摘要
提出了一种基于隐条件随机场的人体行为识别方法.首先,通过目标检测和目标跟踪提取图像序列中人体所在时空区域;其次,提取人体区域的Gist特征作为人体行为视觉描述子;最后,利用隐条件随机场模型对人体行为进行建模和识别.通过大规模试验证明了该方法的有效性,与其他方法的对比实验验证了该方法的优越性.
This paper proposes a hidden conditional random field-based human action recognition method. First,the consecutive spatiotemporal region of human body is extracted by object detection and tracking. Then,the Gist feature is computed for each frame within the extracted spatiotemporal region to serve as visual characteristics. At last,the Hidden conditional random field classifier is constructed to model the visual transition within one human action se-quence. Large-scale experimental results demonstrate that the proposed method can recognize human action accurately and outperform the competing methods.
出处
《天津大学学报（自然科学与工程技术版）》
EI
CAS
CSCD
北大核心
2013年第10期917-922,共6页
Journal of Tianjin University：Science and Technology
基金
国家自然科学基金资助项目(61100124
21106095
61170239
61172121)
天津市应用基础与前沿技术研究计划资助项目(10JCYBJC 25500)
天津大学自主创新基金资助项目
关键词
隐条件随机场
人体行为识别
Gist特征
hidden conditional random field
human action recognition
Gist feature
