Image Stitching Of Aerial Footage

Ng, Wei Haen (2021) Image Stitching Of Aerial Footage. Final Year Project, UTAR.

Preview

PDF Download (3073Kb) | Preview

Abstract

With the advent of modern drones or unmanned aerial vehicles (UAVs), it is used in the application of infrastructure, agriculture monitoring, disaster assessment, etc. It has simplified and automated the site assessment and monitoring procedure. A lot of well-known image stitching software or applications, including Image Composite Editor (ICE), Adobe Photoshop and AutoStitch have been developed to allow users to stitch the images for monitoring or assessment purposes. However, the problems arise when the input data is aerial footage as these software are only taking images as input data. In this project, an image stitching framework is proposed to take aerial footage as input data. The proposed algorithm extracts the frames of the aerial footage and undistorts the bird-eye-effect of the images to remove the noises. ScaleInvariant Feature Transform (SIFT) approach is used to detect and describe the feature points of the extracted frames. The randomized k-d tree of FLANN matcher is utilized to match the feature point pairs between the images. The Lowe’s ratio test is applied to discard the mismatched point pairs. RANSAC is exploited in the homograhy estimation to calculate the corresponding homography matrix and remove the outliers. The images are warped to the key frame of the footage to generate a stitched image by using the computed homography. The algorithm performance is evaluated using the Orchard datasets, consisting of L-shape flight pattern and lawnmower flight pattern. The implemented method successfully stitched the frames extracted from the aerial footage to generate a large scene image beyond the normal resolution.

Actions (login required)