VINS-MKF：A Tightly-Coupled Multi-Keyframe Visual-Inertial Odometry for Accurate and Robust State Estimation

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Fusion

Background:

• Accurate state estimation is critical for robot autonomy.
• Existing visual odometry (VO) methods struggle with limited fields of view (FOV) in complex conditions.
• There is a need for robust and accurate state estimation solutions for indoor robotics.

Purpose of the Study:

• To present a novel tightly-coupled multi-keyframe visual-inertial odometry (VINS-MKF) system.
• To enhance state estimation accuracy and robustness for robots in indoor environments.
• To leverage large FOV from multiple fisheye cameras and IMU data.

Main Methods:

• Modified ORBSLAM (Oriented FAST and Rotated BRIEF Simultaneous Localization and Mapping) for multi-fisheye camera and IMU integration.
• GPU-accelerated feature extraction with parallelized threads for efficient VO framework.
• Tightly-coupled, multi-keyframe, visual-inertial nonlinear optimization with accurate initialization and a novel MultiCol-IMU camera model.

Main Results:

• VINS-MKF demonstrates improved accuracy and robustness compared to state-of-the-art VINS-Mono.
• The system effectively integrates data from multiple fisheye cameras and an IMU.
• Extensive experiments on custom datasets validate the performance gains.

Conclusions:

• VINS-MKF is the first tightly-coupled multi-keyframe visual-inertial odometry system combining multiple fisheye cameras and IMU measurements.
• The proposed system significantly enhances robot state estimation in challenging indoor environments.
• The novel framework offers a promising solution for robust and accurate robot autonomy.