Camera-Radar Fusion
Overview
Camera-radar fusion studies how to combine detailed image information with radar measurements that provide geometry and motion cues and can remain useful in poor visibility. Common approaches align features in bird’s-eye view, use cross-attention or learned queries to associate image regions with radar returns, and add temporal fusion, depth estimation, or calibration modules. Methods are evaluated on driving benchmarks such as nuScenes, View-of-Delft, TJ4DRadSet, and K-Radar, for tasks including 3D detection, depth, segmentation, and occupancy, with increasing attention to adverse weather and long-range settings. A central challenge is balancing cameras’ rich but weather-sensitive cues against radar’s sparse, sometimes noisy measurements while maintaining reliable calibration, generalization, and real-time efficiency.
Latest papers
49 papers4 papers in the last four weeks, down 20% on the four weeks before. 0.0% of all new papers.