A novel point cloud simplification method integrates multi-feature fusion based on intensity variations

Abstract With advancements in 3D laser scanning technology, point cloud resolution has achieved submillimeter precision. The high precision and volume of point cloud data pose significant challenges for storage, processing, and visualization. Therefore, this article presents a novel method for point cloud simplification that integrates intensity variations with multiple features. This method preserves geometric features and introduces the reflection intensity from laser scanning as a texture feature, thereby enhancing the retention of texture features. It integrates normal vector deviation, intensity difference, and curvature to identify feature and non-feature points. Specifically, points with large normal vector deviations, distinct intensity differences, or significant curvature are identified as feature points, while points lacking these characteristics are classified as non-feature points. Additionally, an adaptive voxel sampling method is proposed, which estimates the input grid size based on the spatial boundaries and density of the point cloud, ensuring that the sampled points approximate the desired quantity. Non-feature points are sampled using this adaptive method and combined with feature points to obtain the final simplified point cloud. Testing on open-source datasets shows that the proposed method produces distinct features, a uniform point cloud distribution, and small errors. Engineering tests using rock surface data confirm clear features and regular distributions, with a simplification error of only 0.23%. These results demonstrate the effectiveness and accuracy of the proposed method and highlight its superiority.