Improvement of normal estimation for point clouds via simplifying surface fitting

Jun Zhou, Wei Jin, Mingjie Wang, Xiuping Liu, Zhiyang Li, Zhaobin Liu

Abstract: Highlights•A novel point update strategy is proposed to effectively adjust the points of a given patch to proper locations, making the surface fitting process easier, particularly for a patch involving sharp features.•A dynamic selection mechanism is delicately designed, namely Top-K selection strategy. It is devoted to drive the model to adopt a precise weight learning process, and to achieve a straightforward fitting procedure by attending the critical points.•In addition to the proposed strategies, the collaboration between the weight learning networks and the classic surface fitting method tends to be more effective. Extensive experiments demonstrate the superiority of our method against state-of-the-art methods.