Autonomous launching gantry: Improved monocular vision approach for real-time pose estimation of precast concrete girders

The absence of accurate and real-time 6-DoF pose data for precast concrete girders renders launching gantry operations predominantly manual, thereby impeding further automation. Such limitations pose a critical question: How can we accurately and robustly estimate the pose of precast concrete girders in real-time during launching gantry operations? To address that question, our paper proposes a monocular vision-based approach to estimate the 6-DoF pose of the precast concrete girder in launching gantry operations. The approach detects the ChArUco board regions using the YOLOv11n model, applies GAN-based image deblurring. The 6-DoF pose is then estimated using a Perspective-n-Point solver and transformed to the gantry coordinate system. Field tests demonstrate robust performance, achieving a mean reprojection error of 0.113 pixels and a processing latency of 60 ms per frame. The results validate the approach's robustness and real-time performance, highlighting monocular vision as a cost-effective alternative to LiDAR–IMU fusion for large-scale automation in construction.