Multi-Pass Automotive Synthetic Aperture Radar Image Fusion

In this work we demonstrate an efficient technique for the registration and compositing of multiple automotive synthetic aperture radar (SAR) images as a step towards distributed SAR imaging. Through the use of an efficient sub-pixel Fourier-based iterative correlation and refinement technique, sub-sections of the overall SAR images are aligned. Because only small sub-regions of the full SAR images are aligned at a time, this technique inherently compensates for accelerations experienced when sampling the synthetic aperture, by assuming a quasi-constant velocity over small sub-regions of the image. Because the image fusion is performed after image formation, this technique lends itself well to a distributed architecture where the image formation is performed on-vehicle and transferred to other vehicles or a centralized cloud processor where images from other vehicles can be aggregated. Finally, because the registration of sub-regions of the SAR images are performed, the registration alignment vector can be used as a odometry provider to improve vehicle odometry estimates; if a metrology grade SAR map is used as reference, a globally reference position estimate could be performed.