Human Fall Prediction and Detection Using Low Price IMU Sensor

In this project, we develop a device to warn the elderly when they are unsteady and at a risk of falling, based on a low price IMU sensor. From our preliminary study we find that during the sway, the body vibrates violently in all frequencies. This is different from the body at stationary posture, where the vibration is small, or when a person moves in a regular manner where the body vibrates with some definite frequencies. From these results, it is possible to use the mean amplitude of the Fourier spectrum of the body acceleration to distinguish the vibration state of the body, which can be used in the warning system. We then designed the device using the low price MPU6050 sensor to measure the body acceleration. This device will be attached to the body (hip) for collecting the acceleration data. An ESP8266 board is used to collect and control the sensor operation. Another ESP8266 board is used to receive and send the data to a Raspberry Pi unit. All data transmission is performed wirelessly via ESP NOW protocol, which is a low power consumption 2.4GHz wireless communication with a data package of up to 250 bytes at a time with 100-200m transmission range, which cover the living space area of typical wearer. When the mean amplitude of Fourier spectrum is greater than the warning criteria, the Raspberry Pi will send a signal to the warning unit wirelessly through the ESP8266 board. There are 3 types of alarms in warning unit which are sound alarm, light alarm, and LCD display alarm. We have tested our system with many volunteers. The mean amplitude of Fourier spectrum in the steady, walking, and sway condition can be clearly distinguished. In all cases the state of risky from falling can be detected correctly. We are now in the process of maximizing the data sampling and analyzing time to make the falling prediction as fast as possible.