The goal of this project is to use computer vision classification and object detection to construct 3D building models from satellite imagery, extending the BlenderGIS plugin with more accurate buildings out of the box.

The SpaceNet challenges provide large quantities of labelled data that will likely be suitable for this project. For example: https://spacenet.ai/spacenet-buildings-dataset-v2/ provides labelled buildings, which should be suitable for training building segmentation.

• Research satellite imagery building detection algorithms, build an implementation that works for one country/area.
• Attempt to infer shape of roof from satellite image.
• Extend to be robust with any building types/styles.

• Algorithm to sort detected buildings into categories.
• Feature-based classification - roof type, materials etc.

• Extend BlenderGIS to feed satellite data into CV algorithms.
• Construct a few rudimentary Blender models based on building categories.
• Place basic building models on map with correct orientation and location.
• Improve building model construction - automated generation based on data from classification.