CUAHN-VIO: Content-and-Uncertainty-Aware Homography Network for Visual-Inertial Odometry

Abstract--Learning-based visual ego-motion estimation is promising yet not ready for navigating agile mobile robots in the real world. In this article, we propose CUAHN-VIO, a robust and efficient monocular visual-inertial odometry (VIO) designed for micro aerial vehicles (MAVs) equipped with a downward-facing camera. The vision frontend is a content-anduncertainty-aware homography network (CUAHN) that is robust to non-homography image content and failure cases of network prediction. It not only predicts the homography transformation but also estimates its uncertainty. The training is self-supervised, so that it does not require ground truth that is often difficult to obtain. The network has good generalization that enables "plugand-play" deployment in new environments without fine-tuning. A lightweight extended Kalman filter (EKF) serves as the VIO backend and utilizes the mean prediction and variance estimation from the network for visual measurement updates. CUAHN-VIO is evaluated on a high-speed public dataset and shows rivaling accuracy to state-of-the-art (SOTA) VIO approaches. Visualized outputs of CUAHN-VIO evaluated on Seq. 13 of the UZH-processor to navigate a fast autonomous MAV. CUAHN is robust to sparse non-planar objects and motion blur. The arrows in pink are the networkpredicted optical flow vectors of the four corner pixels. They are plotted according to the uncertainty estimation from the network. The code developed for CUAHN-VIO will be open-sourced at the same link upon the publication of this article. A video of VIO runtime have inherent defects. They are often affected by disadvantageous performance is available at https://youtu.be/ NgDkgON-nE. An alternative is learning to predict camera ego-motion by a deep neural network (DNN). HANKS to the rapid development of computer vision and state estimation techniques, VIO has become a DNNs better cope with visually degraded conditions than their trustworthy component of autonomous robots, such as MAVs.