On Human Motion Recognition Based On Acceleration Characteristics

According to statistics,about one-third of the elderly over the age of 65 will be injured every year because of the fall.If the elderly does not receive timely assistance after the fall,it will cause more serious consequences.With the popularity of smart wearable devices,the monitoring of human motion state has attracted people's attention.Most of the existing wearable devices are used to detect movement data,but they cannot effectively monitor human falling behaviors.Therefore,it is particularly important to design a high-precision fall-detection system.In this thesis,on the Android platform,the acceleration sensor is used to collect human motion information to identify the human motion pattern.On this basis,the machine learning method is adopted to carry out real-time monitoring of falls,which is helpful for the elderly to be timely detected and effectively rescued after falls.Firstly,the motion signal of human body is collected through Android phone.In order to eliminate the influence of friction and jitter on the data,the motion data is filtered and processed,and the peak detection method is adopted for step counting.Extracting 20-dimensional representative features in time-space,in order to avoid dimensional disasters and redundancy of feature information,principal component analysis and kernel-based principal component analysis are used to reduce the dimension of features and select the feature components with higher principal component respectively.Finally,through the comparison experiment between neural network and support vector machine algorithm,SVM is selected as the classifier for training to predict daily exercise and fall patterns,and the accuracy of the classifier is verified.Based on the fall detection in the motion recognition system,when a single SVM is used for prediction,misjudgment and missed judgment will occur.In response to this problem,this thesis proposes a fall detection method based on multiple support vector machine.The grid optimization and particle swarm optimization algorithm are used to optimize the two important parameters of SVM.After finding the group parameters with high accuracy,a combined classifier composed of SVMs is built to comprehensively evaluate the predicted output and reduce the harm to the health of the elderly due to missed judgment.Finally,the MATLAB Libsvm toolbox is used as the experimental simulation platform to verify the proposed method.Finally,simulation experiments are carried out on 1000 sets of experimental data.The experimental results show that under the condition of ensuring that the accuracy rate does not change greatly,according to the determined "safety domain",the missed rate of the fall sample can be reduced to 0.8%.The application of fall detection in human motion recognition has practical significance.