Video abnormal target description based on CRF model

This paper presents a method for video abnormal target description based on Conditional Random Fields (CRF) model. CRF offer several advantages over Markov Random Fields (MRF), including the ability to use contextual information, and CRF allows us to relax the assumption of conditional independence of the observed data often used in generative approaches, an assumption that might be too restrictive for a considerable number of object classes. Feature vectors of target as well as context information are extracted. These feature vectors modeled by CRF. Parameter of model is estimated through train and description abnormal object through inference. The experiment results show that the accuracy rate is 91.7%. To improve the efficiency of method we optimize method by parallel design.