Research On Fall Recognition Based On Machine Learning

With the aggravation of population aging,people pay more and more attention to the life safety of the elderly.Fall has brought serious threat to the life safety of the elderly.The relevant investigation and research show that fall is the primary factor causing accidental injury of the elderly.If no one discovers the fall of the elderly in time,the best time for treatment is often missed,resulting in irreversible and serious consequences.Therefore,it is of great practical significance to study on the fall recognition method and effectively identify the abnormal fall of the elderly.In this paper,by studying the fall recognition method,the machine learning theory method is applied in the field of fall recognition to accurately and effectively identify the abnormal falls that may occur in the elderly,minimize the serious consequences caused by falls in the elderly,and effectively reduce the occurrence of falls in the elderly that no one has discovered.The main contents of this paper include the following aspects:(1)Due to the recognition of the relative lack of experimental data set,fall so in this paper,using self-designed wearable equipment acquisition of the normal activity of the human body and human body falls acceleration,angular velocity and attitude Angle data,the analysis and get data for making experimental data set,to fall identification methods experiment and fall to prepare identification system implementation.(2)In order to enhance the fall recognition effect of correlation vector machine,BA-RVM method is proposed in this paper.For extracting effective characteristics of experimental data centralized data first,then the relevance vector machine the goals of the kernel function selection through likelihood estimation method to improve the convergence condition of nuclear wide,through the bat algorithm for kernel function of relevance vector machine broad global optimal nuclear small and get the optimal target using the kernel function model,experimental data set down recognition accuracy is 98.2%.(3)In this paper,ELM neural network method is proposed for fall recognition.ELM neural network method in the process of training does not need to adjust the weights of input and hidden layer nodes bias,the output layer weights can be implied by calculation the output matrix,possesses the advantages of fast learning speed and generalization ability is strong,fall recognition effect is better than the common machine learning fall identification method,the experimental data set down recognition accuracy is 99.4%.(4)The fall recognition method in this paper is applied to the system to design and implement the fall recognition system.The design of the wearable device is completed,and the functions of data acquisition and wireless data transmission are realized.The server side can receive the data uploaded by the wearable device and recognize the data through the deployed machine learning algorithm model to realize the function of fall recognition.