Wearable IMU-based System for Proactive Fall Detection via Gait Pattern Recognition

For senior citizens, falls pose a serious health risk since they frequently cause serious harm or even death. Current market solutions are predominantly reactive, alerting carers only after a fall has occurred. This paper describes a new proactive prevention and detection system that makes use of a Wireless Body Area Network (WBAN) with three nodes. The system obtains a full kinematic profile of body dynamics by carefully positioning Inertial Measurement Units (IMUs) at the thoracic (T12), lumbar (L3), and foot positions. Trained on SisFall, UpFall, and manually gathered datasets, a Bidirectional Long Short-Term Memory (BiLSTM) neural network analyses sensor inputs in real-time to accomplish dual functionality: (1) fall detection with 98.2% accuracy, and (2) proactive identification of pre-fall risk states through analysis of abnormal gait patterns and centre of gravity (COG) deviations. The system alerts users before falls occur, shifting the paradigm from post-incident response to preventive intervention. Experimental results demonstrate 85% success in early risk detection with mean response time of 400ms prior to loss of balance.