Project: Building Digital Surface Models exploiting the shadows from satellite and aerial RGB images

Description: Aerial and satellite images are widely used in geographic information systems (GIS) for a plethora of interesting tasks, such as: urban monitoring and planning, agricultural development, landscape change detection, disaster mitigation planning and recovery as well as aviation. However, these images are predominantly two-dimensional (2D) and constitute a poor source of three-dimensional (3D) information, hindering adequate understanding of vertical geometric shapes and relations within a scene. Estimating the heightmaps of buildings and vegetation in single remotely sensed images is a challenging problem. Effective solutions to this problem can comprise the stepping stone for solving complex and demanding problems that require 3D information of aerial imagery in the remote sensing discipline, which might be expensive or not feasible to require.

3D info is often obtained with a Light Detection and Ranging Laser Scanner (LiDAR), structure from motion or by using stereo image pairs. These techniques suffer from poor performance and have high cost, due to the need of a flight mission equiped with sophisticated instruments. Height estimation from a single image and monocular vision is a difficult problem, because mapping 2D pixel intensities to real-world height values is a challenging task.

We propose a task-focused Deep Learning (DL) model that takes advantage of the shadow map of a remotely sensed image to calculate its heightmap. The shadow is computed efficiently and does not add significant computation complexity. The model is trained with aerial images and their Lidar measurements, achieving superior performance on the task. We validate the model with a dataset covering a large area of Manchester, UK, as well as the 2018 IEEE GRSS Data Fusion Contest Lidar dataset. Our work suggests that the proposed DL architecture and the technique of injecting shadows’ information into the model are valuable for improving the heightmap estimation task for single remotely sensed imagery.

Techniques used: Deep learning, aerial photography, satellite imagery, modelling, GIS, LIDAR

Started: October, 2020

Status: Finished

Publications: Savvas Karatsiolis, Andreas Kamilaris and Ian Cole, IMG2nDSM: Height Estimation from Single Airborne RGB Images with Deep Learning, In Remote Sensing Journal, vol. 13, no. 12, pp. 2417, June 2021.