| dc.description.abstract | As society faces the phenomenon of aging populations, the proportion of elderly individuals gradually increases, with frailty emerging as a significant factor affecting their health. Frailty is characterized by the body becoming vulnerable in response to stressful events, a result of long-term decline in multiple physiological systems. Although the Clinical Frailty Scale (CFS) is a commonly used assessment tool, it is susceptible to subjective factors and lengthy measurement times. Therefore, this study utilized a 3D camera, Kinect, to collect gait skeleton data, replacing traditional assessment scales and consolidating the top five levels of CFS into four levels. Recognizing the potential discontinuation of Kinect, the research team converted collected data into Nuitrack skeleton detection points compatible with RealSense D435. Leveraging machine learning, three-dimensional skeleton features were extracted, achieving an impressive classification accuracy of 97%. Additionally, the study trained a Long Short-Term Memory (LSTM) classifier with skeleton information, achieving a 77% accuracy rate. Ultimately, by combining skeleton-derived features with time-motion characteristics identified by the LSTM classifier, the classification accuracy increased to 100%. Furthermore, experiments confirmed that depth cameras with Nuitrack skeleton points can be used for gait testing and prediction, showing potential as substitutes for the Kinect depth camera. | en_US |