Implementation of a Fall Prediction System Using Neural Network Algorithms on a Microcontroller

Based on the needs of hospital rehabilitation centers, this study focuses on posture detection for patients undergoing rehabilitation in hospitals, especially the elderly. The goal is to ensure that nearby medical staff receive alerts as soon as possible when a potential fall is detected, allowing them to intervene before an incident occurs. This paper proposes a fall prediction system that uses sensor fusion and a Feedforward Neural Network (FNN). The system consists of a Bluetooth-enabled microcontroller unit (MCU) integrated with a 6-axis Inertial Measurement Unit (IMU), which is placed on the patient's collar. In this study, the potential movements performed by patients during rehabilitation are classified into nine categories: standing, stomping, stair climbing, stretching, waist twisting, bending over, squatting, walking, and running. Motion data from the IMU is processed using an FNN and k-means clustering for movement classification and abnormal motion detection. If a potential fall is detected, an alarm is sent via Bluetooth to notify nearby medical staff. Finally, the proposed fall prediction system will be implemented on a Bluetooth-enabled MCU module.