Researchers at Harbin Engineering University have developed a new method for reconstructing 3D underwater terrain from 2D side-scan sonar images. The approach, based on shape-from-shading (SFS), combines the BFGS quasi-Newton optimization algorithm with the alternating direction method of multipliers (ADMM) to recover depth and surface normals from acoustic shadow information. A preconditioned conjugate gradient (PCG) solver handles linear depth updates, ensuring consistency and smoothness in the final reconstruction.
Key takeaways
- The BFGS-ADMM method iteratively optimizes an energy function derived from a Lambertian reflection model, enabling accurate depth estimation from sonar intensity variations.
- ADMM decomposes the complex optimization into simpler subproblems, improving stability and convergence, while BFGS provides fast convergence for nonlinear parameters.
- PCG efficiently solves the linear depth update step, reducing computational cost compared to traditional approaches.
- Experimental validation shows the method outperforms existing iterative optimization algorithms in terms of accuracy and computational efficiency, meeting the needs of underwater terrain reconstruction.
This work addresses limitations of prior SFS techniques, such as poor handling of noise and high computational complexity, offering a practical solution for applications like pipeline tracking, wreck inspection, and seabed characterization.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2026-05-12 · “基于侧扫声呐图像的水下3维地形环境重建”