Font Size: a A A

Research On Application Of Kernelized Correlation Filters And Particle Filters Algorithm In Video Target Tracking

Posted on:2020-01-10Degree:MasterType:Thesis
Country:ChinaCandidate:G C DuFull Text:PDF
GTID:2428330575489328Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Video tracking has many important applications,such as intelligent transportation,virtual reality,precise guidance and so on.Nowadays,the diversity of video leads to a variety of problems,such as partial occlusion,change of light intensity and complex background,which undoubtedly increases the difficulty of tracking algorithm.So,our work mainly studies and discusses the two problems about partial occlusion and complex background,and integrates the predicted results of improved kernel correlation filtering and particle filtering through online detection algorithm.When partial occlusion and complex background occur,the target can be tracked better.The main work of this thesis is as follows:Firstly,two commonly used video tracking models,discriminant model and generative model,are introduced,whose disadvantages and advantages are discussed.Then,the representative algorithms are selected in the two models,respectively,the kernel correlation filtering and particle filtering.Therefore,the classification result of the classifier in the kernel correlation filtering is improved by introducing the loss function in the online detection algorithm to judge whether the classification result is correct.If the classification is wrong,the relevant parameters of the classifier are updated to get the correct classification result,so as to improve the expression of the target model.The improved kernel correlation filtering is always a kind of discriminant tracking.Particle filter belongs to generative tracking and has good effect in the expression of target model.Finally,the improved correlational filtering and particle filtering are complementary fusion for the shortcomings.The fusion strategy is to use the fusion function to fuse the target position predicted by the improved correlational filtering and the target position predicted by the particle filtering.The result is better tracking.A part of video is selected from the Visual Tracker Benchmark dataset for comparative analysis with other algorithms,and the validity of the proposed method is verified by quantitative analysis and qualitative analysis respectively.In terms of quantitative analysis,the average distance measurement and the average success rate are commonly used to measure the quality of the algorithm.The average distance measurement of the method in this thesis is 91%,which is 11.36%higher than the kernel correlation filtering and 25.8%higher than the particle filtering.The average success rate of the proposed method is 94.5%,which is 17.98%higher than that of kernel correlation filtering and 31.9%higher than that of particle filtering.In terms of qualitative analysis,the comparison of experimental results can directly verify the effectiveness of the proposed method.
Keywords/Search Tags:video target tracking, kernelized correlation filters, Particle filters, Associative function
PDF Full Text Request
Related items