An Improved ORB Algorithm for the Unmanned Aerial Vehicle (UAV) Image Stitching Task

Based on the image chunking processing theory, this paper proposes an improved ORB-based UAV image stitching algorithm that takes the pyramidal optical flow characteristics into consideration to overcome the weaknesses of the traditional algorithm (proneness to feature point aggregation, uneven distribution, and low feature matching accuracy). Firstly, the ORB algorithm is used to detect the feature points using the image chunking method of constructing moving masks. Subsequently, the non-maximum suppression method is used to reject the feature points aggregated in each mask, and the Hamming distance is used for feature matching after traversal. Following the step, the pyramid optical flow method is used to correct the motion displacement vector of the feature points and reject the mis-matched pairs after matching. Finally, the RANSAC algorithm is used to filter the redundant pairs to further improve the accuracy, and then the image is stitched together using the optimal stitching seam strategy and the fade-in-and-fade-out fusion algorithm. Overall, the aforementioned procedure increased the alignment rate to approximately 97% in the alignment stage, achieving a better alignment accuracy and a more accurate stitching effect.