Robust Tracking of Vibrating Rods for Precast Beams via Visual Fusion and Online Self-Correction

The compactness of concrete vibration in precast box beams is a critical determinant of structural durability. Traditional manufacturing processes, however, rely heavily on manual experience and lack precise positioning and quantitative standards, frequently leading to defects such as missed vibration and under-vibration. Achieving precise positioning of vibrating rods presents a significant technical challenge in indoor precasting environments characterized by GNSS-denied conditions, dense reinforcement, and dust occlusion. To address these issues, this study proposes and validates a high-precision positioning and dynamic identification system for vibrating equipment based on multi-modal sensor fusion. This system effectively resolves the challenges of field-of-view coverage and dynamic occlusion across a standard 30-meter precasting pedestal. At the algorithmic level, a two-layer coordinate transformation framework comprising offline calibration and online correction is established. Simultaneously, the YOLOv13 deep learning network is employed for real-time object detection. By integrating point cloud filtering and plane fitting techniques, the system calculates the three-dimensional coordinates and spatial pose of the vibrating rod within the box beam coordinate system. Experimental results demonstrate the system's exceptional robustness in complex, real-world beam plant environments. The dynamic tracking refresh rate exceeds 10 Hz, with both static and dynamic positioning errors maintained within the centimeter-level range. This research holds significant theoretical importance and engineering application value for promoting the intelligent transformation of precast beam production.