Visual Target Tracking Algorithm Research Based On Extreme Learning Machine

Visual target tracking is one of the most important part of computer vision and also an important research topic in recent years.Nowdays,Visual target tracking have very important application in intelligent monitor,manless driving,military,content research,human-computer interaction.This thesis aim at the problem in target tracking,deeply research ELM,then combine ELM to target tracking,achieve to effective tracking in occlusion,deformation,illumination change and so on.The main work of this thesis is:(1)At first,introduce the original ELM.Then summarizing some of improved ELM and analyze each algorithm's drawback and advantage.At last,simple introduce sparse representation and solutions.(2)To address appearance change and partial occlusion,a novel tracking algorithm is presented via combing hierarchical extreme learning machine and adaptive structural local sparse appearance model.HELM is improve of ELM which is good at distinguish the foreground and background.A sparse auto-encoder which is based on ELM can extract robust feature in fast speed,then use online-sequence ELM to classify features.ASLSAM is a sparse representation model which exploits both partial information and spatial information of the target based on a novel alignment-pooling method.The similarity obtained by pooling across the local patches helps not only locate the target more accurately but also handle occlusion.Aim at the appearance change in target tracking,we combine subspace learning and sparse representation to update template which can reduce the influence of noise and occlusion.(3)This thesis propose a multiple kernel learning which is based on Boosting framework.Each Boosting iteration will learn a weak classifier,then combine all weak classifier to a strong classifier use to classifier features.Combining multiple kernel learning and Boosting learning can reduce the computation caused by traditional multiple kernel learning.It can ensure target tracking can keep efficient and effective in complicated scene.In order to decrease the computation and increase the classify performance,instead of using S VM we use ELM as base classifier.