High-precision 3D reconstruction systems serve as critical enablers for robotic machining applications demanding sub-millimeter accuracy in complex manufacturing environments. Especially heavy-load manufacturing scenarios present unique challenges, particularly when handling large-scale irregularly shaped, textureless objects with geometric repetitions. In current, conventional single-view 3D capture methods suffer from incomplete data acquisition, and meanwhile the expensive commercial handheld 3D scanning system requires heavy time-consuming manual placement of fiducial markers and the handheld scanning process. To address these limitations, we propose an automatic robotic 3D modeling system incorporating key innovations. Active projection of speckle patterns onto object surfaces enables stable optical feature encoding, effectively mitigating the challenge of feature scarcity on homogeneous surfaces. With a globally optimized pose estimation framework for multiview data registration, bundle adjustment with temporal consistency constraints has been implemented, to achieve cumulative error reduction across sequential scanning trajectories. Experimental validation through different objects’ reconstruction results demonstrates our proposed method’s capability both in accuracy, robustness and efficiency.