| 摘 要: 随着中国人口老龄化趋势的加剧，跌倒检测成为医疗健康领域的一个研究热点。针对传统基于监督学习的跌倒检测算法中难以获取真实的老年人跌倒数据作为跌倒检测方法的基础训练数据及未考虑个性化适配问题，文章提出了一种基于扩展孤立森林的个性化跌倒检测模型，将跌倒视为一个二分类异常检测问题，通过可穿戴传感器采集大量老年人日常行为数据，经预处理和滑动窗口特征提取后，利用无监督的扩展孤立森林算法对每个老年人的日常行为数据进行单独建模，当数据不符合正常行为模式时，模型判定为跌倒。通过公开数据集SisFall验证模型效果，实验结果表明基于扩展孤立森林的个性化跌倒检测模型的平均识别准确率可达96.76%，平均敏感度和特异度分别为97.91%和94.72%，具有良好的性能表现。 |
| 关键词: 跌倒检测；扩展孤立森林；个性化；异常检测；可穿戴设备 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家重点研发计划项目“典型功能障碍患者智能康复辅具研发及应用示范”(2020YFC2005800)；“认知障碍患者辅助的安全行为监测系统与认知康复训练系统研发”(2020YFC2005802). |
|
| Research on Personalized Fall Detection based on Extended Isolation Forest |
|
XIONG Wentao, ZHENG Jianli
|
(School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China)
xwt9888@foxmail.com; zhengjianli163@163.com
|
| Abstract: With the increasing trend of China's aging population, fall detection has become a research focus in the field of public health. In traditional fall detection algorithm based on supervised learning, it is difficult to obtain the real fall data of the elderly as the basic training data of the fall detection method, and it fails to consider the personalized adaptation problem. In view of these problems, this paper proposes a personalized fall detection model based on the extended Isolated Forest, where fall is regarded as a binary classification anomaly detection problem, and a large number of daily behavior data of the elderly is collected through wearable sensors. After preprocessing and sliding window feature extraction, the daily behavior data of each elderly person is modeled separately using the unsupervised extended Isolated Forest algorithm. When the data does not conform to the normal behavior pattern, the model is determined as falling. The effect of the model is verified by the public dataset SisFall. The experimental results show that the average recognition accuracy of the personalized fall detection model based on the extended Isolated Forest is 96.76%, and the average sensitivity and specificity are 97.91% and 94.72%, respectively, with good performance. |
| Keywords: fall detection; extended isolation forest; personalization; anomaly detection; wearable device |
