The article discusses CenterRadarNet, a joint architecture for 3D object detection and re-identification tasks using 4D radar data. It’s an efficient single-stage 3D object detector that directly infers the bird’s-eye-view (BEV) object distribution confidence maps, 3D bounding box attributes, and appearance embedding for each pixel. An online tracker is also built using the learned appearance embedding for re-identification. The system has shown robust performance in diverse driving scenarios, highlighting its wide applicability in autonomous driving and advanced driver assistance systems (ADAS).
Publication date: 3 Nov 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2311.01423
