| 基于组合特征和SVM的视频中人体行为识别算法 |
| |
| 引用本文： | 陈艳,胡荣,李升健,万彬,孙书会.基于组合特征和SVM的视频中人体行为识别算法[J].沈阳工业大学学报,2005,42(6):665-669. |
| |
| 作者姓名： | 陈艳 胡荣 李升健 万彬 孙书会 |
| |
| 作者单位： | 1. 江西科技师范大学 通信与电子学院, 南昌 330036; 2. 南昌大学 科学技术学院, 南昌 330029; 3. 国网江西省电力有限公司 电力科学研究院, 南昌 330096; 4. 南昌职业学院 工程系, 南昌 330004; 5. 沈阳工业大学 软件学院, 沈阳 110870 |
| |
| 基金项目： | 国家自然科学基金项目(61563034);江西省教育厅科学技术研究项目(GJJ151504,GJJ151505,GJJ151497) |
| |
| 摘 要： | 针对复杂场景中的人体行为识别困难的问题,提出了一种基于组合特征和SVM的行为识别算法.该算法使用光流特征、HOG特征、重心特征和3D SIFT特征构成的组合特征来描述人体的各种行为;使用一对一的方式训练SVM分类器对提取出的特征进行分类,并以投票的方式得到具体的行为类别.使用包含4个场景的KTH数据集进行仿真.结果表明,所提出的算法能适应各种复杂环境,且相比只采用单一特征的识别算法具有更高的分类精度.
|
| 关 键 词： | 行为识别 光流 方向梯度直方图 重心 3D SIFT特征 支持向量机 KTH数据集 行为分类 |
Recognition algorithm for human behavior in video based on combined features and SVM |
| |
| CHEN Yan,HU Rong,LI Sheng-jian,WAN Bin,SUN Shu-hui.Recognition algorithm for human behavior in video based on combined features and SVM[J].Journal of Shenyang University of Technology,2005,42(6):665-669. |
| |
| Authors: | CHEN Yan HU Rong LI Sheng-jian WAN Bin SUN Shu-hui |
| |
| Affiliation: | 1. School of Communication and Electronics, Jiangxi Science and Technology Normal University, Nanchang 330036, China; 2. School of Science and Technology, Nanchang University, Nanchang 330029, China; 3. Electric Science Research Institute, States Grid Jiangxi Electric Power Limited Company, Nanchang 330096, China; 4. Engineering Department, Vocational College of Nanchang, Nanchang 330004, China; 5. School of Software, Shenyang University of Technology, Shenyang 110870, China |
| |
| Abstract: | Aiming at the difficulty in human behavior recognition in complex scenarios, a behavior recognition algorithm based on combined features and SVM was proposed. With the combined features composed of optical flow features, HOG features, gravity center features and 3D SIFT features, various human behaviors were described by the as-proposed algorithm. The SVM classifier was trained in a one-to-one manner to classify the extracted features, and the specific behavior category was obtained by votes. The KTH data set containing 4 scenarios was simulated. The results show that the as-proposed algorithm can adapt to various complex environments, and has higher classification accuracy than those algorithms just employing a single feature. |
| |
| Keywords: | behavior recognition optical flow histogram of orientational gradient gravity center 3D SIFT feature support vector machine(SVM) KTH data set behavior classification |
|
| 点击此处可从《沈阳工业大学学报》浏览原始摘要信息 |
|
点击此处可从《沈阳工业大学学报》下载全文 |
|
