Research On State Estimation Methods In Moving Targets Tracking System

Target tracking is one of the basic problems in nature. While tracking reliably and accurately is always the main goal of designing target tracking system. With the rapidly development technology of computer, sensor manufacturing and information processing, the target tracking problem occurred with unprecedented profound changes. The estimated system is further revealed with the characteristics of non-linear non-Gaussian during system modeling, the multi-platform, distributed and uncertain of sensor measurement, and the multi-mode of moving target. On one hand, the perceived environment and perceived objects are complicated and changeable, the perception methods are gradually diversity, and the perception capability is increasingly enhanced; on the other hand, the requirements for perception task such as real-time, reliability, robust and accuracy are proposed strictly in target tracking system. The development, improvement and application of estimation theory faced with more severe challenges for the emergence of these new features. Taking filter, the basic tool to achieve target state estimation, as an impetus, this paper launches the research of three key technology problems of state estimation in moving target tracking system in order to the further developement and improvement of moving target tracking theory. The main contributions are as follows:Aiming at the adverse effects of sensor measurement noise random characteristic on filter state estimation accuracy, a linear measurement sampling bootstrapping strategy is proposed and applied to standard Kalman filter. In the strategy, considering the multi-sensor fusion theory a measurement sampling set is constructed by combining current measurement and noise priori statistical information. On this basis, the two implementation structures-- the measurement sampling distributed weighted fusion and the measurement sampling centralized consistency fusion-- are designed respectively. Considering the nonlinear problems caused by coordinate conversion in target state evolution model and measurement model, a nonlinear measurement sampling bootstrapping strategy is built by combining with nonlinear measurement characteristics, and dynamically applied to cubature Kalman filter for dealing with nonlinear estimation problem. Combined with weighted fusion theory, a measurement sampling bootstrapping cubature Kalman filter based on distributed weighted fusion is proposed.Aiming at adverse effects result from system error on state estimation precision in linear system, a novel algorithm which jointly estimates system error and state based on Kalman filter is proposed for linear system estimation. In the algorithm, considerating with Kalman filter state prediction mechanism and measurement update mechanism, system error is modeled as a state variable which is to be estimated through state augmentation technique. On this basis, in allusion to the state estimation problem of nonlinear system with measurement system error, combined with the modeling of nonlinear system error augmentation and the reasonable optimization of cubature Kalman filter, an augmentation cubature Kalman filtering algorithm is also proposed for jointly estimating system error and state in nonlinear system.Aiming at the multi-target tracking problem with unknown or time-varying targets number in nonlinear Gauss scene, combined with cubature Kalman filter and Gaussian mixture probability hypothesis density filter, a cubature Kalman probability hypothesis density filter for single sensor measurement system is designed for nonlinear dynamic model estimation. Furthermore, according to membership function in fuzzy set theory, the reliability of measurement is evaluated utilizing the mutual support degree among all measurement. Furthermore, through constructing and calculating confidence distance and consistency matrix denoted support degree between sensor measurements the fused measurement is achieved. On this basis, a multi-sensor consistency fusion cubature Kalman probability hypothesis density filter is finally proposed.Considering the research on state estimation problems in modern moving target tracking system, different state estimation algorithms are proposed or improved respectively according to practical applications in this paper. The theoretical analysis and simulation results verify the validity of proposed algorithms.