Advance Research On Algorithm Of Single Object Tracking In Complicated Scenes

Visual object tracking is a fundamental research topic in the field of computer vision,which plays a critical role in various applications,such as human computer interaction,driverless car,surveillance,and human motion analysis,to name a few.After several decades of visual tracking research,considerable progress has been made.But it still remains challenge for developing an all-situation-handled tracker which successfully handles all scenarios,such as partial occlusions,illumination changes,fast motions,camera motions,background clutter and viewpoint,etc.This dissertation,based on the existing traditional target tracking algorithm,covers the research of the motion model,searching method and appearance model.The main contributions of this work are summarized as follows:(1)A target tracking algorithm in high-order particle filtering is proposed.Aiming at the problems in classical particle filtering,this algorithm applys the high-order Markov Monte Carlo posterior sampling for object tracking.The target is represented by the two dimensional principal component analysis and the trendency information of the target motion.In the tracking process,the estimation of the target position is more accurate because of the information extracted from the multi-frame.To obtain the higher quality particles within short time,weight is introduced into the sequential importance resampling method to estimate the posterior density.To further gain the better samples,K-means clustering is used to select the more typical particles,which reduces the computational cost.Compared with the traditional tracking algorithms,the proposed tracker has a certain improvement on tracking precision and tracking speed according to the experimental results.(2)An advanced Wang-Landau monte carlo-based tracker for abrupt motions is proposed.Conventional tracking methods rarely consider abrupt motions and easily fail to track the abrupt motion object because they are based on smooth motion assumption.To assuage these problems,visual background extractor,the background subtraction technique,is introduced to detect the object roughly.Then,the density-of-states term and Vi Be information are combined into a Wang-Landau Monte Carlo-based tracking framework,which could efficiently track the target in the whole state space within a short time.Moreover,most of the abrupt motion algorithms cannot handle the scale variation of the target,the proposed tracking algorithm applys a seperate correlation filter method which is fast and robust to grapple the challenge of the object's variation.Experimental results demonstrate that our algorithm efficiently samples and tracks the target in not only the position but also the scale variance accurately and robustly.(3)A structured correlation filter tracking algorithm is proposed.Traditional correlation filter based trackers cannot handle the appearance changes scenarios caused by severe occlusions and illumination variations or track the target regarding long-term tracking drifts.The structural correlation filter for visual tracking could takes full advantage of the advantageous information from the holistic patch and local patches of the target.A traditional correlation filter based tracking using the global information gives a coarsely prediction of the position.Then,whether an online learning scheme starts or not is decided by computing and comparing the peak-to-sidelobe ratio for each part.Moreover,an easy but successful estimated scheme is proposed,which plays a pushing role in resulting the final candidates.Both qualitative and quantitative evaluations on challenging image sequences demonstrate that the proposed algorithm performs favorably on tracking speed and robustness in complicated conditions.(4)A superpixel baed tracking algorithm is proposed.For the problem that low level feature could not represent the structure of the object,superpixel,a mid-level feature,is applied to build the appearance model.Based on the superpixel representation of video frames,the haar-like feature is extracted at the superpixel level as the local feature,and the color histogram feature is applied with the combination of background subtraction method as the frame feature.Then,local features are clustered and weighted according to the target label and the location center.Superpixel-based appearance model is measured by using the sum of the voting map,and the candidate with the highest score is selected as the tracking result.Finally,an efficient adaptive template updating scheme is introduced to obtain the robust results and improve the computational efficiency.The proposed algorithm is evaluated on challenging video sequences and experimental results demonstrate that the proposed method can get better performance on occlusion,illumination variation and transformation.