UAV-Based Radar–Camera Fusion for Robust Terrain Analysis

We present an efficient pipeline for the radar-camera fusion from an unmanned aerial vehicle (UAV). Using a commercial off-the-shelf millimeter-wave radar, the altitude of the UAV is estimated absed a synthetic aperture radar (SAR) algorithm. The SAR images serve as input to a lightweight machine learning (ML) model that accurately predicts the UAV’s altitude. To satisfy the computational and power constraints of UAV platforms, we adopt the polar format algorithm (PFA) as a low-complexity SAR processing technique. Leveraging DMs as high-level features enables the use of a simple yet effective ML model for altimetry. Furthermore, because GPS-based altitudes are referenced to a fixed point (e.g., the UAV’s take-off location), we compute a nadir depth profile by subtracting GPS altitude from radar-based estimates. The proposed method is validated through real-world UAV flight experiments, demonstrating its effectiveness for accurate radar-based altimetry and depth profiling. Additionally, the features of the SAR images are fused with their corresponding camera images to specify the terrain type.