Camera self-calibration from translation by referring to a known camera

Area of Science:

• Computer Vision
• Robotics
• Photogrammetry

Background:

• Camera calibration is crucial for 3D scene reconstruction and robotic vision.
• Existing methods often require complex motion constraints or known scene structures.
• A need exists for robust and practical self-calibration techniques.

Purpose of the Study:

• To develop a novel linear method for camera self-calibration.
• To leverage a known reference camera and pure translation for calibration.
• To enable accurate and practical online camera calibration.

Main Methods:

• Compute infinite homography from scene depths.
• Recover scene depths using two images from pure translation of an uncalibrated camera.
• Linearly compute infinite homography between uncalibrated and reference camera images.
• Decompose infinite homography for camera calibration.

Main Results:

• The proposed method demonstrates practical and accurate results in simulations and real-world tests.
• Pure translation motion is shown to be more maneuverable than strict motions like pure rotation.
• The method successfully calibrates uncalibrated cameras using a known reference camera.

Conclusions:

• The novel linear camera self-calibration method is effective and accurate.
• The technique offers a more flexible motion requirement (pure translation).
• This method has significant potential for online camera calibration in multi-camera and zooming systems.