LiRaFusion: Deep Adaptive LiDAR-Radar Fusion for 3D Object Detection

LiRaFusion is a technique proposed to improve 3D object detection in autonomous vehicles by fusing LiDAR and radar data. It combines an early fusion module for joint voxel feature encoding, and a middle fusion module to adaptively fuse feature maps via a gated network. The study highlights the complementary information of LiDAR and radar and demonstrates how LiRaFusion leverages this for improved object detection. The research was extensively evaluated on the nuScenes dataset.

Publication date: 20 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.11735