An algorithm built to stitch drone aerial imagery coined "QuickMap". The program runs locally on iOS (written in Swift, Objective C++). QuickMap is a feature I worked on in the summer of 2018 & 2019 at 3DR which is currently integrated into BETA SiteScan. The algorithm itself was built using OpenCV and C++. QuickMap uses the SIFT descriptor algorithm to localize keypoints and generate affine warp transformations that neatly align images. It uses GPS points to fill in missing images and uses multi-image referencing to reduce accumulated error. Once finished stitching, it blends the result and finds an appropriate geographical transformation to display it on a map.
It's recommended to use the latest version of OpenCV (4.1.1) with the contribution modules. You can download it here. Alternatively, you can build it yourself, however the former is much easier. Once downloaded, add it to Aerial-Image-Stitcher/QuickMap/OpenCV
GPS Points are returned upon generation of the map. To display, I suggest using MapBox via Cocoapods a developer access key must be added to your info.plist configuration to work properly. You can then display the image as a raster overlay.
The stitcher accepts images at any resolution (it will automatically resize), however there must be GPS encoded in the exif data. It's highly recommended that images are shot 150ft above with minimal distortion and 30-80% overlap. Some good data sets to test:
Once downloaded, add the JPG image folder to Aerial-Image-Stitcher/Images. In StitcherViewController.swift specify the dataset path, then build.
Museo Ambar data set - 54 Images shot ~200ft altitude.
The stitcher was heavily inspired by A Real-time Stitching Algorithm for UAV Aerial Images by Peng Xiong, Xianpeng Liu, Chao Gao, Zan Zhou, Chunxiao Gao, and Qiongxin Liu. Multiple image referencing or "predicted region matching" is a key technique integrated in the algorithm that increases geographical accuracy and reduces accumulated error.