Event-Frame-Inertial Odometry Using Point and Line Features Based on Coarse-To-Fine Motion Compensation
Byeongpil Choi, Hanyeol Lee, Chan Gook Park
AI summary
Problem
Standard cameras degrade under fast motion and high dynamic range, while event cameras lack static signal and struggle with noisy line detection. This work addresses how to accurately extract and fuse line features from event streams with frame-based data to enhance visual-inertial odometry in structured environments.
Approach
The method uses a coarse-to-fine motion compensation technique to generate sharp event frames, enabling robust line detection via event intensity and RANSAC. These event features are adaptively synchronized and fused with frame-based points in an MSCKF backend.
Key results
- Novel coarse-to-fine motion compensation yields sharp event frames
- Event line features paired with point features eliminate traditional descriptors
- Adaptive synchronization ensures consistent tracking across event and frame timestamps
- Validated on real experiments and 19 public sequences with improved pose accuracy
Why it matters
Enhances the reliability of visual-inertial odometry for robots and drones operating in challenging conditions like fast motion, HDR, or static scenes, particularly in indoor or structured environments.
Abstract
An event camera is a vision sensor that captures pixel-level brightness changes and outputs this information as asynchronous events. These events are primarily generated from geometric structures such as edges, which are sensitive to variations in brightness. In this letter, we aim to leverage this line structure information alongside point features to enhance the robustness and accuracy of localization in indoor or human-made environments. To obtain precise line measurements from events, we propose a novel line detection method that incorporates a coarse-to-fine motion compensation scheme, which generates highly sharp event frames. The extracted line features are paired with point features, eliminating the need for traditional line descriptors. Finally, the event features are effectively fused with frame-based point features within a multi-state constraint Kalman filter-based backend, fully exploiting the complementary advantages of both sensors. The performance of the proposed method is verified through an author-constructed experiment and two public datasets, demonstrating improved accuracy in line detection and pose estimation. Open-source implementation is available at: https://github.com/choibottle/C2F-EFIO.