Introduction

3D object detection is becoming an indispens-able functional module in environmental perception forautonomous driving recently. LiDAR-based methods havemade remarkable progress on accuracy; however, pointclouds often fail to distinguish objects with similar struc-tures, which leads to false detection. Therefore, cameraand LiDAR fusion is naturally considered a solution dueto the rich texture information of camera images. Never-theless, current fusion methods are either limited to poorprecision or efficiency. To this end, we propose a plug-and-play module named RI-fusion to achieve the effectivefusion of LiDAR and camera, and the module can be easilyaccessed to existing LiDAR-based algorithms. The pointcloud is converted into a compact range view representa-tion through the spherical coordinate transformation. Andthen, the range image is integrated with the correspond-ing camera image based on an attention mechanism. Theoriginal range image is concatenated with fusion featuresto retain the point cloud information, and the results areprojected to a spatial point cloud. Finally, the feature-enhanced point cloud can be fed into a LiDAR-based 3Dobject detector.