Leveraging Photogrammetry for Efficient Multiview Image Labeling
Li Tang, Nicholas Maxwell, John Papadakis, S. Craig Stutts, Benjamin L. Raskob · 2023
We present a method for efficient image-segmentation labeling of Very High Resolution (VHR) multi-view overhead imagery, applied in the original non-orthorectified views. We forward-project orthographic space labels into aligned images using a registered VHR Digital Surface Model (DSM). We demonstrate that this method can improve the accuracy of a building segmentation network using minimal manual effort, making it possible to create large scale datasets to train learning-based approaches for various semantic segmentation tasks with satellite imagery.