<p>Understanding 3D point clouds is a key challenge in the field of computer vision, with significant implications for applications such as autonomous navigation, robotics, and augmented reality. Traditional methods often rely on parametric models and heavily trained programs, which can be computationally expensive and difficult to generalize. To address these issues, we propose a new nonparametric approach to understanding 3D point clouds, utilizing curvature guidance to enhance geometric feature extraction. Our method comprises three key components: using spatial curvature information for local sampling to select information points; local encoding without training parameters, a robust encoding scheme that bypasses the need for parameter learning; and a knowledge warehouse that stores basic geometric information for efficient retrieval and processing. We validated the proposed method on the 3D classification dataset ScanObjectNN (three variants) and ModelNet40, as well as on the component/scene segmentation datasets ShapeNet and S3DIS, with and without parameter training. The experimental results demonstrate that our method achieves competitive accuracy in various point cloud segmentation and classification tasks while maintaining low computational overhead.</p>