Monocular Visual-Inertial SLAM:Continuous Preintegration and Reliable Initialization

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous systems.
• Real-time and robust pose estimation in unknown environments remains a challenge.
• Integrating visual and inertial sensors offers complementary data for improved accuracy.

Purpose of the Study:

• To propose a novel visual-inertial SLAM algorithm.
• To achieve robust and real-time sensor pose estimation.
• To enhance the performance of monocular camera and IMU fusion.

Main Methods:

• A tightly coupled sensor fusion approach combining a global shutter monocular camera and an Inertial Measurement Unit (IMU).
• A parallel processing framework featuring a novel IMU initialization method.
• Incorporation of a novel IMU factor, continuous preintegration, directional error vision factor, separability trick, and robust initialization criterion.

Main Results:

• The algorithm provides robust and real-time sensor pose estimates in unknown environments.
• The proposed methods enable efficient and reliable real-time output on modern CPUs.
• Experimental results demonstrate performance comparable to state-of-the-art methods.

Conclusions:

• The developed visual-inertial SLAM algorithm offers a significant advancement in real-time localization and mapping.
• The novel fusion techniques and parallel framework contribute to enhanced robustness and efficiency.
• The algorithm is validated to be competitive with existing leading SLAM solutions.