"Geometry Constrained Camera and LiDAR Fusion in Underground Confined Spaces"

"Camera sensors are highly susceptible to environmental disturbances in complex underground scenarios, such as variable illumination, dust, mist, and occlusions, which introduce measurement noise and severely degrade the accuracy of 3D reconstruction. To address these challenges, a robust denoising method is proposed, in which single-line LiDAR and depth camera data are fused within a unified geometric framework. The complementary characteristics of sparse but geometrically accurate LiDAR and dense but noisy depth data are exploited through surface parameterization and geometry-constrained residual correction. A joint calibration process is conducted to align point clouds from sensors within a unified coordinate system, which ensures reliable geometric correspondence. The denoising pipeline is composed of principal-axis-guided surface parameterization of the fused point cloud, followed by residual optimization in the parameter domain, in which local geometric constraints and prediction consistency are jointly incorporated to correct noisy depth observations.In real tunnel environments, the proposed method achieves up to 25\\% reduction in average noise and over 59\\% improvement in processing time compared to a 3D LiDAR fusion baseline, while also outperforming traditional filtering and other regression methods. Even under multiple complex operating conditions, noise and dimensional errors are reduced by more than 89\\% relative to the original depth camera data, indicating that accurate 3D reconstruction can still be maintained with strong robustness and adaptability."