IMG2nDSM: Height Estimation from Single Airborne RGB Images with Deep Learning

Estimating the height of buildings and vegetation in single aerial images is a challenging 12 problem. A task-focused Deep Learning (DL) model which combines architectural features from 13 successful DL models (U-NET and Residual Networks) and learns the mapping from single aerial 14 imagery to a normalized Digital Surface Model (nDSM) is proposed. The model is trained on aerial 15 images whose corresponding DSM and Digital Terrain Maps (DTM) are available and is then used 16 to infer the nDSM of images with no elevation information. The model is evaluated with a dataset 17 covering a large area of Manchester, UK, as well as the 2018 IEEE GRSS Data Fusion Contest LiDAR 18 dataset. The results suggest that the proposed DL architecture is suitable for the task and surpasses 19 other state-of-the-art DL approaches by a large margin.