Night Scene Image Stitching and Image Recognition Based on Improved SIFT

Image stitching is an important task in the field of computer vision, aiming to automatically merge multiple overlapping or adjacent images into a seamless panoramic image. This paper proposes an improved and fast image stitching method specifically designed for unmanned aerial vehicle (UAV) nighttime aerial photography in high-altitude environments. The method is implemented using the Python programming language and the OpenCV library. First, mean filtering is applied to the original images to reduce noise and minimize the impact of detail differences on the stitching result. Subsequently, the Scale-Invariant Feature Transform (SIFT) algorithm is used for feature extraction and matching to identify corresponding feature points between different images. To eliminate incorrect matches, the Random Sample Consensus (RANSAC) algorithm is employed for robust estimation and model verification, ensuring the accuracy of the stitching result. Finally, a weighted fusion algorithm blends the two images, considering the contribution weights of each pixel, resulting in a seamlessly stitched panorama without noticeable artifacts. Additionally, leveraging the capabilities of the YOLOv5 object detection algorithm, this paper achieves the recognition and precise localization of high-rise buildings in nighttime urban scenes. Experimental results demonstrate that the proposed method achieves a feature extraction speed improvement of at least 54% compared to the traditional SIFT algorithm. Moreover, by integrating the YOLOv5 object detection algorithm, accurate localization and recognition of nighttime urban high-rise buildings are achieved, further enhancing the functionality and application value of image stitching. This paper provides a reliable solution with broad prospects for the application of nighttime aerial photography.