基于KPCA和结构化支持向量机的视频目标跟踪  

Video object tracking based on KPCA and structured support vector machine

作　　者：龙君芳 马琳娟 李庆珍[3] Long Junfang;Ma Linjuan;Li Qingzhen(School of Data Science and Computer Science,Guangdong Peizheng University,Guangzhou 510000,China;School of Computer Science,Beijing Institute of Technology,Beijing 100081,China;Institute of Data Rule of Law,China University of Political Science and Law,Beijing 102249,China)

机构地区：[1]广东培正学院数据科学与计算机学院,广东广州510000 [2]北京理工大学计算机学院,北京100081 [3]中国政法大学数据法治研究院,北京102249

出　　处：《南京理工大学学报》2023年第5期671-677,共7页Journal of Nanjing University of Science and Technology

基　　金：广东省普通高校特色创新项目(2021KTSCX159;2019KTSCX250);广东省重点建设学科科研能力提升项目(2022ZDJS133);中国高校产学研创新基金资助课题(2021FNA04013)。

摘　　要：为了提高视频目标跟踪性能,采用结构化支持向量机用于视频目标跟踪,并借助核主成分分析用于视频目标特征降维及去冗余处,以增强视频目标分类适应度。首先,提取视频目标特征,经过核主成分分析映射至更利于目标分类的特征向量。接着,建立基于结构化支持向量机的视频目标判别分类模型,从而充分挖掘目标内数据特征的紧密性,有效提高了目标跟踪的精准率。最后,针对结构化支持向量机进行关键参数求解,并获得稳定的视频目标跟踪分类结果。根据分类结果判定对应的目标以完成视频目标跟踪。试验结果表明,对于4类公共视频数据集,对比其他3种视频跟踪算法,所提方法在跟踪精准率和跟踪速率等4个关键性能指标方面具有明显优势,在应对大规模视频目标跟踪时具有较强的适应度。In order to improve the performance of video target tracking,structured support vector machine(SVM)is used for video target tracking,and kernel principal component analysis(KPCA)is used for dimensionality reduction and redundancy removal of video target features to enhance the adaptability of video target classification.Firstly,the video target features are extracted and mapped to feature vectors which are more conducive to target classification through kernel principal component analysis.Then,a video target discrimination and classification model based on structured support vector machine is established,so as to fully mine the compactness of data features in the target and effectively improve the accuracy of target tracking.Finally,the key parameters of structured support vector machine are solved,and the stable video target tracking and classification results are obtained.According to the classification result,the corresponding target is determined to complete video target tracking.Experimental results show that,for four kinds of public video data sets,compared with the other three video tracking algorithms,the proposed method has obvious advantages in four key performance indicators,such as tracking accuracy and tracking speed,and has strong adaptability in dealing with large-scale video target tracking.

关 键 词：视频目标跟踪 结构化支持向量机 核主成分分析 精准率 跟踪速率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]