Tracking Algorithm Based On The Struck

Target tracking is an important step in computer vision,computer image processing,in the broad fields of study,learning target tracking is one important branch of study,the target tracking for the image processing of two layers of semantic analysis provides a wealth of information.But in the target tracking algorithm,due to the nonrigid structure by tracking model,namely the irregular deformation target,scene illumination,target transformation,separation of foreground and background in the traditional algorithm,occlusion and target,is a great challenge for the target tracking,target tracking has become a need comprehensive ability strong problem.However,the information provided by the target tracking provides important information for the subsequent application and the accuracy of the application itself.Since the target detection module classic simple,only for a feature vector of the target model,image film,matching image films,such as Cameshift Mean-shift(mean shift),and robust tracking algorithm,only for emphasis on the characteristics of the target,but for complex scenes,target frequent transformation.Even the light intensity changes,will cause irreversible mistakes and judgment of this kind of algorithm.For this article:(1)For strong tracking performance of simple tracking module,discussed machine learning based target tracking algorithm,which is the process of target tracking,into continuous online learning,classification,add,or delete the pre target model to target the transformation of the queue,and the target tracking and online learning are highly combined,forming structured output target function,will no longer use the tracking module,learning module,detection module,real-time online learning,the branch target transform for real-time monitoring,target movement changes,or scene changes with real-time learning.(2)The paper is based on the machine algorithm of support vector machine,which isa method of using high dimensional kernel machine learning algorithm,for the past two according to the classification is to determine whether the target tracking algorithm is proposed a multi target tracking algorithm based on structured output target classification SVM by maximizing the confidence function,change the corresponding target do not transform,based on the two classification,but multi classification,but also for the number of boundary can be classified,can be dynamically set.As long as the target can be extremely close to a transform boundary through the algorithm,you can judge for a change in the target model,and add the classified learning dynamic set the transformation in the queue number.(3)For the target tracking algorithm due to severe occlusion or out of the scene caused by the target drift,but also for multi classification algorithm due to its high dimensional function transformation,high computational multi classification structured output function of the complex degree caused by the low tracking performance,we proposed a tracking algorithm of structured multi classification SVM and Hausdorff distance based on the target.The canny feature extraction operator between adjacent frames,calculate the target contour feature points Hausdorff distance image sequence tracking integration of adjacent frames,and the study of sample collection anticipation,to avoid unnecessary samples in the traditional algorithm of online learning.Secondly,by using the structured multi class SVM target output prediction function to increase the target transformation,it greatly enhances the robustness and accuracy of the target tracking.The experimental results show that the proposed algorithm not only extends the good generalization ability of SVM,but also can effectively track the transformation of the target.