Boundary-Aware Consistent Normal Orientation for Point Clouds

Inferring a globally consistent normal orientation for point clouds remains challenging, especially for noisy and non-watertight point clouds. To improve accuracy and robustness in orientation inference, a Boundary-Aware Consistent Normal Orientation (BACNO) method is proposed. Its main idea is to transform the normal orientation problem into a boundary-aware narrow band grid partitioning problem. This processing process is as follows: First, an unsigned distance field for an input point cloud is computed, which is defined on a regular grid. The field is then trimmed as a boundary-aware narrow band grid around the point cloud. Next, the narrow band grid is segmented into two parts, with each part located on one side of the input point cloud. Finally, a coarse-to-fine normal orientation strategy is presented to achieve the globally consistent orientation. Extensive experimental results demonstrate that the proposed method outperforms state-of-the-art methods, particularly for noisy and non-watertight point clouds.