Dense Matching Method for UAV SAR Images Without Epipolar Rectification

Dense matching is an important step in radargrammetry. As approximating epipolar lines in synthetic aperture radar (SAR) images is difficult, conventional dense image matching (DIM) algorithms are unsuitable for these images. This study proposes a DIM algorithm for unmanned aerial vehicle (UAV) SAR images that does not require epipolar rectification. The proposed algorithm uses tie-point matching results to construct a search window for corresponding points and utilizes an improved DAISY descriptor incorporating the ratio of exponentially weighted averages (ROEWAs) operator for cost calculation, which suppresses errors caused by speckle noise. Cost aggregation was performed using computationally efficient superpixel segmentation and the guided filter (GF) algorithm, and the winner-takes-all (WTA) strategy was applied for dense matching. Finally, experiments were performed on six pairs of UAV SAR images containing different terrains and ground objects, and an average root mean square error (RMSE) of 4.3 pixels was obtained, demonstrating that the proposed method is superior to conventional DIM algorithms and has excellent precision and accuracy.