ROI-Optimized KAZE Image Stitching Technology for UAV Reconnaissance

In the fields of military reconnaissance and remote sensing detection, visible light image stitching technology has become increasingly important due to the limited field of view of single-frame images captured by UAVs at high altitudes, which fails to meet the needs of large-angle and large-area reconnaissance imaging. However, existing traditional algorithms generally suffer from the problem of low computational efficiency or insufficient stitching accuracy in UAV visible light image stitching scenarios, making it difficult to balance both. Therefore, this thesis aims to propose a fast stitching method for UAV high-altitude visible light images, which can significantly improve the stitching speed while maintaining the original stitching quality. This thesis proposes a new improved method based on the KAZE algorithm: introducing the combination of Region of Interest optimization and the RANSAC algorithm to solve the problems of slow stitching calculation speed while satisfying image stitching quality. Simulation experiment results show that the improved ROI-KAZE algorithm can achieve higher matching accuracy and reduce stitching time compared with the traditional KAZE algorithm, and the time used under scale transformation is lower than that of the traditional KAZE algorithm; the test results of Structural Similarity Index and Peak Signal-to-Noise Ratio of the stitched images also perform well. The research results prove that the ROI-KAZE-based image stitching algorithm has spatial and scale optimization effects, ensures the original image quality after stitching, and improves the operation efficiency of the stitching algorithm.