A new sparse reconstruction framework integrating parallel photography with laser-coplanar directional control for constrained underground spaces

In the field of underground engineering, particularly in water conservancy, hydropower, and transportation tunnel projects, the high-resolution reconstruction of deep-buried long tunnels and other confined underground spaces serves as a fundamental data source for precise, efficient, and safe measurement of geological parameters. Due to the lack of control points and GNSS, important orientational data such as geological spatial attitude are difficult to be measured. Parallel photogrammetry can quickly acquire highly overlapping image data, but it still suffers from significant cumulative model drift and lack of absolute true geo-referenced pose in long tunnel reconstruction. To address these challenges, a novel sparse reconstruction method is proposed that integrates parallel photography with laser-coplanar directional control. First, an absolute orientation control system is established, in which the underground spaces with true directional geo-reference can be determined by combination of a geological compass and a laser level, i.e., a geological compass is used to measure absolute direction and a laser level is used to determine the vertical and horizontal pose angles parameters. Laser feature correspondences are extracted from both vertical and horizontal laser planars through a deep convolutional matching network guided by projected laser lines. These absolute orientation parameters, together with the global planar constraints, are incorporated as novel geometric constraints for the bundle adjustment optimization. The proposed method not only provides an absolute direction reference in constrained underground scenes without relying on GNSS, effectively mitigates model drift and enhances sparse reconstruction accuracy.