An Intelligent Mixed Population Optimization Filtering Tracking Method

Abstract

The invention discloses an intelligent mixed group optimization filtering method. Firstly, the particles are layered according to the weight; then according to the number of particles in different layers, different movement modes are selected for the particles in different layers; and then the state of the particles is selected. Estimation, using the conditional mean or the state with the maximum posterior probability density as the estimated value of the system state; then updating the particle state to generate a suitable proposal distribution, so as to accurately estimate the position of the target at the current moment; finally, the particle state The purpose of prediction is to estimate the state of the target more accurately at the next moment, that is, to design a suitable prior distribution function. The method of the invention can more accurately estimate the posterior state in the nonlinear system, and exhibit higher tracking accuracy in complex and changeable scene environments.

Description

technical field [0001] The invention belongs to the field of computer vision and relates to state estimation in target tracking, in particular to an intelligent mixed group optimization filter tracking method. Background technique [0002] The most popular method for estimating the posterior state in object tracking is the particle filter algorithm (PF). The PF algorithm adopts the sequential Monte Carlo method (SMC), uses a group of samples (ie particles) to approximate the posterior distribution of the nonlinear system, and then uses this approximate representation to estimate the state of the system. Compared with several other algorithms, the PF algorithm is more suitable for nonlinear systems, has a wider range of applications, and has better actual results. Among the current mainstream visual tracking algorithms, such as CNT algorithm and IOPNMF algorithm, are all tracking algorithms based on particle filter. However, the particle filter algorithm cannot avoid partic...

AI Technical Summary

Problems solved by technology

However, increasing the number of particles will lead to an increase in the amount of calculation, which doubles the running time, and the real-time performance of the algorithm disappears.

The method of resampling is to take out particles with smaller weights and only copy particles with larger weights. However, as resampling proceeds, particles with large weights are copied continuously, and finally the types of particles drop sharply, resulting in poor samples. problem

[0003] Another problem that the particle filter algorithm needs to solve is that during the particle state transition process, the transferred particles must be able to appear in all possible positions of the target, otherwise the tracking may gradually move away from our tracking target, and eventually lead to the loss of the tracking target

Increasing the number of particles can also solve this problem, but obviously increasing the number of particles will lead to an increase in the amount of calculations, making the running time doubled, and the real-time performance of the algorithm disappears.

Images