Building Footprint Extraction

Building footprint extraction (BFE) aims to automatically identify and delineate the ground-level boundaries of buildings from aerial or satellite imagery, crucial for urban planning and other applications. Recent research emphasizes improving accuracy, particularly in densely built areas and off-nadir imagery, using deep learning models such as encoder-decoder networks (including U-Net variations), graph convolutional networks, and novel architectures designed for polygon prediction and offset vector learning. These advancements leverage techniques like self-supervised learning, super-resolution, and prompt-based methods to enhance performance and address challenges related to data scarcity and complex building arrangements. The resulting improvements in BFE accuracy have significant implications for various fields, including urban modeling, disaster response, and infrastructure management.