AdaSplats: Adaptive Splatting of Point Clouds for Accurate 3D Modeling and Real-Time High-Fidelity LiDAR Simulation

LiDAR sensors provide rich 3D information about their surroundings and are becoming increasingly important for autonomous vehicles tasks such as localization, semantic segmentation, object detection, tracking. Simulation accelerates the testing, validation, deployment of while also reducing cost eliminating risks testing in real-world scenarios. We address problem high-fidelity simulation present a pipeline that leverages point clouds acquired by mobile mapping systems. Point-based geometry representations, more specifically splats (2D oriented disks with normals), have proven ability to accurately model underlying surface large clouds, mainly uniform density. introduce an adaptive splat generation method models handle variable densities, especially thin structures. Moreover, we fast sensor simulator, working splatted model, GPU parallel architecture acceleration structure focusing on efficiently handling clouds. test our conditions, showing qualitative quantitative results compared basic splatting meshing techniques, demonstrating interest modeling technique.