Detection Of Human Abnormal Interaction In Video Scene

The purpose of this study is to identify the human interaction behavior of the two people involved in the video scene and to accurately detect the abnormal behavior of the human body in the video scene.This paper focuses on the extraction of the salient features of human interaction in the video scene,and the analysis and detection of abnormal interaction behavior using Bayesian hierarchical networks.Through research and experiment,this paper presents a hierarchical recognition framework for human interaction behavior recognition and abnormal behavior detection in video scene.Experiments show that the algorithm framework of this paper realizes the detection of abnormal behaviors of human body in video scene,and improves the accuracy and sensitivity of the algorithm in some degree compared with other former algorithms.The specific research contents are as follows:(1)Human body segmentation algorithm based on hierarchical fusion.Since the state changes in the body part in the interaction behavior have a significant effect on the behavioral abnormality,it is necessary to process the lower-level image elements into the upper body part before performing the gesture recognition.Therefore,this paper proposes a hierarchical fusion and segmentation framework for the two-person interactions.Using the foreground extraction algorithm,the whole human foreground is extracted.And then three levels of fusion are performed including pixel level,image block level and object level.Through the integration of all levels,human body will eventually splited into three parts including the head,upper body,lower body.(2)Aiming at the posture of human body part in interactive behavior,this paper proposes an algorithm of human pose recognition based on hierarchical frame.Firstly,the human body part is modeled and extracted by ellipse and convex polygon.Then,by using the hierarchical Bayesian network,the pose estimation sequences are generated for each part of the human body.Then the pose of each part of the human body is combined to form the whole body pose,which bridge between the underlying model and the video and images semantics.(3)To study the recognition method of human interaction behavior,a new algorithm based on Hidden Markov Model(HMM)is proposed to detect the abnormal behavior of human body.First,after acquiring the pose estimation sequence,we use Hidden Markov Model to train the observation sequence with time information.Then input the test video set to verify that it is possible to generate a description of the human interaction behavior.Finally,the classification of abnormal interaction behavior and statistical generation results are peformed.After setting the evaluation index,the experimental results of this paper are compared with other schemes to verify the effectiveness of the layered identification framework algorithm proposed in this paper.