Research Of Acceleration-based Human Activity Recognition Methods

It is easy to see acceleration exists in our life,and the micro electro mechanical system develops very rapid,especially the sensor,therefore,using acceleration sensor to collect the acceleration data and the recognition of human action are getting more and more attention.Human action recognition has a good application in health monitoring,fall detection and rehabilitation,etc.fro the elderly.Human action recognition also has a good application in motion monitoring,human-computer interaction,positioning and navigation,etc.for the young people.Morever,human action recognition also has important research meaning and a wide range of applications in other areas.This article introduces the research background,domestic and abroad research status of the human action recognition at first,and then introduces the main technology of the human action recognition based on acceleration,which are data collection system,singal analysis methods and classification.Although there are a lot of meaningful and valuable research about human action recognition based on acceleration sensor,it also can be improved.Therefore,this paper proposes a novel human action recognition method based on graph.This paper put forward a human action recognition framework based on data preprocessing,feature analysis,multi-graph representation learning,multi-graph embedding learning and k nearest neighbor classifier through in-depth study of the structure and properties of the acceleration signal.And in order to improve the accuracy of the human action recognition,after verifing the feasibility and effectiveness of the algorithm,this article compares the k nearest neighbor with the SVM,combines with different features and combines the data of different parts of the body in the end.This paper uses the open database SCUT-NAA,and the experimental results show that,in the case of k nearest classification,the proposed algorithm can achieve an average recognition of 83.1%,higher than PCA?LDA?LPP?SLPP?LSDA and MFA which are dimensionality reduction method and can all be unified in the framework of graph embedding.