Design Of A Neural Network Based Fall Detection And Alert System

Ng, Yong Jie (2017) Design Of A Neural Network Based Fall Detection And Alert System. Project Report. UTeM, Melaka. (Submitted)

Text Design Of A Neural Network Based Fall Detection And Alert System - Ng Yong Jie - 24 Pages.pdf - Submitted Version Download (525kB)

Abstract

Accidental falls are considered the major cause of accidents that could lead to paralysis, accidental deaths or psychological damage. In most of the fall accidents, external support is crucial in order to prevent major injuries. Thus, a system that automatically detects fall event could help to reduce fall events and efficiently improve the prognosis of fall victims. This project proposes a neural network based fall detection and alert system with SMS alert and GPS function. A GY-80 10 Degree of Freedom (DOF) Inertial Measurement Unit (IMU) module is mounted on a wearable waist-worn device to continuously record body movements and detect body postures. The GY-80 10DOF IMU module consists of BMP085 barometer, HMC5883L magnetometer, ADXL345 accelerometer and L3G4200D gyroscope. For this project, we only use the accelerometer and gyroscope for the fall detection. The tri-axial accelerometer measures the static acceleration of gravity with high resolution (4 mg/LSB) which enables measurement of inclination changes less than 1.0°. The gyroscope is device that measure or maintains rotational motion. A new neural network algorithm has been developed to accurately distinguish falls from different postural transitions during activities of daily living (ADL) including standing, walking, jumping, running, sitting and lying. A body temperature and heart pulse monitoring device was developed for this system to assist the rescue team know the body condition of the user during the fall occurs. The application of the system is implemented on the Android platform. Once a fall accident happens, the alert system will be triggered and send emergency messages, the actual location and body conditions of the user to the recipient. Fall and ADL simulations were performed by a group of subjects to test and to validate the performance of the system. The experiment results showed that the proposed system could obtain sensitivity of 95.5%, specificity of 96.4% and accuracy of 96.3%.

Actions (login required)

View Item