Concept for Automatic Annotation of Automotive Radar Data Using AI-Segmented Camera and Lidar Reference Data

This work presents a concept for automatic annotation of automotive radar data by combining AI-segmented camera images and 3D lidar point clouds. The AI-interpreted images are used to detect objects or features, which are then localized within the lidar point cloud. This enables a precise labeling of various radar data representations. The concept’s effectiveness is validated on a test vehicle equipped with an automotive radar and off-the-shelf camera and lidar sensors. As this concept relies on standard processing techniques, it integrates seamlessly into other setups and is beneficial along the complete radar development chain.