IMU raw and processed data for computing human joint angles

The goal of this study was to compute the relative angle of human joints such as the knee flex/extension angle using two IMUs. To do so, we utilized two 6-axis (accelerometer, gyroscope) low-cost IMUs (MPU6050, TDK-Invensense, CA, USA) that were mounted on a custom developed test apparatus that replicated the human knee motion. The custom test apparatus contained a single motor that repeatedly rotated one of the IMUs from 0 ~ 180 degrees at a one of three predefined speeds (slow - max speed of 25 deg/s, medium - max speed of 100 deg/s, fast = max speed of 200 deg/s) about one of the three rotation axes (yaw - rotation about gravity vector, pitch - rotation orthogonal to yaw and roll, roll - rotation orthogonal to yaw and pitch) for 25 minutes. One IMU was stationary and mounted securely on the side of the test apparatus while the other IMU was mounted on the motor axis and rotated about the motor-axis. The motor speed could be programmed via a microcontroller. The rotation axis could be changed by re-configuring the orientation of the IMUs. An optical encoder, which was placed on the motor shaft, was used to measure the true rotated angle. As the test apparatus rotated the IMU, raw accelerometer and gyroscopic data from the two IMUs as well as the encoder data were collected using a microcontroller. Five computational algorithms (i.e., accelerometer inclination angle, gyroscopic integration, Complementary Filter, Kalman Filter, Digital Motion Processing) IMU data were used (see https://github.com/ssong47/get_joint_angles_using_imus) to calculated the relative angle between the two IMUs. These computed angles were compared to the gold standard (i.e., encoder angle). A total of 9 trials were collected = three rotation axes (yaw, pitch, roll) x three speeds (slow, medium, fast). Each trial contained two phases: 1) a short calibration phase where the IMUs were stationary and placed such that the origins were parallel to each other, and 2) test phase when the one of the IMUs started to move at the pre-defined speed and rotation axis.

本研究的目标为利用两台惯性测量单元（IMU, Inertial Measurement Unit）计算人体关节的相对角度，例如膝关节屈伸角度。为此，我们采用了两款6轴（集成加速度计与陀螺仪）低成本IMU，型号为MPU6050，产自美国加利福尼亚州的TDK-Invensense公司，将其安装于定制开发的模拟人体膝关节运动的测试装置上。该定制测试装置搭载单台电机，可围绕三种旋转轴之一（偏航——绕重力矢量旋转；俯仰——垂直于偏航与滚转的旋转；滚转——垂直于偏航与俯仰的旋转），以三种预设转速中的一种（慢速：最大转速25 deg/s；中速：最大转速100 deg/s；快速：最大转速200 deg/s），将其中一台IMU反复旋转0~180度，单次测试时长为25分钟。其中一台IMU固定安装于测试装置侧面，另一台IMU安装于电机轴上并随电机轴同步旋转。电机转速可通过微控制器进行编程设定，旋转轴可通过重新调整IMU的安装方位进行切换。安装于电机轴上的光学编码器用于采集真实旋转角度。当测试装置带动IMU旋转时，我们通过微控制器采集两台IMU的原始加速度计与陀螺仪数据以及编码器数据。我们将IMU数据输入五种计算算法（即加速度计倾角法、陀螺仪积分法、互补滤波法、卡尔曼滤波法、数字运动处理法），以计算两台IMU之间的相对角度（详见https://github.com/ssong47/get_joint_angles_using_imus）。将上述计算得到的相对角度与金标准（即编码器测得的真实角度）进行对比分析。本次实验共采集9组测试数据，即三种旋转轴（偏航、俯仰、滚转）与三种转速（慢速、中速、快速）的全组合测试。每组测试包含两个阶段：1）短校准阶段：两台IMU保持静止，且其坐标系原点相互平行；2）测试阶段：其中一台IMU按照预设转速与旋转轴开始旋转运动。