闫肃 肖明波

摘 要：针对目前跌倒检测系统准确率低、应用场景单一等问题，提出一种基于特征参数变化量的跌倒检测算法。该算法通过采集合加速度幅值的实时变化量，为人体运动剧烈程度提供更直观准确的分析。由于运动过程中人体竖直方向相对于水平地面的角度随步伐呈周期性变化，算法对走路和上下楼梯行为进行准确区分，通过实验确定不同运动状态下的报警阈值，大大提高了跌倒检测准确率。系统采用Protothread模型实现并行检测，在减小系统开销条件下满足实时多任务需求。实验证明，改进后的跌倒检测算法将准确度和特异性分别维持在98.4%和99.1%，在保证实时性和便携性的同时能更有效地识别跌倒。

关键词：Mpu6050;变化量;并行检测;Arduino;跌倒检测

DOI：10. 11907/rjdk. 181753

中图分类号：TP312文献标识码：A文章编号：1672-7800（2019）001-0077-04

Abstract： Aiming at the shortcomings of the current fall detection system such as low accuracy and single application scenario， a fall detection algorithm based on the variation of feature parameters is proposed and the system parallel detection is realized. The algorithm provides a more intuitive and accurate analysis of the intensity of human motion by collecting the real-time variation of the combined acceleration amplitude. Due to the vertical direction of the human body to the horizontal ground changes periodically with the pace， the walking and going upstairs and downstairs can be accurately distinguished. Through a large number of experiments to determine the respective alarm threshold under different state of motion， the accuracy of the fall detection is greatly improved. The Protothread model is also adopted to meet the demand of real-time multitask under the condition of reducing the system overhead. The experiment shows that the improved fall detection algorithm maintains the accuracy and specificity of 98.4% and 99.1% respectively， and the system can identify the fall more effectively while ensuring the real time and portability.

0 引言

老年人及醫院内处于康复中的病人极易突发意外而跌倒[1]，而老年人骨质较疏松，若未及时得到专业救助，产生的后果无法估量。为帮助跌倒后当事人及时发出求救信息，科研人员研究了多种跌倒检测装置。

目前跌倒检测装置[2]分为基于视频图像[3]、基于环境[4]和基于可穿戴式设备3种跌倒检测系统[5]。基于视频图像系统准确度较高，但限制在有摄像头的室内使用，算法复杂度高、计算量巨大，不适合市场推广普及;基于环境系统依靠声学检测判断跌倒事件的发生，需要在检测范围内铺设种类繁多的传感器，不仅增加检测成本，还将被检测者的活动范围局限在室内某些区域，给日常生活带来很多不便。……