Research On Non-degenerated Particle Filter Algorithm And Its Application In Target Tracking

The traditional particle filter has the problem of degeneration and the sampleimpoverishment caused by re-sampling, which, to a certain extent, restrictes thedevelopment of the particle filter. For particle filter, the application environment isoften complex, which also put forward higher requirements for the estimationaccuracy and robustness of the particle filter. Based on the depth exploration of thecurrent situation of the particle filter, the dissertation researches on thenon-degenerated particle filter algorithm, and put it into the combination withtarget tracking in a dynamic context. The main works of the dissertation can beorganized as follows:1. For the degeneracy and the sample impoverishment caused by re-sampling inthe particle filter, a non-degenerated particle filter algorithm is proposed byintegrating the idea of particle swarm optimization into the sample stage of theparticle filter. The algorithm selects the particle state with the maximum weight ofthe last moment as the optimal value, and then moves the particle re-sampled to thehigh likelihood region around the optimal value according to the designed strategy,which can increase the diversity and effectiveness of the particles. The dissertationproposes the proposal distribution, and analyzes the computability of the newproposal distribution theoretically.2. For the deficiency of the method of discarding/keeping particles and sampleimpoverishment caused by duplicating particles, a non-degenerated particle filterre-sampling algorithm is proposed. The algorithm uses a new method to discard andreserve particles, to ensure that the remaining particles are the ones with largerweights, and then uses the Gauss method to generate new particles, rather than copythe particles repeatedly, which increases the diversity and effectiveness of theparticles at the stage of re-sampling, and then relieves the degeneration of theparticles effectively.3. Researches the non-degenerated particle filter on the application of targettracking in the dynamic context, implements the target tracking algorithm based onthe non–degenerate particle filter. The proposed algorithm is still able to keepbetter tracking results in the situations of rapid movement, partial occlusion andmorphological changes, illumination changes. The comparative experiments showthat the proposed algorithm is superior to the standard particle filter trackingalgorithm and is more adaptive to the dynamic context.