A Multi-Camera System-Based Relative Pose Estimation and Virtual–Physical Collision Detection Methods for the Underground Anchor Digging Equipment

This work proposes a novel multi-camera system-based method for relative pose estimation and virtual–physical collision detection for anchor digging equipment. It is dedicated to addressing the critical challenges of achieving accurate relative pose estimation and reliable collision detection between multiple devices during collaborative operations in coal mines. The key innovation is that the multi-camera multi-target system is established to collect images, and the relative pose estimation is completed by the EPNP (Efficient Perspective N-Point) algorithm based on multiple infrared LED targets. At the same time, combined with the characteristics of a roadheader and anchor drilling machine, AABB (Axis Alignment Bounding Box) with a simple structure and convex hull with a strong wrapping are selected to create the mixed hierarchical bounding box, and the collision detection is carried out by combining SAT (Split Axis Theorem) and GJK (Gilbert–Johnson–Keerthi) algorithms. The experimental results show that the relative pose estimation error of the multi-camera system is within 20 mm, with an angular error within 1.002°. The position error in the X -axis direction is within 1.160 mm, and the maximum deviation in the Y -axis direction is within 0.957 mm in the virtual–physical space. Compared with the existing methods, our method integrates digital twin technology, and has a simple system structure, which can meet the requirements of relative attitude estimation and collision detection between equipment in the process of heading face operation, and at the same time improve the system performance.