RCM-Depth: Depth Estimation Based on Camera and Sparse Radar Expansion

Existing depth estimation methods fail to fully utilize the intrinsic features of radar data. To address this, we propose a two-stage network called RCM-Depth, designed to dynamically adjust radar point clouds for more accurate spatial representation. We dynamically expand the radar point clouds using intrinsic feature information, enhancing their density and accuracy to improve spatial representation. By leveraging a Transformer model and the RSA module, we capture deep relationships between radar point clouds and images, achieving high-quality multimodal data fusion. In the NuScenes benchmark, our method achieved a 17.42% reduction in mean absolute error (MAE) and an 18.67% reduction in root mean square error (RMSE), validating the effectiveness of our approach in improving depth estimation quality.