LifeSenior – A Health Monitoring IoT System Based on Deep Learning Architecture

This paper proposes an efficient and reliable elderly health monitoring system based on a low power IoT communication service inside a watch type wearable device. The watch senses motion (accelerometer, gyroscope, and magnetometer) and vital signs (heart rate variability, oxygen saturation, breathing rate, and blood volume pressure) to detect falls and other possible risk situations estimated by the EAEWS (Elderly Adopted Early Warning Scores) algorithm. Sense data collected are continuously fed into an embedded bi-LSTM (bidirectional Long Short-Term Memory) deep-learning neural network that bases the LifeSenior AI (Artificial Intelligence) health monitoring system. As there are no databases with motion and vital signs collected in the same environment, we design the LifeSenior Database Project (LDP); a motion-vital signs correlated database explicitly developed to the neural network training phase. Our experimental results in a simulated environment show that this architecture presents a 84,63% of accuracy in fall situations detection and can keep the user alert about his health.