Research On The Old People’s Fall Detection System Based On SVM-KNN Optimized By Grid Search Method

Since the reform and opening up, China’s rapid economic development has led to a comprehensive upgrade of medical technology. In the process of rising living standards,people’s health awareness has gradually increased. Thus, with the growth of the average life expectancy of our population, the social demographic structure is changing step by step towards aging. Falls is one of the major risks of life, the elderly fall will not only cause physical harm, but also because there is no timely treatment and lead to crisis life safely.Therefore, the design of a precise fall detection system, used in real-time detection of the elderly is falling, and the fall event to inform the family in order to facilitate the elderly in a timely manner, has important application value.In recent years, domestic and foreign research on the fall detection technology is divided into four categories: (1) based on video image fall detection; (2) based on environmental sensor fall detection; (3) based on wearing device fall detection; ) Based on the fall of the smart phone detection. As the first three methods of design and implementation of a certain degree of complexity, for the privacy of people being monitored and the daily life of inconvenience, and inconvenient in outdoor testing.Therefore, this article for the equipment to carry the convenience and accuracy of the system to consider, decided to use a smart phone as a fall data collector, and collected data sent to the server in real time, relying on the server’s data processing capabilities to achieve fall detection algorithm. This method avoids the problem of the computational power ofthe mobile phone to limit the complexity of the fall detection algorithm to a certain extent.The main work and achievements of this paper are as follows:(1) Building a human behavior data collection platform collect the nine kinds of daily behavior and four kinds of fall behavior. By analyzing the acceleration of each behavior data and the barometer, the five empirical eigenvalues of this paper are proposed.(2) On the Matlab platform, the parameters of SVM algorithm are optimized by grid search method, genetic algorithm and particle swarm algorithm respectively. According to the simulation results,select the parameter optimization algorithm suitable for this paper-grid search method.(3) According to the problem of SVM classifier optimized by grid search, the SVM algorithm is further optimized, and the KNN algorithm is introduced into the SVM classifier. Considering the multidimensional and unbalanced feature set, this paper introduces the standardized Euclidean distance to replace the traditional Euclidean distance.(4) The overall implementation framework structure of the fall detection system is proposed, and the SVM-KNN algorithm based on grid search is realized based on the smart phone and server. The test data of the 810 daily behavior data sets and 360 fall data sets which were worn on the chest, waist and buttocks were tested by 15 volunteers, and the test verification system was designed and tested. , Communication and other functions as well as front-end display.