Objective Ground drilling rescue serves as a critical technical approach within mine disaster emergency response systems. To accurately assess the passage capacity of large-diameter rescue shafts during rescue operations, it is essential to construct precise three-dimensional models of rescue shafts for rapid reconstruction of rescue scenarios.

Methods Based on multi-camera 3D model reconstruction theory, this study investigated the deployment methods of four Intel D435i cameras within a large-diameter rescue shaft. A multi-camera 3D model reconstruction system for large-diameter rescue shafts was established, involving joint calibration of the camera array and optimization of shaft wall image feature extraction and matching algorithms. Experimental research on 3D model reconstruction was conducted using a large-diameter rescue shaft simulation platform. By fitting point cloud data to the shaft diameter, the system enables accurate assessment of rescue well passage capacity.

Results and Conclusions  (1) The proposed four-camera array joint calibration method significantly improves calibration accuracy, with errors below 0.5 pixels. (2) The optimized multi-camera system can rapidly capture well wall images at speeds up to 0.16 m/s. (3) The optimized combination of SFM camera motion and pose estimation, SIFT well wall texture feature extraction, and FLANN false match point removal algorithms enables efficient matching of large-diameter rescue well wall images and rapid 3D model reconstruction, enhancing the realism and visualization quality of the rescue well 3D model. (4) Comparison with the actual radius of a large-diameter rescue shaft verifies that the point cloud model's radial error within one circumference remains below 1.37%, the overall model's point cloud radial error stays under 4%, and 94% of point cloud radial errors fall within 2.5%. This conclusion provides scientific basis for assessing the passage capacity of large-diameter rescue shafts during mine disaster emergency rescue operations.