This is the official implementation of ICLR2025 paper: RobuRCDet: Enhancing Robustness of Radar-Camera Fusion in Bird's Eye View for 3D Object Detection.
In this paper, we first conduct a systematic analysis of robustness in radar-camera de�tection on five kinds of noises and propose RobuRCDet, a robust object detection model in bird’s eye view (BEV). To mitigate inaccuracies in radar points, including position, Radar Cross-Section (RCS), and velocity, we design a 3D Gaussian Expan�sion (3DGE) module to achieve radar denoising. We further introduce a weather-adaptive fusion module, which adaptively fuses radar and camera features based on camera signal confi�dence. RobuRCDet achieves competitive results in both regular and noisy conditions.
This project is based on excellent open source projects: