Pre-Impact Fall Detection with Wearable Sensors Using Deep-learning Algorithms

Falls among elderly individuals have become a significant public health concern due to the associated injuries and health implications. This thesis presents a comprehensive study on pre-impact fall detection using wearable glasses equipped with accelerometer, gyroscope, and magnetometer sensors in three dimensions. Data collection involved 11 subjects who participated in simulated fall scenarios and various non-fall activities. The wearable glasses captured signals from three sensor types, enabling the recording of precise motion data during these activities. Moreover, the study included video recordings of all activities, which were utilized for fall detection event labeling through computer vision methods. The proposed approach leverages machine learning and deep learning algorithms to detect falls as early as possible, thus reducing the response time for assistance.