Research On Fall Detection Method And System Based On Convolutional Neural Network

With the improvement of people’s living standards and social medical conditions,the average life expectancy is gradually lengthening,and the proportion of the elderly in the global population is also increasing.As a large country with a large population,China’s aging problem is becoming increasingly prominent and severe.When the elderly live alone at home without other people to take care of them,falling is the most common and dangerous phenomenon due to the decline of physical function and the influence of some diseases of their own.Therefore,it is of great significance to quickly,efficiently and accurately identify and judge human falling movements,which can effectively alleviate the life and health threats and social medical burden of falling to the elderly.This thesis mainly studies the fall detection method and system based on convolutional neural network.The purpose is to improve the accuracy of human fall detection through the natural advantages of convolutional neural network in image recognition,and provide a better solution for medical monitoring and care of the elderly.This thesis studies the fall detection method based on convolutional neural network and the design and implementation of alarm system based on inertial sensor human posture detection.The main research contents are as follows:(1)This thesis discusses and introduces the realistic background environment and mainstream research technology at home and abroad of fall detection research,analyzes the posture of human body,and divides it into daily behavior activities and fall actions.Through the comparative study of convolutional neural network model and other algorithms,its advantages and feasibility in fall detection are highlighted.(2)The commonly used public fall data sets such as Sisfall and Mobifall are used as the object of this study to compare the performance of the algorithm.The human posture detection module is designed and made independently.The collected experimental data set and public data set are used as the input of the network model to verify the performance of the model from multiple angles.(3)The normalized mapping of triaxial pose data to 0～255 RGB color gambit data makes the convolutional neural network perform better in the processing of fall detection.A light fall recognition algorithm based on convolutional neural network,FR-CNN algorithm,is designed.The accuracy,sensitivity and specificity indexes reached 98.48%,98.55%and 99.70%,respectively,and compared with other mainstream algorithms,the advantages of this method in fall detection and recognition are proved to be innovative.(4)For the research of human fall detection and alarm system based on inertial sensor,the circuit design and host computer design are carried out in this thesis.Through the off-line test and on-line test of the self-test data set,the performance of the system is evaluated.From the perspective of practical simulation,it is proved that the designed human fall detection and alarm system has good fall recognition performance.Finally,the results and conclusions are summarized and analyzed,and the deficiencies in the number of classification results,the coverage of data sets and hardware detection techniques are pointed out,and the future improvement direction is proposed.