Research On Real-time Recognition Technology Of Human Movement

As a significant biological feature of the human body,human motion features can convey a lot of information about the human body,and have been well applied in various industries in society.It is a research hotspot in the field of computer vision.However,in the commonly used human motion recognition,two-dimensional motion information is often used,which leads to the error of motion recognition due to the diversity of perspectives.It is easy to be interfered by the environment and background when extracting the 3D features of the action,which leads to the inaccuracy of the action recognition.The diversity and dynamic characteristics of human movement itself also lead to difficulties in the study of movement recognition.In order to solve the above problems,this paper first proposed to use the human motion capture device Kinect to obtain the three-dimensional motion characteristics of the human body.Based on the motion features obtained by Kinect,two kinds of human motion recognition schemes were designed:Real-time human motion recognition based on 3D space features and human motion recognition based on STN-Res Net.Specific research contents are as follows:Firstly,for complex scenarios that exist in the shade,incomplete body posture,thehuman body under the action of easy registration form of movement characteristic,putforward the improvement of the STN-Res Net network model,to join the STN model as attention mechanism of human action recognition,reduce the weight of the secondary action features,more pay attention to the movement characteristic of weighting,to reduce the time complexity,and finally use Softmax,based on the classification of the multiple classification function to human body action,in real time under the premise of guarantee human action recognition,good recognition rate.Second,the real-time recognition of human 3D motion is studied in this paper.Comprehensive utilization of the purpose of this paper is to depth camera to access the body capture hardware equipment to extract human body skeleton data,using linear function normalization and standard deviation of the normalized method,based on spatial features of the Euclidean distance and spatial location of hamming distance characteristics,the space Angle cosine of the Angle characteristics such as theory,the real-time dynamic and real-time actions match as a standard action.It can meet the real-time requirements of posture matching and dynamically match with real-time characters to meet the requirements of action recognition.This article research results using low-cost equipment to 3d human motion recognition,adopt the method of normalization,improvement action scoring method and method of Res Net network at the same time to join the STN model solution space complexity of human action recognition,time difference and the accuracy of the 3d data obtained from three aspects,in order to achieve rapid,efficient,accurate action provide new real-time identification method.