Optimization-Based Online Initialization and Calibration of Monocular Visual-Inertial Odometry Considering...

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Online system state initialization and spatial-temporal calibration are crucial for monocular Visual-Inertial Odometry (VIO).
• Existing methods primarily focus on filter-based VIOs, with limited options for optimization-based VIOs.
• Lack of online spatial-temporal calibration methods for optimization-based VIO hinders performance.

Purpose of the Study:

• To propose an optimization-based online initialization and spatial-temporal calibration method for VIO.
• To estimate initial states (metric-scale, velocity, gravity, IMU biases) and calibrate sensor transformations without prior knowledge.
• To improve the accuracy and robustness of VIO systems through enhanced calibration.

Main Methods:

• Utilizes a time offset model and motion interpolation to align camera and IMU data.
• Employs an incremental estimator for initial state and spatial-temporal parameter estimation.
• Integrates bundle adjustment to refine the accuracy of estimated results.

Main Results:

• The proposed method effectively estimates initial states and spatial-temporal parameters for VIO.
• Experiments on synthetic and public datasets demonstrate robust performance.
• Outperforms contemporary methods in VIO initialization and calibration accuracy.

Conclusions:

• The developed method provides a comprehensive solution for online initialization and spatial-temporal calibration in VIO.
• It addresses the limitations of existing methods for optimization-based VIO.
• Offers a significant advancement for achieving accurate and reliable VIO systems.