This website requires JavaScript.

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

Jean Pierre RichaJean-Emmanuel DeschaudFran\c{c}ois GouletteNicolas Dalmasso
Mar 2022
摘要
LiDAR sensors provide rich 3D information about their surrounding and arebecoming increasingly important for autonomous vehicles tasks, such as semanticsegmentation, object detection, and tracking. Simulating a LiDAR sensoraccelerates the testing, validation, and deployment of autonomous vehicles,while reducing the cost and eliminating the risks of testing in real-worldscenarios. We address the problem of high-fidelity LiDAR simulation and presenta pipeline that leverages real-world point clouds acquired by mobile mappingsystems. Point-based geometry representations, more specifically splats, haveproven their ability to accurately model the underlying surface in very largepoint clouds. We introduce an adaptive splats generation method that accuratelymodels the underlying 3D geometry, especially for thin structures. Moreover, weintroduce a physics-based, faster-than-real-time LiDAR simulator, in thesplatted model, leveraging the GPU parallel architecture with an accelerationstructure, while focusing on efficiently handling large point clouds. We testour LiDAR simulation in real-world conditions, showing qualitative andquantitative results compared to basic splatting and meshing techniques,demonstrating the interest of our modeling technique.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答