基于极限学习快速分类方法的人体跌倒检测方法

为了准确快速检测人体跌倒状态,提出基于惯性测量单元(inertial measurement unit,IMU)测量和处理数据的极限学习机(extreme learning machine,ELM)快速分类判别方法。分析了人体运动行为特征,构建了腿部运动参数提取模型;通过IMU采集人体腿部运动特征数据,并进行姿态解算;采用ELM方法对人体运动特征的加速度、角速度和姿态进行分类,判断人体是否处于跌倒状态;根据机器学习评价指标对ELM参数进行优化,得到最佳参数。进行了人体运动状态测量实验,结果表明,ELM方法能够对IMU测量和处理数据进行准确快速地分类。当隐含层结点为1 000时,ELM检测方法跌倒检测的准确率为96. 45%,灵敏度为97. 32%,特异性为89. 32%。因此,采用ELM快速检测方法,可有效地对人体运动特征数据进行分类,实现对人体跌倒行为的准确检测。

Research on Human Fall Detection based on ELM Rapid Classification Method

In order to accurately and quickly detect the fall state of human body, an ELM fast classification and discrimination method based on IMU measurement and processing data is proposed. The characteristics of human exercise behavior were analyzed, and the model of leg motion parameter extraction was constructed. The movement data of the human leg is collected by the IMU, and the attitude solution is performed. The ELM method is used to classify the acceleration, angular velocity and attitude of human motion characteristics to determine whether the human body is in a falling state. The ELM parameters are optimized according to the machine learning index evaluation method to obtain the optimal parameters. The human motion state measurement experiment was carried out. The results show that the ELM method can accurately and quickly classify IMU measurement and processing data. When the hidden layer node is 1000, the accuracy of the fall detection of the ELM detection method is 96.45%, the sensitivity is 97.32%, and the specificity is 89.32%. Therefore, using the ELM rapid detection method, the human motion characteristic data can be effectively classified to achieve accurate detection of human fall behavior.