PointEMRay: A Novel Efficient SBR Framework on Point Based Geometry

TL;DR

PointEMRay addresses the challenge of performing high-frequency EM scattering simulations directly on point-cloud geometry by introducing a screen-based PRI that generates geometric frame buffers (GFBs) via CNN refinement of coarse ray-tube depth maps, and a GFB-assisted MBC that fuses multi-view geometry to recover complete targets. The GO and PO components of SBR are preserved, with a novel pipeline that computes the total scattered field by aggregating per-ray contributions, leveraging efficient ray tracing on GFBs and SLAM-inspired data fusion. The method demonstrates state-of-the-art accuracy and real-time capability on PEC targets, substantially outperforming traditional mesh- or point-based priors such as Poisson reconstruction and splatting in key scenarios (e.g., OctAR) while handling large point clouds efficiently. This work enables practical point-cloud EM simulations with potential impact on rapid prototyping, vehicle/chassis sensing, and scene-level scattering analysis, while outlining avenues for handling finer textures and extending to non-far-field regimes.

Abstract

The rapid computation of electromagnetic (EM) fields across various scenarios has long been a challenge, primarily due to the need for precise geometric models. The emergence of point cloud data offers a potential solution to this issue. However, the lack of electromagnetic simulation algorithms optimized for point-based models remains a significant limitation. In this study, we propose PointEMRay, an innovative shooting and bouncing ray (SBR) framework designed explicitly for point-based geometries. To enable SBR on point clouds, we address two critical challenges: point-ray intersection (PRI) and multiple bounce computation (MBC). For PRI, we propose a screen-based method leveraging deep learning. Initially, we obtain coarse depth maps through ray tube tracing, which are then transformed by a neural network into dense depth maps, normal maps, and intersection masks, collectively referred to as geometric frame buffers (GFBs). For MBC, inspired by simultaneous localization and mapping (SLAM) techniques, we introduce a GFB-assisted approach. This involves aggregating GFBs from various observation angles and integrating them to recover the complete geometry. Subsequently, a ray tracing algorithm is applied to these GFBs to compute the scattering electromagnetic field. Numerical experiments demonstrate the superior performance of PointEMRay in terms of both accuracy and efficiency, including support for real-time simulation. To the best of our knowledge, this study represents the first attempt to develop an SBR framework specifically tailored for point-based models.

Figures (13)

• Figure 1: Illustration of PO and SBR methods. (a) The PO method where only central rays are traced. (b) The SBR method.
• Figure 2: The PointEMRay framework initiates with a raw point cloud input and proceeds to simulate the corresponding EM scattering field. To pinpoint intersections on the point cloud, PointEMRay adopts a screen-based approach. It begins by projecting a multitude of cylindrical rays to generate coarse depth maps. Subsequently, a neural network refines these maps, producing corresponding oriented normal maps and intersection masks. These datasets collectively constitute sets of GFBs. Depending on the application, users can choose between a swift PO simulation or execute multiple GFB fusions to compute the multi-path effects.
• Figure 3: Structures of (a) a ConvNeXt block and (b) our neural network.
• Figure 4: When dealing with ray-splat intersections, additional testing becomes essential. In Figure (a), two splats overlap, resulting in the ideal intersection point $\mathbf{P}$ rather than the closer one. Moreover, a blending technique is employed to achieve a smoother normal estimation. In Figure (b), shifting the origin of the secondary ray aids in avoiding erroneous intersections.
• Figure 5: Some of the meshes included in our dataset.