Differentiate Between Human and Non Human Fall Using Floor Vibration and Artificial Neural Network

Abstract

Human falls is when a body loss its balance of standing and falling down where it can cause serious injury such as bone fracture, coma and death. Usually person at age 60 and above has highest death rate of falling due to weak physical body. There are various type of fall detection including wearable, non wearable device and sensor. In this research accelerometer were used to determine fell response. Various activity were chosen such as dummy falling with a weight of 26.48kg, sand bag dropping with the weight of 6 kg, ball dropping with a weight of 0.54kg dropping height of 1m, and mechanical movement. Sample data were averaged from each accelerometer and generated pattern graph of acceleration and peak acceleration. Then, three feature dataset which are maximum, mean and variance from total 385 test sample was chosen to analyze by using MATLAB software and using Artificial neural network to determine the training data prediction of performance. The testing result data prediction overall had great potential to classify human and non human fall with accuracy of 97.1%, sensitivity with 92.7% and specificity 97.1%. Therefore, Artificial neural network in MATLAB provide high accuracy of data prediction to the target data.