机器学习辅助IMU人体跌倒状态识别

Machine Learning Assisted IMU Human Fall State Recognition

针对单一传感器在人体运动状态监测中误差较大的问题,本文采用了高精度的陀螺加速度计MPU6050模块。该模块使用陀螺仪输出的人体运动信息对加速度传感器采集到的姿态角信息进行修正,提高测量精度。采用机器学习的方法对样本数据进行分析,获得分类阈值。当人体由静止到跑步或由站立到坐下等状态转移时,加速度也会突然增大,有可能达到跌倒时的阈值。当采集的加速度信息的特征值大于阈值时,通过分析其产生峰值时的加速度值和该时刻及之后的角度变化,来确定是否发生跌倒事件。

For the problem of high bias in the monitoring of human motion in a single sensor,the high precision gyro accelerometer MPU6050 module is adopted. The module uses the human body motion information output from the gyroscope to correct the attitude angle information collected by the acceleration sensor to improve the measurement accuracy. The method of machine learning is used to analyze the sample data and obtain the threshold of classification. When the human body moves from static to running or standing to sit and down,the acceleration will suddenly increase and it is possible to reach the threshold when it falls. When the feature value of the acquired acceleration information is bigger than the threshold value,it is determined whether or not a fall event has occurred by analyzing the acceleration value at the time of generating the peak value and the angle change after that time and thereafter.