Implicit calibration method for underwater stereo cameras.

Area of Science:

• Robotics
• Computer Vision
• Optical Engineering

Background:

• Underwater stereo cameras offer precise perception in challenging environments.
• Refraction causes nonlinear distortions in stereo imaging, complicating calibration.
• Standard single viewpoint projection assumptions are inadequate for underwater stereo vision.

Purpose of the Study:

• To propose a data-driven implicit calibration method for underwater stereo cameras.
• To address nonlinear distortions and aberration distributions unique to underwater imaging.
• To enhance the accuracy of 3D reconstruction and camera posture estimation in underwater applications.

Main Methods:

• Developed a corresponding coordinates regression network and fusion strategy for network-based calibration.
• Designed an underwater self-luminous calibration target system for sample dataset acquisition.
• Implemented an underwater corner point extraction strategy for efficient data collection.

Main Results:

• The proposed implicit calibration method demonstrated superiority over explicit calibration techniques.
• Comprehensive evaluation showed improved performance in measurement, camera posture estimation, and 3D reconstruction.
• Experimental results validate the method's effectiveness for real-world underwater vision tasks.

Conclusions:

• The developed implicit calibration method offers an efficient and accurate solution for underwater stereo cameras.
• This data-driven approach overcomes limitations of traditional calibration methods in complex underwater environments.
• The technique significantly advances the capabilities of underwater vision systems.