In order to achieve high-precision robot machining requirements for large workpieces, a 3D reconstruction process of the workpiece is necessary. This paper proposed an accurate 3D reconstruction method for large workpieces based on 3D vision. The method contains two steps: Considering the substantial size of large workpieces, in the first step we established a measurement-machining system that captures point cloud pieces using a point cloud camera. Using coordinate transformations, we obtained a coarse registration result. Due to factors including calibration errors and robot motion inaccuracies, etc., misalignments and displacements exist between point cloud pieces; To address this, we employed an improved Iterative Closest Point (ICP) algorithm to perform precise registration as the second step and ultimately completed the accurate 3D reconstruction process. Experimental results showed significant reductions in both point-to-tangent plane and point-to-point errors after precise registration, all within acceptable error margins. The point-to-tangent plane errors are close to point-to-point errors, which means fine alignments lies in both position and pose. These results validate the reliability and effectiveness of the measurement-machining system for large workpieces 3D reconstruction and demonstrate the high accuracy of the proposed 3D reconstruction method.