Inferring the height map of aerial images

The Shadows2DSM project harnesses deep learning to infer height maps from aerial RGB images, enhancing remote sensing applications and providing detailed height data for diverse landscapes without the need for auxiliary information.

The primary objectives of the Shadows2DSM project were to:

• Develop a deep learning model to predict height maps from aerial RGB images.
• Train the model using imagery from the Manchester area and the University of Houston campus.
• Evaluate model accuracy against ground truth data and LiDAR technology.
• Explore the potential of the model in predicting height maps in varying landscapes with different architectural features.
• Improve model robustness by enhancing its reliance on shadows for accurate height predictions.