Sharp Feature-Preserving 3D Mesh Reconstruction from Point Clouds Based on Primitive Detection

High-fidelity mesh reconstruction from point clouds has long been a fundamental research topic in computer vision and graphics. Traditional methods require dense triangle meshes to achieve high fidelity, but excessively triangles may lead unnecessary storage computational burdens, while also struggling capture clear, sharp, continuous edges. This paper argues that the key high-fidelity lies preserving sharp features. Therefore, we introduce novel sharp-feature-preserving framework based on primitive detection. It includes an improved deep-learning-based detection module two splitting selection modules propose. Our can accurately reasonably segment patches, fit each patch, split overlapping at level ensure true sharpness obtaining lightweight models. Quantitative visual experimental results demonstrate our outperforms both state-of-the-art learning-based traditional methods. Moreover, designed are plug-and-play, which not only apply detectors be combined with other cloud processing tasks such as edge extraction or random sample consensus (RANSAC) results.