Research On Human Action Recognition Based On Adaptive Center

Human action recognition based on 3D skeleton has been a popular topic in computer vision,the goal of which is to automatically segment,capture,and recognize human action,it has been used in video surveillance,health care,intelligent robot and virtual reality,and has been widely explored by researchers since the 1960 s.Although there are many ways to obtain 3D data in the study of behavior recognition,it is difficult to obtain data information in real environment,such as it is difficult to solve the problems of difficult foreground and background division,image color and texture similarity,light,angle,occlusion,clutter,or diversity of data sets.The rapid development of depth sensors,such as Microsoft Kinect sensor,in recent years it has provided not only color image data but also 3D depth image information,so it has promoted the rapid development of the research based on the 3D skeleton algorithm.But the present action recognition algorithm selects a fixed joint as the coordinate center,which results in the accuracy of the recognition method usually not very ideal.This algorithm aims at the problem that only one fixed coordinate center is selected in the current motion recognition algorithm,leading to the problem that the recognition rate is not ideal.This paper proposes an adaptive bone center human behavior recognition study and improves the motion recognition rate.In the algorithm,frames of skeleton action sequences are loaded onto a human action dataset,redundant frames are removed from the sequence frame information,and the original coordinate matrix is obtained by preprocessing the sequences.Then,the original coordinate matrix is mapped to the Lie group space by relative position relation.Rigid vector and joint angle features are generated by extracting the original coordinate matrix.The adaptive value can be determined on the basis of changes in rigid vector and joint angle values,feature extraction is done by extracting the original coordinate matrix to generate rigid body vectors and joint angle eigenvalues,namely the angular velocity and angular acceleration of rigid joint angles.In this algorithm,20 human joints are reduced to 16 when dealing with the eigenvalues,so that the feature dimension of the rigid joint angle is less.The coordinate center can be adaptively selected according to the adaptive value and used to normalize the original matrix.The action coordinate matrix is denoised by using a dynamic time warping method.The Fourier time pyramid method is used to reduce the time displacement and noise problems of the action coordinate matrix.The matrix is classified by using support vector machine.Unlike existing algorithms,the proposed algorithm exhibits improved performances on MSR-Action3 D,UTKinect-Action and Florence3D-Action datasets.On the MSR-Action3 D dataset,the action recognition rate of the proposed algorithm is higher than the HO3 DJ algorithm,the profile HMM algorithm and the Eigenjoints algorithm.On the UTKinect-Action dataset,the action recognition rate of the proposed algorithm is higher than the HO3 DJ algorithm and the CRF algorithm.On the Florence3D-Action dataset,the recognition rate of the algorithm is higher than Multi-part Bag-of Poses algorithm,Motion trajectories algorithm and Elastic Function Coding algorithm.The proposed algorithm solves the low accuracy problem of the existing action recognition algorithm that the coordinate center of a fixed joint is adopted.The experimental results show that this algorithm can effectively improve the accuracy of action recognition,the recognition rate is higher than the existing algorithms of human action recognition.The main contributions of this paper are as follows:First,the algorithm maps the existing 3D data sequence of the human skeleton from the European space to the Lie group space SE(3)to reduce the error caused by direct transformation of the data in the Euclidean space when the skeleton data is preprocessed.Second,the adaptive eigenvalue of bone center is constructed when the algorithm is extracted from the human skeleton data,and a method of adaptive analysis and understanding of motion is proposed under the motion state analysis.