High-precision scene-level metaverse virtual space construction and update technology integrating point cloud semantic information

Aiming at the problems of low automation, insufficient modeling accuracy, and low scene update efficiency of traditional methods in current metaverse scene construction, this study proposes a high-precision scene-level metaverse virtual space construction and update technology that integrates point cloud semantic information. Firstly, this technology optimizes data quality through point cloud preprocessing and achieves fine feature extraction by combining the improved SLAM algorithm. Subsequently, it obtains scene semantic information using the DeepLab V3+ algorithm and completes semantic association via the improved Bayesian method. Then, it constructs a 3D semantic model through secondary clustering corner extraction and format conversion. Finally, it realizes dynamic scene update with the visual bag-of-words and pyramid TF-IDF algorithms. Taking the urban area of Yanshan District in Guilin as the experimental object, the point cloud data acquired by a 3D laser scanner and Gaofen-2 satellite images were used for verification. The results show that the virtual scene constructed by this technology has a planar accuracy of 0.1m and an elevation accuracy of 0.15m, which are 22.5% and 18.8% higher than those of traditional methods, respectively. The scene update time is 40.3% shorter than that of manual iterative update. Compared with advanced methods such as PointNet++ based modeling and ORB-SLAM3, this technology improves semantic matching accuracy by 8.3%–12.1% and reduces update time by 25.6%–33.2%.This technology effectively solves the core problems of constructing and dynamically updating high-fidelity, large-scale metaverse virtual spaces, and provides technical support for the application of the metaverse in urban management, cultural tourism, and other fields.