Fall Detection based on Walking Gait

Abstract

Serious falling can cause fatal consequences, such as fractures, head injuries or even death, if no immediate action to look after the injured person. A fall detection system, which is able to identify the walking condition and fall status of a walking person, is vital to avoid tragedies. Most of the falling happens during walking. The aim of the project is to develop an algorithm to detect falling based on walking gait. In the project, a 3-axis-accelerometer sensor with Bluetooth capability is used. The sensor measures accelerations due to gravity in X, Y and Z planes. The walking period is used to quantify the walking gait. A short time moving average (moving window) period is computed from the data. Experimental results show the accelerations have significant changes in at least two axes when falling. Moreover, there are noticeable changes in walking period before and after falling. With the observation, a detection algorithm based on the values and changes of accelerations and walking periods is developed. To improve the accuracy of estimating the periods, cubic-spline smoothing is used to smooth the received accelerations. In the algorithm, statuses of fall, such as stand, walk, lie and sit, and falling direction are analysed.