欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
研究视觉的动作识别特性,由于使用分层结构来对应行为的分层特性是一种广泛使用的视频人体行为理解方法,但对行为运动特性难以正确表达,主要困难在于行为划分的歧义问题,一般需要提供粗略层、中间层和细微层的每一个细节。在不提供各层细节的前提下,为克服行为划分的歧义行为将流形学习方法应用到行为分层的过程中,利用m-1维超平面对m维空间的二值划分性质获得行为划分的边界特征点,并使用实测数据进行了仿真。实验结果表明,使用方法获得的行为分层结构具有明确的物理含义,消除了行为划分的不确定性。
英文摘要：
Layer model is an efficient way for video action recognition.However,the major difficulty of this approach is the ambiguity of behavior layer,which generally requires designers to provide a gross layer,intermediate layer,and the fine layer of every detail.To solve these problems,the manifold learning method was applied to the behavior layer classifying process.Using the binary division nature of m-1 dimension hyper plane with m dimension space,the unsupervised action gross layer and the action distribution graph in embedding space were obtained,and an experiment with measured data was implemented.The preliminary experiment shows that this method makes each layered action a physical meaning,which helps to eliminate the ambiguity of behavior understanding.
同期刊论文项目
同项目期刊论文
期刊信息
位置：