Fast and Accurate Pose Estimation with Unknown Focal Length Using Line Correspondences.

Area of Science:

• Computer Vision
• Photogrammetry
• Robotics

Background:

• Camera pose estimation is crucial for computer vision, photogrammetry, and Simultaneous Localization and Mapping (SLAM).
• Traditional methods often rely on 2D-3D point or line correspondences.
• Simultaneous focal length estimation is necessary when using zoom lenses.

Purpose of the Study:

• To propose a new, fast, and accurate method for camera pose estimation with unknown focal length.
• To utilize two 2D-3D line correspondences and camera position for pose estimation.
• To address limitations of existing methods, particularly with line correspondences.

Main Methods:

• The core contribution converts the perspective-n-line (PnL) problem into a 3D-3D point correspondence problem.
• A key geometric characteristic involving planes defined by 3D lines and camera position is exploited.
• The method establishes a transform between plane normal vectors, analogous to 3D point projection, to estimate pose.

Main Results:

• The proposed method achieves fast and accurate focal length estimation by leveraging the invariance of the angle between planes.
• Experimental results demonstrate good numerical stability and robustness to noise in camera position.
• The method shows strong performance in computational speed and noise sensitivity with both synthetic and real-world data.

Conclusions:

• The new method offers an efficient and robust solution for camera pose and focal length estimation.
• It effectively transforms line-based problems into point-based ones for computational advantage.
• The technique shows significant promise for applications in computer vision, photogrammetry, and SLAM.