Enhancing MIMO Radar Data Quality via Stereo-Camera-Guided Diffusion Models

MIMO radar systems are recognized for their robustness in low-visibility and adverse weather conditions, yet they face challenges such as low spatial resolution and incomplete data capture, particularly in static regions. To address these limitations, this study introduces an innovative framework designed to enhance the quality of MIMO radar data by integrating super-resolution and inpainting techniques utilizing stereo camera data as ground truth. The proposed approach includes sensor fusion system that spatio-temporarily synchronize heterogeneous data, integrated inpainting and super-resolution model, and optimized masking strategy to restore missing static regions. Quantitative evaluations demonstrate that our model substantially outperforms existing baseline methods, achieving a Peak Signal-to-Noise Ratio (PSNR) of 13.7232 dB, an Structural Similarity Index (SSIM) of 0.7597 and a Learned Perceptual Image Patch Similarity (LPIPS) score of 0.1814, indicating a marked enhancement in both the detail and integrity of radar images. These advancements extend the utility of MIMO radar across various applications including autonomous driving, object detection, and human activity recognition, providing a foundation for advancing radar-based spatial sensing technologies.