Calib-Net: Calibrating the Low-Cost IMU via Deep Convolutional Neural Network

Area of Science:

• Robotics and Sensor Fusion
• Deep Learning for Sensor Calibration

Background:

• Low-cost Inertial Measurement Units (IMUs) are crucial for orientation sensing but suffer from calibration inaccuracies.
• Poor IMU calibration leads to significant angular velocity errors and rapid orientation drift.

Purpose of the Study:

• To introduce Calib-Net, a novel deep convolutional neural network for accurate low-cost IMU calibration.
• To dynamically compensate gyroscope measurements using a sophisticated mathematical model.

Main Methods:

• Developed Calib-Net, a deep convolutional neural network incorporating dilated convolution for spatio-temporal feature extraction.
• Implemented a mathematical calibration model to generate dynamic compensation components for gyroscope data.

Main Results:

• Calib-Net demonstrated superior quantitative and qualitative calibration performance compared to existing methods.
• Orientation estimation accuracy using Calib-Net was found to be comparable to Visual Inertial Odometry (VIO) systems.

Conclusions:

• Calib-Net offers an effective and efficient solution for accurate low-cost IMU calibration.
• The proposed deep learning approach significantly enhances orientation estimation reliability for IMU applications.