Skip to main navigation menu Skip to main content Skip to site footer

Fast, Accurate and Object Boundary-Aware Surface Normal Estimation from Depth Maps


In this paper, we introduce a fast and precise surface normal estimation technique designed explicitly for depth maps, also known as organized point clouds. Our approach formulates the surface normal estimation as a closed-form expression, effectively mitigating the impact of measurement noise through multi-directional averaging. We then streamline the multi-directional normal estimation process for efficiency. Additionally, we propose a straightforward yet powerful method to eliminate inaccurate normal estimations at depth discontinuities, making our approach object boundary-aware. Comparative analyses with established surface normal estimation algorithms reveal that our method not only excels in accuracy but also exhibits the speed required for real-time applications.