TY - GEN
T1 - Underwater Depth Estimation using Pinax Model for Refraction Modeling and Correction
AU - Elsayed, Hussein H.
AU - Abdelaal, Amr S.
AU - Farouk, Ahmed R.
AU - Selem, Hossam A.
AU - Abdelmohsen, Mustafa M.
AU - Nabawy, Mahmoud Y.
AU - Zaki, Eyad M.
AU - Mohamed, Samer A.
AU - Awad, Mohammed I.
AU - Abd El Munim, Hossam E.
PY - 2025/11/14
Y1 - 2025/11/14
N2 - Accurate depth perception is essential for autonomous underwater vehicles (AUVs), particularly in GPS-denied environments where SLAM depends on precise distance measurements to visual landmarks. However, stereo vision systems struggle underwater due to light refraction at the air-housing-water interfaces, which distorts image geometry and violates the assumptions of the pinhole camera model. In this work, we present a stereo perception module based on an extended implementation of the Pinax model, designed to correct refractive distortion in flat-port camera systems. By generating physically grounded correction maps, our work transforms distorted images into geometrically consistent projections compatible with standard stereo pipelines. The system is deployed onboard an AUV and validated in a controlled underwater environment using synchronized 1D sonar as ground truth. Results show that the corrected depth estimates are consistently accurate, with errors bound below 18 cm and an average error of just 10 cm, demonstrating the practicality of real-time underwater depth estimation using passive stereo vision.
AB - Accurate depth perception is essential for autonomous underwater vehicles (AUVs), particularly in GPS-denied environments where SLAM depends on precise distance measurements to visual landmarks. However, stereo vision systems struggle underwater due to light refraction at the air-housing-water interfaces, which distorts image geometry and violates the assumptions of the pinhole camera model. In this work, we present a stereo perception module based on an extended implementation of the Pinax model, designed to correct refractive distortion in flat-port camera systems. By generating physically grounded correction maps, our work transforms distorted images into geometrically consistent projections compatible with standard stereo pipelines. The system is deployed onboard an AUV and validated in a controlled underwater environment using synchronized 1D sonar as ground truth. Results show that the corrected depth estimates are consistently accurate, with errors bound below 18 cm and an average error of just 10 cm, demonstrating the practicality of real-time underwater depth estimation using passive stereo vision.
KW - Underwater 3D Reconstruction
KW - Underwater Depth Estimation
KW - Underwater Refraction Correction
UR - https://www.scopus.com/pages/publications/105027207600
U2 - 10.1109/NILES68063.2025.11232148
DO - 10.1109/NILES68063.2025.11232148
M3 - Chapter in a published conference proceeding
AN - SCOPUS:105027207600
T3 - 7th Novel Intelligent and Leading Emerging Sciences Conference, NILES 2025 - Proceedings
SP - 223
EP - 226
BT - 7th Novel Intelligent and Leading Emerging Sciences Conference, NILES 2025 - Proceedings
PB - IEEE
CY - U. S. A.
T2 - 7th International IEEE Novel Intelligent and Leading Emerging Sciences Conference, NILES 2025 - Proceedings
Y2 - 25 October 2025 through 27 October 2025
ER -