实时跌倒预测保护能够显著降低老人跌倒致伤的风险，提高老人独居自理能力和身心健康水平。为了提高跌倒预测算法的识别准确率、召回率、特异度，减少跌倒判别和气囊保护系统的充气时间，设计了一种基于SVM的多级阈值跌倒预测算法及气囊保护系统，实现对跌倒行为的实时预测和保护。首先，利用佩戴在腰部的加速度传感器实现运动数据的采集；然后利用SVM算法得到分类跌倒和日常行为的合加速度、加速度、姿态角阈值，最后在单片机上预测算法进行重构，实现跌倒行为的实时预测，并根据预测结果判定是否触发气囊保护系统。实验结果表明，本文算法对跌倒的识别准确率、召回率、特异度分别为97.3%、99%和96.1%，保护气囊的平均充气时间为350.4ms，具有识别准确率高，充气时间短的优点，从而加强了该系统在实时跌倒预测与保护中的应用。

Real-time fall prediction and protection can significantly reduce the risk of fall injury and improve the self-care ability and physical and mental health of the elderly living alone. In order to improve the recognition accuracy, recall rate and specificity of the fall prediction algorithm, and reduce the fall discrimination and inflating time of the air bag protection system, a SVM-based multi-threshold fall prediction algorithm and air bag protection system were designed to realize the real-time prediction and protection of the fall behavior. Firstly, the acceleration sensor worn on the waist is used to collect motion data. Then, SVM algorithm is used to obtain the acceleration, acceleration and attitude Angle threshold of classified falls and daily behaviors. Finally, the prediction algorithm is reconstructed on the single chip microcomputer to realize the real-time prediction of falls and determine whether the air bag protection system is triggered according to the prediction results. The experimental results show that the recognition accuracy, recall rate and specificity of the proposed algorithm for falls are 97.3%, 99% and 96.1%, respectively. The average inflating time of the protective air bag is 350.4ms, which has the advantages of high recognition accuracy and short inflating time, thus strengthening the application of the system in real-time fall prediction and protection.

TP23

重庆市科卫联合项目(2020MSXM111)，中央高校基本科研业务费医工融合项 （2021CDJYGRH011），重庆市科卫联合课题面上项目（2023MSXM023）