Research On An Action Recognition Algorithm By Analyze Hidden Semantic

With the development of computer technology,image recognition technology has been paid more and more attention by academia.Especially the technology of human activity recognition,has become one of the most popular fields in academia.How to accurately and efficiently recognize activities has become a challenging problem.This paper is based on three-dimensional data of human skeleton points that collected and extracted from Kinect devices,researches on activity representation and activity recognition,which are the two main steps of human activity recognition.Firstly,this paper researches on the activity representation algorithm of human activity recognition,introduce several traditional activity representation algorithms,and analyze their advantages and disadvantages.Traditional representation algorithms only focus on the main features of activity data,but ignore the details of human activity.This paper proposes a human activity representation algorithm,which fuses the main semantic and the hidden semantic of human activity.This paper use skeleton point sequences as main semantic,at the same time extracts and process the variety of human body parts displacement on the time-based characteristic,the variety of human body parts interval on the space-based characteristic and the acceleration of human body parts on the time-based and space-based characteristic as hidden semantic.Finally,use the fusion of main semantic and hidden semantic as human activity representation.Secondly,this paper researches on the activity recognition algorithm of human activity recognition,introduces several traditional activity recognition algorithms,and analyze their advantages and disadvantages.Traditional activity recognition algorithms use principal component analysis or kernel principal component analysis and support vector machine as the main data processing methods.When theses algorithms are applied in human activity recognition,it's very easy to lose the details of human activity data,and reduces the accuracy of recognition rate.This paper proposes a human activity recognition algorithm,which is the fusion of two dimensionality class mean kernel principal component analysis and support vector machine,and improve the algorithm in many aspects,which makes the algorithm more suitable for human activity recognition.This algorithm can maximally protect the details of human activity during the process,and improve the accuracy of human activity recognition.Finally,this paper use three kinds of data sets to do the experiment,then compares the accuracy with different kinds of activity representation algorithms,activity recognition algorithms and excellent algorithms in academia circles.The experimental result shows that the proposed activity representation algorithm and activity recognition algorithm can effectively improve the accuracy of human activity recognition.