Pervasive human fall detection and alert system based on multilayer perceptron neural network

• Main Author: Kerdegari, Hamideh
• Format: Thesis
• Language: English
• Published: 2012
• Online Access: http://psasir.upm.edu.my/id/eprint/38563/1/FK%202012%2052R.pdf; http://psasir.upm.edu.my/id/eprint/38563/

Falls are major health problems which affect the human life by restricting their movements and independency. With the increase in the human population, more attentions are required in order to prevent the serious effects and problems caused by fall. A system which can identify the occurrence of falls in almost every situation and alert the care center is a helpful solution to care for the human safety. The motivation behind this work is to develop a pervasive system for monitoring the human activities and identifying the occurrence of falls. In this work, a waist worn fall detection system has been developed. A tri-axial accelerometer (ADXL345) was used to capture the movement signals of the human body and detect events such as walking and falling to a reasonable degree of accuracy. A set of laboratorybased falls and activities of daily living (ADL) were performed by healthy volunteers with different physical characteristics while the sensor was attached to their waists. Personal information features which are the volunteers’ personal physical features and acceleration features which are taken from acceleration data were considered as feature set. Decision tree method was used to find out the effective features for classification. In order to classify the collected falls and ADL acceleration patterns,Multilayer Perceptron (MLP) neural network was applied for precise classification of motions and determination of fall events and ADL. The results showed that MLP can detect the falls with accuracy of 91.6 %. Finally, a pervasive fall detection system was developed as a smart phone-based application under the name of Smart Fall Detectionc (SFD). SFD is a standalone Android-based application that works using smart phone resources such as accelerometer sensor and GPS which utilizes proposed trained MLP for fall detection. When SFD detects the fall, a help request will be sent to the specified emergency contact using SMS and subsequently whenever GPS data is available, the exact location of fall will be sent. The SFD performance showed that it can detect the falls with accuracy of 91.25 %. This work resulted in the most accurate, first and only smart phone-based fall detection application which uses MLP neural network to determine the occurrence of fall.