Tackling Clutter in Radar Data -- Label Generation and Detection Using PointNet++

Johannes KoppDominik KellnerAldi PiroliKlaus Dietmayer

Mar 2023

摘要

Radar sensors employed for environment perception, e.g. in autonomousvehicles, output a lot of unwanted clutter. These points, for which nocorresponding real objects exist, are a major source of errors in followingprocessing steps like object detection or tracking. We therefore present twonovel neural network setups for identifying clutter. The input data, networkarchitectures and training configuration are adjusted specifically for thistask. Special attention is paid to the downsampling of point clouds composed ofmultiple sensor scans. In an extensive evaluation, the new setups displaysubstantially better performance than existing approaches. Because there is nosuitable public data set in which clutter is annotated, we design a method toautomatically generate the respective labels. By applying it to existing datawith object annotations and releasing its code, we effectively create the firstfreely available radar clutter data set representing real-world drivingscenarios. Code and instructions are accessible atwww.github.com/kopp-j/clutter-ds.

展开全部

图表提取

论文十问

参考文献

被引用

提问与回答

暂无人提供速读十问回答

被引用

笔记

问答

领域