Research On Human Activity Recognition And The System Development Based On SVM

With the development of modern science&technology and the great improvement of life quality,the human activity recognition based on sensor technology highlights the broad application prospects in sports medical treatment,disaster relief rescue,virtual reality,human-computer interaction and so on,and attracts large number of researchers' favor.In this paper,the three-axis acceleration and gyroscope sensor are used to recognize the motion state,which overcomes the privacy protection problem based on computer video,and avoids the influence of low precision of motion state recognition based on single sensor,and is easyto transplant and popularize.The main researches of this paper are as follows:1)This paper focused on the comparative analysis of SVM,KNN and RVM algorithms in human activity recognition,and introduced the svm+and[2-svm+algorithms used in the field of image recognition.Considering the real-time requirements of the human activity recognition system in the online phase,we compared and analyzed the process running time fully.By comparing experiments on six kinds of behavior,such as walking,standing,lying,upstairs,downstairs and trotting,the recognition rate could be obtained 99%by SVM algorithm,and the running time met the requirements of system design.Experiment showed that the recognition of human motion state based on SVM was accurate and effective.2)The feature engineering was used to extract the features from raw data,and evaluating the quality of feature set,including the study of data windowing technology and verification of classification accuracy and efficiency.We used median filter for data denoising and smoothing,and wavelet transform technology to separate the gravitational acceleration and gravity,time domain and frequency domain characteristic calculation,obtaining the optimal features set.3)Design and implementation an online real-time recognition system.Based on the above researches on the human activity recognition technology,we designed and implement a complete online real-time human recognition system based on SVM through Java.The system online recognition accuracy could reach 99%,and the function and performance met the requirements of this target..It could also be used on the following areas such as fall detection and scheduling reminders according to the different of training data samples.After the testing of the system,the research of human activity recognition system based on SVM provided important reference value for relevant researchers and software developers.