Research On Human Behavior Recognition Method Based On Vision

With the development of artificial intelligence,human behavior recognition has become one of the important research directions in the field of robot vision.It is widely used in the fields of security monitoring system,medical diagnosis monitoring,human-computer interaction,etc.,and it has also been widely concerned by the academic circle,business circle and industry circle.However,how to enable robots to recognize different human behaviors is a prerequisite for serving human beings.In reality,different human behaviors are bound to be affected by various environmental factors,which makes it a worthy research direction for robots to accurately recognize different human behaviors.This paper focuses on the recognition of human behavior.This paper briefly introduces the research situation of artificial intelligence in human behavior recognition,application and frontier problem,respectively from the bottom of the target detection and recognition algorithm,the feature extraction of human body key point,the design of the human behavior recognition model,top goal of retrieval and the qualitative relationship of time and space of human posture,said the behavior of human conduct the thorough research to the potential said,etc.The main contents of this paper are as follows:(1)In order to improve the accuracy of object detection and recognition and human key points detection and recognition,the IoU method is used as the performance index to measure the accuracy of target detection,and the I oU method is promoted and improved,the neural network structure was redesigned,and different data sets were used as data samples for training research.In the research of human key points feature extraction,the bottom-up feature extraction method is adopted.The experiment proves that the algorithm based on the redesigned neural network structure and feature extraction can adapt to the feature extraction of human key points.Compared with other models,the method implemented in this paper not only improves the accuracy of object detection and recognition and the detection and recognition of human key points,but also improves the application ability of the algorithm.(2)The feature extraction algorithm model of human behavior recognition is redesigned.The detection box of human body is combined with the skeleton of human body,and the human behavior is divided into different categories by the algorithm of classification and recognition.The experiment proves that the model can improve the accuracy of human behavior recognition compared with other human behavior recognition models.(3)For further research on human behavior recognition,the Kinect device was used to obtain information about indoor scenes.The concave-convex segmentation algorithm is used to classify the target,and the concave-convex segmentation algorithm is improved.The feature extraction of the target is carried out.The lexical tree model is used to represent the information features,and the human body is segmented.Based on human behavior characteristics of qualitative spatial relations,the abstract become to interval graph,describe people's behavior by using the method of coding,in the end,the method of using dirichlet distribution to low level expression of human behavior,identify the different behavior of the human body,and the different behavior in the human body,calculate the different human behavior,the accuracy of the recognition of human behavior different task.This topic focuses on the recognition of human behavior,adopts the method of deep learning,and further studies the method of qualitative spatial representation without supervision,which has certain research and application value for the improvement of the accuracy of human behavior recognition and the understanding of human behavior recognition.