Efficient minimal solvers for relative pose estimation in autonomous driving applications.

Area of Science:

• Computer Vision
• Robotics
• Autonomous Systems

Background:

• Computer vision is crucial for autonomous driving and robot navigation.
• Relative pose estimation in multi-camera systems is vital for localization and perception.
• Current methods are computationally expensive and require many features, hindering real-time applications.

Purpose of the Study:

• To develop a unified framework for efficient relative pose estimation.
• To introduce novel translation parameterization and first-order rotation approximation.
• To propose three efficient minimal solvers tailored for autonomous vehicles.

Main Methods:

• Developed a unified framework for efficient relative pose estimation.
• Introduced novel translation parameterization and first-order rotation approximation.
• Proposed three minimal solvers leveraging vertical direction prior (IMUs), rotation axis prior, and planar motion assumptions.

Main Results:

• The proposed solvers reduce the number of point correspondences and algebraic complexity.
• Methods enable faster hypothesis generation in RANSAC-based pipelines for real-time systems.
• Achieved a favorable balance between speed and accuracy on synthetic and KITTI datasets.

Conclusions:

• The novel framework and solvers enhance real-time relative pose estimation for autonomous vehicles.
• The proposed methods offer a practical solution for time-sensitive driving scenarios.
• Demonstrated superior performance compared to existing state-of-the-art algorithms in speed and accuracy.