Method and device of particle filtering and target tracking

Abstract

The invention discloses a method and device of particle filtering and target tracking. The method of particle filtering comprises the following steps that a probability density function of multiple integral points at the observation time of the last target is built by the adoption of the Gauss-Hermite integral point, an approximate particle set of the integral points is acquired according to the probability density function of the integral points, a predicted particle set is acquired by amending the approximate particle set according to relative features of the target, a predicted probability density function of a target state at the observation time of the current target is acquired according to the predicted particle set, and a posterior probability density function of the target state at the observation time of the current target is acquired according to the predicted probability density function of the target state at the observation time of the current target. According to the method, diversity and accuracy of particles are enhanced, and filtering accuracy and target state estimation performance are improved greatly.

Description

technical field [0001] The invention relates to the field of nonlinear filtering, in particular to a particle filtering method and device, and a target tracking method and device. Background technique [0002] During the movement of aircraft, aviation vehicles, vehicles and other targets, it is often necessary to estimate the real-time state of the target in order to realize the tracking of the target, and the motion system model of the target such as aircraft is generally a nonlinear stochastic system. Nonlinear filtering technology is a common method for state estimation in nonlinear stochastic systems. [0003] According to different application backgrounds, existing nonlinear filtering technologies are mainly divided into two categories: the first category is for state estimation problems in nonlinear Gaussian environments, such as Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), integral Kalman filter (QKF), truncated unscented Kalman filter (IUKF), these m...

AI Technical Summary

Problems solved by technology

The second category is for state estimation problems in nonlinear non-Gaussian environments, such as Gaussian sum filter (GSF), Gaussian and integral Kalman filter (GS-QKF). This type of Gaussian sum method mainly uses multiple mixed Gaussian The posterior probability density function of the state is approximated as a single Gaussian function. However, similar to the above-mentioned EKF and other methods, such Gaussian sum methods must be linearized. For strongly nonlinear non-Gaussian systems, the filtering accuracy of such filters is not as good as is not high, and the number of Gaussian mixture terms of the filter grows rapidly with time

[0004] In addition, the existing technology also adopts another nonlinear filtering method: the particle filter method. Due to the existence of particle degradation, resampling is required for a type of particle filter method used in the prior art, which affects the parallel implementation of the particle filter.

Another type of particle filter method in the prior art does not require resampling, such as Gaussian particle filter (GPF), fast Gaussian particle filter algorithm, quasi-Monte Carlo-Gaussian particle filter (QMC-GPF) algorithm, etc. When the time is updated, the state transition function is simply used for particle sampling. When the sampling time interval of the target observation point is large or the target motion model is not accurate enough, the diversity and accuracy of the particles are poor, and the particles cannot effectively represent the target. test probability distribution, thereby reducing the performance of the particle filter

Images