🤖 AI Summary
This study addresses the challenges of autonomous driving object detection arising from the sparsity and disorderliness of radar point clouds by proposing a translation-rotation invariant graph representation coupled with a virtual node message-passing neural network. Through invariant feature reconstruction, an enhanced message-passing mechanism, and residual connections, the method effectively strengthens feature propagation and global modeling capabilities while balancing model robustness with inference efficiency. Experimental evaluations on the RadarScenes dataset demonstrate that this approach outperforms state-of-the-art methods and significantly reduces computational and memory overhead. Consequently, this work establishes a novel paradigm for efficient radar-based object detection in autonomous systems.
📝 Abstract
Perception is a fundamental component of autonomous driving systems. While LiDAR-based methods have achieved remarkable progress in object detection, their reliability can degrade under adverse weather conditions. Radar point clouds provide a robust alternative due to their resilience to bad weather and low-illumination scenarios. However, radar point clouds are typically sparse, unordered, and less informative than LiDAR data, making it challenging to directly apply existing LiDAR-based perception methods. To address these challenges, we propose IRGNN, an Invariant Radar Graph Neural Network for radar point cloud object detection. IRGNN first reconstructs radar point clouds into graph representations using translation- and rotation-invariant feature designs, enabling robust modeling of sparse radar measurements. It then employs an improved message passing neural network (MPNN) with residual connections and a virtual node layer to enhance local feature propagation and global context modeling. Finally, task-specific heads are applied to the learned graph representations for object classification and bounding box prediction. Experimental results on the RadarScenes dataset show that IRGNN outperforms existing radar-based object detection methods and achieves competitive performance. In addition, IRGNN significantly reduces computational cost and memory usage during inference, demonstrating its effectiveness and practical potential for efficient radar-based perception in autonomous driving.