suppressing the computational burden,the artificial delay compensation approach is established by one-step prediction.The simulation results on dynamic tracking systems are performed to show the proposed filter has the ability to track the actual state.Finally,3D photoelectric sensing and positioning system is introduced,and the spatial localization method is analyzed based on linear charge-coupled device(CCD).For tracking and estimating the dynamic trajectory from the measured spatial target,the problems of modeling and estimation for a class of discrete-time uncertain systems are investigated,which are composed of network-induced random delays,packet dropouts and out-of-order packets.To drop packet disorders and improve system performance,the logic ZOH is used for establishing the system model.Based on the proposed model,a distributed measurement and centralized fusion estimation approach(DFERFH)is designed,which uses the robust finite horizon Kalman-type filtering for estimation variance constraints.Moreover,to obtain the optimal linear estimate,a weighted fusion estimation approach is used for information collaboration by the error cross-covariance matrices.Simulation results demonstrate that the proposed DFERFH method is able to alleviate the communicational burden,suppress the influence for measurement with communication constraints.Moreover,the objective of improving the localization precision can able to be achieved,and shows the potentials of the proposed method for real-life applications.