An Autonomous Calibration for IMUs Using Sensor Architecture and Machine Learning Techniques

Inertial measurement units (IMUs) are extensively used in biomechanical research to develop wearable devices for monitoring biomechanical parameters during rehabilitation and disease progression. The accuracy of these measurements critically depends on the calibration process. Our study introduces an autonomous calibration approach for IMUs using sensor architecture and machine learning models to reduce computational complexity and calibration time for accelerometers, gyroscopes and magnetometers in real-time. Our hybrid algorithm combines adaptive bias and scale factor correction with machine learning techniques. Accelerometer was calibrated using Linear Regression and Decision Trees to handle linear and non-linear complexities, while the gyroscope was calibrated using a Forest Regression model and the magnetometer was proposed to be calibrated using Support Vector Machines. Preliminary results demonstrated high accuracy and stability. The sensor architecture approach achieved a Mean Absolute Error of 0.009g for the accelerometer and 0.011 to 0.018 °/sec for the gyroscope, with an overall standard deviation close to zero. The machine learning approach resulted in an accuracy of 0.009g for the accelerometer and 0.011 to 0.012 °/sec for the gyroscope. The total calibration times were ~ 1.16 minutes for the architecture approach and ~ 9 seconds for the ML-based autonomous calibration approach. This innovative approach demonstrates the potential for real-time applications, enhancing the reliability and efficiency of wearable devices in medical and biomechanical fields.