Intelligent Human Action Recognition Algorithm Based On Support Vector Machine

Human action recognition is an important research topic in the field of human computer interaction.It has broad application prospects in smart home,virtual reality,video game and so on.Especially in the smart home scene,human motion recognition has attracted more and more attention.However,the current recognition algorithms do not meet the practical requirements in the accuracy,real-time and robustness of human motion recognition,especially for the real-time recognition of complex human movements.There are many problems to be solved in motion capture and recognition.In order to solve the difference between different height and size of users in human-computer interaction of human action recognition,based on the existing framework,this paper makes a deep research on human body modeling,feature extraction and recognition algorithm,proposed an action recognition algorithm based on machine learning algorithm,design a complete human action recognition process to solve this problem.Firstly establish the 3D human body model based on Kinect,The feature data of the motion data are constructed by the feature selection and feature extraction based on the angle and modal ratio of the motion vector.The validity of the feature is verified from two aspects: distinguishing and clustering.Secondly this paper proposes a real-time recognition algorithm based on Support Vector Machine(SVM)and Hidden Markov Model(HMM)and an initial recognition algorithm based on multi-class SVM.It can make accurate recognition of most data based on the difference of motion data structure.By introducing the penalty factor and kernel parameter for the outliers can achieve better recognition results.Optimized motion recognition algorithm based on HMM probabilistic constraint relations between misidentified data and real tags.The accuracy of the parameters determined by the grid search algorithm is the highest.Finally,this paper builds an interactive simulation system for intelligent home,and establishes a local database with 13 categories and 1300 actions.The local data is used for real-time action algorithm training,testing.The experimental results show that the proposed algorithm can improve the accuracy and robustness of complex motion recognition.The final action accuracy rate of 95.6%,while meeting the requirements of real-time identification.