High-density point cloud simplification for steel elements with active profile features preservation

Abstract Laser scanning-based virtual trial assembly and quality inspection are emerging in steel structures. However, point cloud data of steel components generated through multi-view, fine-resolution acquisition often has high density and large data volume, which imposes substantial computational burdens on downstream tasks such as assembly and inspection. The simplification methods commonly used in engineering mainly adopt uniform simplification, which may result in the loss of crucial information. This paper presents a high-density point cloud simplification method for steel components with active profile features preservation. A novel profile-feature-oriented dual-factor point importance evaluation method is proposed. Based on information entropy theory, this method can evaluate the importance of each point with respect to the profile features of steel components from both the spatial distribution and the statistical characteristics of the point cloud data. Subsequently, smoothing and closing operations are adopted to ensure a comprehensive evaluation. A profile information recognition approach is also developed, which utilizes the importance evaluation results to generate virtual points containing both positional and importance attributes, and actively recognizes point cloud data corresponding to the profile features according to the geometric distribution of the virtual points. On this basis, a simplification process is designed that considers both sampling uniformity and differences in point cloud density, prioritizing the simplification of non-feature data. This enables the generation of simplified results that preserve the profile features of steel components, improving computational efficiency while maintaining accuracy in downstream tasks. A series of simulation tests and application cases are conducted to examine the proposed method.