Elderly Fall Risk Assessment Based on Kalman Filter and CNN-LSTM-Attention Model

Falls are common among the elderly and can lead to physical injuries, psychological fear, and social isolation, all of which severely affect their quality of life. Therefore, accurately assessing fall risk in older adults is crucial for timely preventive measures to reduce fall incidents and related injuries. This study proposes a novel method for predicting fall risk in the elderly based on Kalman filtering and binocular vision for 3D motion reconstruction combined with a CNN-LSTM-Attention model. The model includes a feature extraction layer that uses MediaPipe to identify keypoint coordinates, while Kalman filtering is applied to reduce noise caused by endpoint jitter and irrelevant movements. Based on the reconstructed 3D data, motion features related to fall risk, such as gait speed and stride length, are calculated. The CNN-LSTM-Attention model is then employed to capture spatial and temporal features, with the attention mechanism enabling the network to dynamically adjust the weight of each time step by modeling global dependencies during training. Experimental validation of the proposed evaluation system demonstrates a model accuracy of 0.831 and an F1-score of 0.813, providing a comprehensive and reliable assessment for fall prevention in the elderly. This system can be widely applied in settings such as nursing homes, home monitoring, and hospital rehabilitation departments. By accurately monitoring the motion status of the elderly and assessing their fall risk, it enables early warning and timely intervention, thereby effectively reducing the incidence of falls and enhancing the quality of life and safety of elderly individuals.