The article discusses the challenges in applying visual-inertial odometry (VIO) to a micro air vehicle (MAV) with a downward-looking camera. The main issues are incorrect initialization during take-off and significant cumulative drift during flight. A homography-based initialization method is proposed to resolve these issues, which takes advantage of the fact that features detected by the camera at take-off are approximately on the same plane. The method introduces a prior normal vector and motion field for more accurate states. To manage the cumulative drift, a strategy for dynamically weighting visual residuals is proposed. The method was tested on real-world datasets and showed promising results in improving initialization and positioning errors.
Publication date: 20 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.09622
