Research On Human Behavior Recognition Method Based On Action Three Views

Human action recognition is the most challenging research direction in the field of computer vision research.It is the current research hotspot.Human action recognition based on Kinect bone data brings a new way of human-computer interaction,in intelligent monitoring,virtual reality,human-computer intelligent interaction and other aspects have broad application prospects.How to use the computer to realize human action recognition efficiently has become a topic worth studying.The two most critical steps in human action recognition are feature extraction and classifier construction.The traditional human action recognition method based on Kinect bone data is mostly based on action data in three-dimensional space,and less take into account the role of the three views of action in action recognition,and take into account the hierarchical problems inherent in the action and time series relationships of actions,the method of combining the action front view with the long short term memory network and the method of combining the three views of the action with the layered framework are proposed.In the third chapter,this paper combines the front view of the action with the long short term memory model to achieve the classification of the action.Aiming at the problem that the traditional manual selection of data features is insufficient,the data is input into the long short term memory network model,and the parameters are adjusted.The recognition results on the public dataset show that the method has better recognition performance.In the fourth chapter,the three-view and the layered framework are combined.In the first layer,the mean,variance,full-range and skewness of the five-point center of gravity of the human body are extracted.Test with the support vector machine in three views of the action,and then use the voting mechanism to determine the category of the first layer.In the second layer,the relative position and velocity of the five parts of the human body are extracted in the local coordinate system and the global coordinate system respectively.Hidden markov model is used to subdivide each major class,and the public data set MSRAction3 D is respectively used in the user independent and user dependent experimental verification methods.The experimental algorithm has goodrecognition results on both public data sets and self-collected data sets.