In this talk we introduce use cases for feature extraction with aerial and satellite imagery such as turn lane marking detection, road and building footprint detection. In doing so, we show the potential of large-scale object detection and segmentation on aerial and satellite imagery.
We give insights into how imagery training sets can be built with little or no human annotations needed making use of available datasets such as OpenStreetMap.
We then focus on modeling aspects for object detection and segmentation. In doing so we give an insight into state-of-the-art detection systems and adaptions we had to make for the aerial and satellite imagery domain.
We conclude the talk with lessons learned in building these large-scale object detection and segmentation pipelines, and show potential for future work in this domain.