Research On Target Tracking And Sensor Registration Problem Based On Kalman Filtering

Target tracking has been widely used in navigation, aviation, transportation, and military-related fields. In these areas, maneuverability of the target is becoming more and more complex, the performance requirements of target tracking system is also increasing. In order to improve the performance of target tracking system, this article mainly discussed about three issues: select suitable filtering algorithm, optimize target model and eliminate possible deviations of sensors.Firstly, the performances of EKF, UKF and UEKF algorithm in two different systems were theoretically analyzed and compared. The simulation results showed that when the truncation error of the system was small and could be ignored, the three algorithms had the same precision.When truncation errors could not be ignored, the performance of the UKF algorithm was much better than EKF and UEKF.Secondly, we analyzed the defects of Current Statistical Model(CS). In the CS model, the variance of acceleration is determined by the extremum acceleration and the estimate value of acceleration at the previous time. When the target's acceleration varies suddenly and extremum acceleration remains invariant, the change of acceleration variance will lag, thus increases the tracking error. In order to improve tracking accuracy, parameter-adaptive algorithms for CS model are widely used. In this article, we analyzed and compared the performance of several parameter-adaptive algorithms, and optimized one of the acceleration extremum self-adaptive algorithms by adding inequality constraints for the acceleration extremum. The inequality was used to avoid acceleration extremum tends to zero, which might cause failure of the whole adaptive algorithm. Simulation results show that this can improve the accuracy of the system when strong maneuvering occurred.Finally, an algorithm for mobile multi-platform multi-sensor bias registration was proposed as the motion states of the sensors platforms were unknown. In the algorithm, the UKF(Unscented Kalman Filter) was used for state estimation. The state vector in UKF algorithm was an augmented state vector formed by the motion state of mobile sensor platforms and the target. At each time of UKF filtering, filtering bias was obtained, constituting a filtering bias set. In addition, Gaussian mean-shift(Mean-shift) algorithm was used to get the estimate of bias from the filtering bias set obtained by UKF filtering. The estimate of bias would in turn be an argument of the UKF filtering. Repeat UKF filtering and Mean-shift iterative calculation would finally get the convergence value of the bias. Simulation results showed that the bias of sensors could precisely obtained when the initial motion state in a given was precise enough.