摘要
研究视频场景中人体行为自然语言描述的实现方法。首先建立三维人体的语义模型和主要的关节点运动模型,并建立人体运动语义描述基本数据库。应用图像自动场景标注技术来描述背景图像。通过人体简单动作的语义逻辑运算,得到人的组合动作和相互动作。将人的行为动作组合场景语义,从而准确描述出人在复杂场景的语义行为。最后建立简单的中文语法规则,得到人在场景中行为的自然语言描述。实验结果表明:与传统的二维模型相比,三维模型结合了场景语义并能解决遮挡问题,可以准确表达更为复杂的人类行为。
The implementation approach of natural language description of human body behaviour in video scene is studied in this paper. First,the 3D semantic human body model and the main joint point motion model are built,and the basic database of human body motion se-mantic description is also established.Automatic image scene annotation technology is applied to describe the background image.The combi-nation actions and mutual actions of human are derived from semantic logic operation of simple human body actions.Human behaviour motions are combined to the scene semantics,and then the human semantic behaviour in complex scenes are precisely described.Finally,the natural language description of human behaviour in scene is obtained by setting up the simple grammatical rules in Chinese.Experimental results show that the 3D model combines the scene semantics and can overcome the occlusion problem in comparison with traditional 2D model.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第2期177-181,共5页
Computer Applications and Software
基金
国家自然科学基金项目(61105020)
关键词
三维人体语义模型
图像自动标注技术
人体运动
人行为自然语言描述
3D semantic human body model
Automatic image annotation technology
Human body motion
Natural language description of human behaviour
