SIA OpenIR  > 光电信息技术研究室
Static human behavior classification based on LLC features and GIST features
Wang ED(王恩德); Hou XK侯续奎; Li XP李学鹏
作者部门光电信息技术研究室
会议名称7th IEEE Annual International Conference on CYBER Technology in Automation, Control, and Intelligent Systems, CYBER 2017
会议日期July 31 - August 4, 2017
会议地点Hawaii, USA
会议主办者IEEE Robotics and Automation Society
会议录名称2017 IEEE 7th Annual International Conference on CYBER Technology in Automation, Control, and Intelligent Systems, CYBER 2017
出版者IEEE
出版地New York
2017
页码651-656
收录类别EI ; CPCI(ISTP)
EI收录号20183905873613
WOS记录号WOS:000447628700119
产权排序1
ISBN号978-1-5386-0489-2
关键词Behavior Recognition Gist Llc Spm Max Pooling
摘要

This paper presents a method for recognizing human behavior in static images based on LLC and GIST features. The feature points in the image is densely located in sub-region of images and we extract SIFT feature from each sub-region. Then using the LLC method to encode the extracted dense SIFT features of each sub-region and each feature point descriptor is assigned to several nearest words, so that each descriptor can be expressed by the linear correlation coefficient of vocabulary. In order to increase the spatial information, we use the SPM model to divide the image into many blocks under different levels. We can get the pooled feature from each sub-region using the max pooling method. Then these pooled features from sub-region are concatenated and normalized as the final pooled features. In order to get GIST features, we divide the image into blocks and construct the Gabor filter group of 32 different scales and directions. Then each Gabor filter in the Gabor filter group does convolution operation with image blocks and we take the mean value of each blocks under each Gabor filter as the feature description. The features from each block under each Gabor filter are concatenated and normalized as the final GIST feature representation of the image. Then concatenating the final pooled features and the GIST features and take as the final feature representation. Finally, the intersection kernel function of SVM is used for classification. We performed experiments on several databases. By contrast, this algorithm has a good effect in accuracy and efficiency.

语种英语
引用统计
文献类型会议论文
条目标识符http://ir.sia.cn/handle/173321/22830
专题光电信息技术研究室
通讯作者Wang ED(王恩德)
作者单位Shenyang Institute of Automation, Chinese Academic of Sciences, Key Lab of Image Understanding and Computer Vision, Key Laboratory of Optical Electrical Image Processing, Shenhe District, Shenyang, Liaoning, China
推荐引用方式
GB/T 7714
Wang ED,Hou XK侯续奎,Li XP李学鹏. Static human behavior classification based on LLC features and GIST features[C]//IEEE Robotics and Automation Society. New York:IEEE,2017:651-656.
条目包含的文件 下载所有文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
Static human behavio(336KB)会议论文 开放获取CC BY-NC-SA浏览 下载
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[Wang ED(王恩德)]的文章
[Hou XK侯续奎]的文章
[Li XP李学鹏]的文章
百度学术
百度学术中相似的文章
[Wang ED(王恩德)]的文章
[Hou XK侯续奎]的文章
[Li XP李学鹏]的文章
必应学术
必应学术中相似的文章
[Wang ED(王恩德)]的文章
[Hou XK侯续奎]的文章
[Li XP李学鹏]的文章
相关权益政策
暂无数据
收藏/分享
文件名: Static human behavior classification based on LLC features and GIST features.pdf
格式: Adobe PDF
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。