校园安全平台中高层特征融合的目标跟踪技术研究  

Study on target tracking technology based on Mid-high level feature fusion in campus security platform

作　　者：刘蓝蓝 司马海峰[1] LIU Lanlan;SIMA Haifeng(School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 454000,Henan,China;Faculty of Arts and Law,Henan Polytechnic University,Jiaozuo 454000,Henan,China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000 [2]河南理工大学文法学院,河南焦作454000

出　　处：《河南理工大学学报（自然科学版）》2022年第6期163-168,共6页Journal of Henan Polytechnic University（Natural Science）

基　　金：国家自然科学基金资助项目(61602157);河南省科技攻关计划项目(202102210167);河南省高校创新团队项目(19IRTSTHN012);河南理工大学博士基金资助项目(B2016-37)。

摘　　要：视频中的目标检测与跟踪属于交叉学科研究的内容,是安全与智慧校园建设的重要监测手段。针对跟踪技术中底层特征缺乏高层知识的问题,提出一种将计算机视觉中的超像素约束融入SVM的跟踪框架。首先提取每一帧图像的超像素,计算局部平均特征与降维的密集SIFT特征,然后融合超像素约束的局部秩变换特征并通过在线学习优化分类器,最后得到优化目标跟踪结果。实验结果表明,基于超像素约束的融合特征进一步提高了目标检测的准确率和成功率,验证了算法的有效性。Video target detection and tracking is an interdisciplinary research task,it is the important monitoring method for construction of security and smart campus. Aiming at the lack of high-level knowledge of low-level features in tracking,a tracking model was proposed by integrating superpixles constraints into SVM in this paper. Firstly,superpixels of each frame were extracted. Then the local mean feature and dense SIFT feature were computed based on superpixels constraints and dimensionality reduction. The mean feature of color space,SIFT and local rank transform features were fused to optimize the classifiers through online learning. Finally,the optimized bounding boxes of tracking target were obtained. Experiments conducted on OTB2013 database demonstrated the effectiveness of the proposed algorithm.

关 键 词：校园安全 目标跟踪 超像素约束 密集SIFT 特征融合 在线学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]