Research On Target Tracking Algorithms Based On Wireless Multi-Sensor Fusion Estimation

The target tracking technology has been applied broadly and successfully to the military and civilian applications,such as the national defense,navigation,guidance,detection,localization,transportation,manufacturing,medicine,etc.Wireless sensor network(WSN),which is of low cost,low power consumption and self-organization,etc.,can be used to apperceive mobile targets cooperatively by a large number of distributed nodes,and provide rich environment information and accurate positioning services.The problem of moving target tracking or localization in WSNs has become hot research topics in society of information fusion estimation.Target tracking based on ranging technology,which is a problem of nonlinear multisensor fusion estimation,is typically affected by the sensor failure,inter-sensor interference,model uncertainty,etc.Moreover,for a target tracking system in WSNs,new problems and difficulties in the analysis and synthesis are introduced due to the limitation of energy and bandwidth,as well as the communication uncertainty,such as packet loss,delay,etc.Based on the Bayesian estimation theory,Lyapunov theory,linear minimum mean square error criterion(LMMSE)and consistent estimation criterion,the problem of multirate/asynchronous fusion estimation is investigated in target tracking systems,and target tracking estimators are designed for different scenes to track mobile target in WSNs.The main results are obtained as follows.1)The problem of distributed maneuvering target tracking in multi-rate WSNs is investigated.The covariance of process noise is considered adequately to compensate for modeling error.Firstly,local estimates satisfying the requirement of consistency are obtained by a modified strong tracking filter(MSTF)method.Secondly,the covariance intersection(CI)fusion estimation method is used to reduce the uncertainty of the estimate.The target in multi-rate WSNs is tracked by the method in the hierarchical fusion framework,which merges the merits of the MSTF method and CI method.2)Considering sensor failure and limited measurement range,an event triggered sampling and transmission mechanism is designed for sensor selection.Firstly,some processes noise covariances are generated randomly,and a decentralized extended information filter(DEIF)is used to generate local estimates with the different hypotheses of modeling error.Furthermore,the CI fusion estimation method is used to generate the fused estimates which is more consistent with the current motion characteristics.Similarly,such a estimation structure is extended to the case where the measurement accuracy is not satisfactory and the target is tracked by fusing the RSSI measurements.3)The problem of target tracking in asynchronous WSNs is investigated.By introducing a time-varying fading factor,the convergence and stability of the unscented Kalman filter(UKF)are improved,and this kind estimator is termed as unscented strong tracking filter(USTF).Moreover,the switching condition of the filters is derived from the consistency criterion,which merges the merits of the traditional UKF method and the USTF method.Finally,the target in asynchronous WSNs is tracked by the hybrid sequential fusion estimation method.4)The problem of RSSI-based target tracking with a sequential Gaussain method is investigated.Firstly,by the Gauss hypothesis,a sequential Gaussian filter method is derived from the sequential Bayesian filter.Moreover,a sequential cubature Kalman filter method is obtained by applying the cubature integral rule.Secondly,an adaptive factor is introduced by hypothesis testing method to improve the stability and convergence of the estimator.Finally,by the analysis of the convergence of the estimator,it has been proved that the adaptive factor is helpful to the convergence of the estimator.5)The problem of a WSNs-assisted self-localization of mobile targets is investigated.To reduce communication conflicts and relieve channel competition,an event-triggered mechanism is employed by mobile targets in the process of sampling.A lower bound is given to select the parameters of process noise covariance,and accordingly,a target localization estimator with multiple cubature points is designed for self-localization of mobile targets in event-triggered WSNs.The effectiveness of the proposed methods are verified by computer simulations and E-puck based target tracking experiments.Finally,the conclusion and future work are presented.