Fall Detection Based on an Inertial Sensor and a Customized Artificial Neural Network Algorithm

With an aging population, falls have become a significant safety hazard, especially for the elderly. This paper proposes a fall detection system based on a 6-axis inertial sensor to collect body movement information and a customized artificial neural network to process the signals. Both the acceleration and the angular velocity are utilized for accuracy measurement. Instead of the typical threshold-based algorithm, a MIMO neural network is customized for fall detection. Rather than simply distinguish between falls and other activities, the system is able to recognize the non-fall behaviors, including running, sitting and walking. The whole device is implemented on a pegboard. Experiment results show that the detection specificities of these behaviors are all above 96% and the whole system accuracy reaches 96.8%. Keywords-fall detection, inertial sensor, neural network, behavior recognition.