2026 International Conference on Networks Computers and Communications (ICNCCom 2026)

Physics-guided Shape-from-X (SfX) reconstruction has emerged as a promising approach for improving three-dimensional (3D) scene understanding under challenging imaging conditions where conventional computer vision algorithms often fail. This paper proposes a robust physics-guided SfX framework that integrates shape-from-shading, shape-from-polarization, shape-from-texture, and depth cues with physics-constrained deep learning models for accurate surface geometry estimation in adverse environments. The proposed method incorporates illumination modeling, reflectance consistency, atmospheric scattering correction, and geometric priors to improve reconstruction accuracy under low-light, foggy, noisy, and non-Lambertian imaging conditions. A hybrid convolutional-transformer architecture is employed to fuse multimodal visual features and enforce physical constraints during optimization. Experimental evaluations are conducted using synthetic and real-world datasets captured in autonomous driving, industrial inspection, and remote sensing scenarios. Results demonstrate that the proposed framework achieves superior reconstruction accuracy, lower depth estimation error, and improved robustness compared with conventional Shape-from-X and purely data-driven methods. The framework also exhibits strong generalization capability under varying environmental conditions. The study highlights the potential of physics-guided vision systems for reliable 3D perception in next-generation intelligent imaging and autonomous systems.

Reconstruction;Intelligence Imaging;Artificial Intelligence;Process Innovation