基于 RBF 神经网络的跌倒检测算法-《计算机技术与发展》

文章信息/Info

作者:: 刘勃; 孔韦韦; 肖家钦; 王明伟; 1. 西安邮电大学研究生院,陕西西安 710121;
2. 西安邮电大学计算机学院,陕西西安 710121;
3. 八 O 二台,江西吉安 343600;
4. 陕西科技大学电子信息与人工智能学院,陕西西安 710021

Author(s):: LIU Bo1 ; KONG Wei-wei2 ; XIAO Jia-qin3 ; WANG Ming-wei4; 1. Graduate School,Xi’an University of Posts & Telecommunications,Xi’an 710121,China;
2. School of Computer Science & Technology,Xi’an University of Posts & Telecommunications,Xi’an 710121,China;
3. 802 Radio Station,Ji’an 343600,China;
4. School of Electronic Information and Artificial Intelligence,Shaanxi University of Science & Technology,Xi’an 710021,China

Keywords:: fall detection; movement characteristics; classification recognition; acceleration sensor; RBF neural network; gradientdescent method

摘要:: 为了利用便携式可穿戴设备精确监测老年人运动状态,及时识别老年人突发跌倒等意外行为,针对传统算法中阈值设计的经验性、随机性等不足,提出基于径向基函数(radial basis function,RBF) 神经网络的跌倒检测算法。通过分析研究人体日常行为和跌倒动作的运动特征,对人体日常运动状态进行分类。运用部署在人体腰部的三轴加速度传感器采集人体运动状态数据,构建关于加速度均值、标准差、极大值与极小值幅度差和极大值与极小值时间差的组合特征向量,采用梯度下降法进行 RBF 神经网络训练和优化,通过 RBF 神经网络分类器实现对日常行为和跌倒动作的识别。实验结果表明:基于 RBF 神经网络的跌倒检测算法在跌倒和非跌倒的分类识别中,准确率、灵敏度和特异性均保持了较高的水平,达到了较好的分类性能。

Abstract:: In order to make use of portable wearable devices to accurately monitor the movement status of the elderly and promptlyidentify unexpected behaviors such as sudden falls,a fall detection algorithm based on radial basis function ( RBF) neural network isproposed to solve the problems of experience and randomness of threshold design in traditional algorithms. By analyzing and studying themovement characteristics of human daily behavior and falling action,the daily movement status of human body is classified. A triaxial acceleration sensor deployed at the waist is used to collect the motion state data,and a combined feature vector about acceleration mean,variance,amplitude difference between maximum and minimum,and time difference between maximum and minimum is constructed. Thegradient descent method is used to train and optimize the RBF neural network,and the fall and daily behavior are recognized by the RBFneural network classifier. It can be found from the results that the algorithm based on RBF neural network maintains high accuracy,sensitivity and specificity in the classification of falls and non-falls,then prefect classification performance can be achieved.