| 邵延华,郭永彩,高潮.基于稠密轨迹特征的红外人体行为识别[J].光电子激光,2015,26(4):758~763 |
| 基于稠密轨迹特征的红外人体行为识别 |
| Infrared human action recognition using dense trajectories-based feature |
| 投稿时间：2014-11-26 |
| DOI： |
| 中文关键词: 行为识别(HAR) 稠密轨迹(DT) 信息融合 红外 |
| 英文关键词:human action recognition (HAR) dense trajectories (DT) information fusion inf rared |
| 基金项目:教育部重点科研项目(108174)、教育部博士点基金(20130191110021)资助项目 (重庆大学 光电技术及系统教育部重点实验室,重庆 400044) |
|
| 摘要点击次数: 1997 |
| 全文下载次数: 6 |
| 中文摘要: |
| 提出了一种使用基于稠密轨迹(DT)融合特征的红外 人体行为识别(HAR)方法。主要流程 如下:1) 通过稠密采样获得输入行为视频的DT；2) 计算DT的方向 梯度直方图 (HOG)、光流直方图 (HOF)和运动边界描述子(MBH)3个描述子；3) 基于DT的HOG 、HOF和 MBH,并采取词袋库模型和一定的融合策略,构建融合特征；4) 以第3步所构建的融合特 征为k近邻分类器(k-NN)的输入,完成人体HAR。实验以IAD B红外行为库为研究对象,正确识别率达到96.7%。结 果表明,提出的特征融合及识别方法能有效地对红外人体行为进行识别。 |
| 英文摘要: |
| Human action recognition (HAR) in videos is a challenging problem in computer vision and pattern recognition with wide applications.Most works on HAR have been visible-spe ctrum oriented.However,this paper uses thermal infrared imaging for HAR.In order to overcome the deficien cy of single scale and individual representation method of action on HAR,a new recognition algorithm of human act ion using dense trajectories-based multi-feature fusion is presented.Our method consists of th e following steps:The dense trajectories (DTs) of the input action video are obtained by using dense samplin g ;Three dense trajectories-based descriptors are constructed,namely histogram of oriented gradient (HOG), histogram of optical flow (HOF) and motion boundary histograms (MBH);The fusion feature is constructed by u sing the popular bag-of-features (BoF) representation of HOG,HOF and MBH,respectively.And fr om the pattern recognition perspective,action recognition can commonly be viewed as a multiclass classific ation problem.Consequently,a k-NN classifier is employed to recognize the human acti on using the computed d ense trajectories-based fusion features.The intensive experimental results show that the proposed method can a chieve the correct recognition rate above 96.67% on the benchmark infrared action dataset of IADB.Interrelated analy ses conclude that the proposed algorithm is effective and promising for visible and infrared human action recognition. |
| 查看全文 下载PDF阅读器 |
| 关闭 |
|
|
|
|
|