Abstract: For human behavior recognition in monocular video, a method for recognizing human behavior based on action subspace and weighted condition random field is presented in this paper. This method combines kernel principal component analysis (KPCA) based on feature extraction and weighted conditional random field (WCRF) based on activity modeling. Silhouette data of human is represented more compactly by nonlinear dimensionality reduction that explores the basic structure of action space and preserves explicit temporal orders in the course of projection trajectories of motions. Temporal sequences are modeled in WCRF by using multiple interacting ways, thus increasing joint accuracy by information sharing, and this model has superiority over generative ones (e.g., relaxing independence assumption between observations and the ability to effectively incorporate both overlapping features and long-range dependencies). The experimental results show that the proposed behavior recognition method can not only accurately recognize human activities with temporal, external and internal person variations, but also considerably robust to noise and other factors.