Font Size: a A A

Research On Object Tracking Algorithms Based On Improved Particle Filter

Posted on:2018-04-05Degree:MasterType:Thesis
Country:ChinaCandidate:Z W SuFull Text:PDF
GTID:2348330518486558Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Moving object tracking is a very important research direction in the field of computer vision.With the research on moving object tracking technology has been carried out extensively,the moving object tracking technology has been developed rapidly.At the same time,People's requirements on moving object tracking technology increase with each passing day,how to track object accurately and stably under complex surrounding is always a difficult problem in moving object tracking field.The research work of this paper mainly includes the following aspects:1?In this paper,a multi-region sampling object tracking algorithm based on optimization weight is proposed in order to solve the problems of regional diversity and the tracking precision decreased,object tracking instability introduced by object tracking method based on multi-region sampling.In our method,region optimization weight and improved sub-region resampling method is proposed,the regional confidence level of each sub-region is optimized by using the optimization weight to appropriate increase the number of particles low regional confidence level region acquired during resampling phase,under the premise of ensuring the particles is effective allocation according to the regional confidence level,the regional diversity decreasing phenomenon is impactful restrained.In sub-region,the particle weight optimization weight is used to optimize particle weight and set resampling threshold,so as to alleviate particle impoverishment and make full use of effective particle information.Experimental results show that the proposed method can effectively improve the object tracking accuracy and stability.2?In order to solve the problem of particle impoverishment phenomenon caused by particle resampling and the poor robustness problem of using single feature in the object tracking process,an adaptive multi-feature fusion object tracking algorithm based on information retention is proposed.The proposed strategy of information retention can effectively alleviate particle impoverishment and the weight values of the particles with small weight values are increased,by optimizing the distribution of particle weight values,more particle information is reserved by improving particle resampling method.Feature effectiveness is affected by changing environment and contribution to tracking object of each features,so the weight of each feature component for multi-feature model is adaptively adjusted.Experimental results show that the proposed algorithm can effectively deal with some challenging scenarios such as object deformation,partial occlusion,and interference of similar objects,which has high tracking accuracy and good robustness.3?Aiming at the problem that track moving object is easily influenced by the complex environment and occlusion,an object tracking algorithm based on global multi-feature fusion and local meanshift is proposed.In this algorithm,the object region is divided into several sub-regions,the particle filter method is applied to track the global region of the object,at the same time,the meanshift method is used to track the sub-region of the object.The improved particle filter object tracking algorithm with color feature and FDF feature fused is used to track the global region of the object,The meanshift object tracking algorithm with color feature and LBP feature fused is used to track the sub-region of the object.The robustness and occlusion resistance of the tracking algorithm was improved by adaptively adjust the contribution of the global and local information according to degree of occlusion and multi-feature fusion.Experimental results show that the proposed algorithm can effectively deal with object deformation,object occlusion,and interference of complex background,which has pretty high tracking accuracy and good robustness.
Keywords/Search Tags:Object tracking, Particle filter, Multi-region sampling, Multi-feature fusion, Local space information
PDF Full Text Request
Related items