Robust filtering method based on p norm optimization

Abstract

The invention discloses a robust filtering method based on p norm optimization. The robust filtering method comprises the following steps: representing process noise and observation noise through generalized normal distribution; introducing a cost function in a minimized p norm form based on a maximum posteriori estimation theory; obtaining observation update steady to the outliers; and obtainingbrand-new p norm sequential filtering. The beneficial effects of the robust filtering method are that: generalized normal distribution is used for representing process noise and observation noise; a cost function in a minimized p norm form is introduced on the basis of a maximum posteriori estimation theory; and steady observation update for the outliers is derived according to an M-estimation theory, so that a brand-new p norm sequential filtering method is obtained, and steady estimation of the ground target state in the presence of the observation outliers is realized.

Description

Technical field [0001] The invention relates to the technical field of ground target tracking systems, in particular to a robust filtering method based on p-norm optimization. Background technique [0002] The ground target tracking system (lidar, camera system, etc.) is affected by the observation conditions, and errors are introduced in the later observation data extraction, which leads to a greater probability of measurement abnormal data, and abnormal data often deviates far from the true value, that is, the observation field Value (outlier). Outliers cause the uncertainty of the observation to no longer satisfy the Gaussian distribution. [0003] Therefore, the use of Gaussian model assumptions will cause significant disturbances in the estimation of the ground target state, resulting in large estimation deviations, or even destroying the convergence of the filter. Common ground target trajectory estimation methods, including traditional batch least squares and sequential Ka...

AI Technical Summary

Problems solved by technology

However, these methods assume that the observed data errors satisfy the Gaussian distribution, so they are sensitive to the observed outliers

Obviously, the least squares method cannot weaken the influence of outliers in data fitting, so it cannot provide high-precision solutions, and even cannot guarantee the stability and convergence of the algorithm.

Images