The virtual pre assembly process of bridge steel components faces challenges such as time consumption, large footprint, and high costs. This paper proposes an automatic virtual pre assembly technology for bridge steel components, employing supervoxel segmentation and a region growing algorithm to achieve precise segmentation of component point cloud feature surfaces. The method accommodates the unique geometry of carling and crossbeam members. Using singular value decomposition algorithm, random consistency fitting algorithm and spatial geometry principle to automatically extract the feature points needed for Mosaic, the traditional point cloud Mosaic control points in the manual selection of large subjectivity, accuracy is difficult to control and other problems effectively solved. Moreover, the paper introduces data threshold and design value auxiliary line methods to ensure accurate assembly, even in scenarios with missing component scanning point cloud data. Engineering application results demonstrate an average error of 0.62 mm, a maximum error of only 1.32 mm, and a maximum angle error of 0.139°, indicating reasonable accuracy control and strong applicability. Based on the research results, a group of algorithm plug ins are developed on the Cloud Compare platform to further realize the automation and standardization of the stitching process. The research results can provide theoretical and algorithmic support for virtual preassembly of bridge steel and similar engineering projects, and reflect better production and application value.