摘要
针对老年人跌倒检测的准确性和实时性需求,该文首先建立了基于姿态角的活动描述模型,研发了集成加速度传感器、陀螺仪和蓝牙的活动感知模块,从而实时采集运动变化数据并使用蓝牙发送到智能手机。其次,选取姿态角及加速度信号向量模作为特征量,通过卡尔曼滤波对数据进行去噪与融合,并应用滑动窗口和k-NN算法实现了可实时感知老年人跌倒并报警的系统。实验证明系统在二分类场景下的跌倒检测准确率为98.9%,而敏感度和特异性分别达到98.9%和98.5%,验证了系统具有良好的实时性和较高的准确率。
According to the accurate and real-time requirement for fall detection. An activity model based on attitude angles is firstly established. A sensor board integrated with trial-axil accelerator and gyroscope is developed, which can capture the accelerations and angular velocities of human activities and transmit them to a smart phone by Bluetooth. Secondly, the three-dimensional attitude angle and acceleration signal vector magnitude are selected as features for fall detection. The collected data is preprocessed using Kalman filter to reduce noise and enhance the precision of attitude angle calculation. The k-Nearest Neighbor（k-NN） algorithm and appropriate sliding window are introduced to develop the fall detection and alert system. At last, the experimental results show that the system discriminates falls from the activities of daily living with accuracy of 98.9%, while the sensitivity and specificity are 98.9%, and 98.5% respectively. It proves that the method has favorable accuracy and reliability.
出处
《电子与信息学报》
EI
CSCD
北大核心
2017年第11期2627-2634,共8页
Journal of Electronics & Information Technology
基金
国家自然科学基金(61602016)~~
关键词
计算机应用技术
跌倒检测
数据融合
卡尔曼滤波
k-NN算法
姿态角
信号向量模
Computer application technology
Fall detection
Data fusion
Kalman filter
k-NN algorithm
Attitude angle
Signal vector magnitude
