Visual-Inertial Odometry with Robust Initialization and Online Scale Estimation.

Area of Science:

• Robotics
• Computer Vision
• Navigation Systems

Background:

• Visual-inertial odometry (VIO) is crucial for unmanned aerial vehicle (UAV) ego-motion estimation.
• Optimization-based VIO methods offer high accuracy but can diverge and are sensitive to initialization.
• Inertial measurement unit (IMU) parameter accuracy significantly impacts VIO performance.

Purpose of the Study:

• To propose a novel VIO algorithm for accurate UAV motion state estimation.
• To enhance the robustness and stability of VIO systems.
• To address challenges in VIO initialization, particularly scale and gravity ambiguity.

Main Methods:

• Fusion of visual information and pre-integrated inertial measurements within a joint optimization framework.
• Stable initialization of scale and gravity using relative pose constraints.
• Adoption of a local scale parameter in online optimization to handle initialization uncertainty.

Main Results:

• The proposed VIO algorithm demonstrates high accuracy in estimating UAV motional states.
• The method shows improved stability compared to existing VIO algorithms.
• Quantitative comparisons on the EuRoC dataset validate the algorithm's efficacy.

Conclusions:

• The novel VIO algorithm effectively estimates UAV motion with high accuracy and stability.
• The joint optimization framework and robust initialization contribute to improved performance.
• The approach offers a reliable solution for VIO challenges in UAV applications.