A Novel Stitching Method for High-Precision Low-Overlap Thermal Infrared Array Sweeping Images
Xiangbo Jin, Chongru Wang, Guicheng Han, Yueming Wang, Jianxin Jia · IEEE Transactions on Geoscience and Remote Sensing · 2023
Automatic image stitching and high-precision photogrammetry have been active research fields. This article proposes a novel stitching method for high-precision, low-overlap thermal infrared array scanning (TIRAS) images. The approach corrects the TIRAS camera’s internal parameters, external orientation (EO) elements, and planar projection errors by calculating three homography matrices. First, we establish collinearity equations using the calibrated internal parameters and EO elements with errors and further compute the direct projection transformation model of the original image. Then, we propose a global error optimization model based on a fixed framework according to the midpoints of the alignment point pairs of the correcting image and neighboring images. The homography matrix calculated by this model can relatively correct the EO element errors in the directly projected image. Finally, we calculate the homography matrix for the planar projection error correction using the universal transverse mercartor (UTM) coordinates of geometric targets and their corresponding image coordinates and use the matrix to perform absolute error correction on the stitched wide-field image of the reprojected image. Comparing the proposed method with AutoStitch, Liu’s, and PTGui for mid-wave infrared (MWIR) image stitching results, it exhibits superior image stitching quality, without noticeable distortion or stretching, and with intact object structures. Besides, our method demonstrates outstanding positioning accuracy. Without ground control points, the optimal positioning accuracy is 0.7538 m (RMSE), less than 1 pixel. In conclusion, the proposed stitching method showcases stable, automated, and highly accurate global error optimization capabilities, and it is also suitable to splice array remote sensing images equipped with a high-accuracy POS system.