基于可穿戴惯性传感器的跌倒预先识别方法

Pre-Impact Fall Detection Method Based on Wearable Inertial Sensor

跌倒是老年人意外伤亡的主要原因,使用防护产品是预防其伤亡的重要措施。为提升跌倒防护产品的适用性,提出一种结合阈值和支持向量机(SVM)多分类的跌倒预先识别方法。利用置于腰部的惯性传感器采集人体动作的加速度和角速度,并提取合加速度、水平合角速度和姿态角特征。通过设定特征阈值对样本进行初步检测,并对疑似跌倒样本提取时间窗内特征的均值、范围和方差来构建分类特征向量,通过训练的SVM多分类器对疑似跌倒样本进行复检和方向识别。结果表明:该方法对跌倒预先识别的前置时间为256ms,准确率为98.9%,可有效预先识别跌倒行为及其方向。

Falls are a major cause of death and injury in elderly people,using protective products is an important measure to prevent casualties.To improve the applicability of fall protection products,a pre-impact fall detection method combined threshold and multi-class support vector machine(SVM)is proposed.A wearable inertial sensor located in waist is used to collect 3-axis ac⁃celeration and 3-axis angular velocity of human behavior,and then resultant acceleration,horizontal resultant angular velocity and attitude angle features are extracted.Samples are preliminarily detected by threshold classifier and suspected fall samples are ex⁃tracted mean,range and variance of features by a time window to construct classification feature vector.Finally,the trained multi-class SVM classifier is used to carry out secondary inspection and direction identification of suspected fall samples.The re⁃sults show that the pre-impact fall detection method with 256 ms lead time and 98.9%accuracy,which can effectively pre-identify fall behavior and its direction.