Attitude normalization based line feature point cloud extraction approach

Extracting the point cloud corresponding to the edges from point cloud of a part is a basic work in the part 3D reconstruction, also a key point of point cloud processing. The commonly used curvature based LFPC(line feature point cloud) extraction approach may have large errors in curve surface fitting and curvature estimation due to the arbitrary attitude of original point cloud and the way of curvature estimation. An attitude normalization based LFPC extraction approach is proposed. First the attitude normalization is accomplished by adjust the principle direction of the point cloud to the z axis, then, after the curve surface fitting, the main curvature with the maximal absolute value is treated as curvature, finally, the LFPC is obtained by applying a curvature threshold. Experiments on different models show that the proposed approach has better efficiency and is adaptive to different kinds of models.