Building Height Estimation from RGB Satellite Imagery

Information regarding a building?s height is really valuable for a wide range of problems.This information could be used to determine the tax bracket applied to a particular building. Height of buildings could also help us reach a fair estimate of the energy consumption of a particular build

Project Title

Building Height Estimation from RGB Satellite Imagery

Project Area of Specialization

Artificial Intelligence

Project Summary

Information regarding a building’s height is really valuable for a wide range of problems.This information could be used to determine the tax bracket applied to a particular building. Height of buildings could also help us reach a fair estimate of the energy consumption of a particular building. The most obvious answer to obtaining the height of a building would be through field work. However, this method would require people to visit the site of the particular building, which is a time consuming and expensive process, and hence is not feasible to carry out this process on a larger scale. Other approaches involve estimating building 3D models using LiDAR data or Digital Surface Model but these might be expensive or might not even be available in certain scenarios. To cater to this problem, in our project, we hope to use RGB satellite imagery (obtained through sources such as google earth) to determine the height of a number of buildings present in the satellite image. The idea behind our planned process is to have two Deep Neural Network Models, one for building segmentation, and the other for shadow segmentation. Using the obtained shadow and building information, and angles and positions of the sun and satellite, we should be able to extrapolate the height of a building. 

Project Objectives

Obtain actual height of building in meters using satellite image. 

Differentiate between ground and not-ground from a satellite image of residential area. 

Determine the number of stories of a building. 

Obtain Building height using stereo method and/or shadow method.

Correlating mathematical formulation to obtain building height from satellite imagery and Deep learning methods. 

Project Implementation Method

In our method, we aimed to extend the methodology employed by Feng Qi et al. The process outlined by Feng Qi involves a large amount of manual collection of information from google earth images. As a result, this can lead to a number of human errors. For this reason, we try to automate the process of selecting building / shadow lengths on a 2D satellite image. 

The method of height estimation we have devised is divided into three steps. 

1. Building segmentation

2. Shadow segmentation

3. Height estimation

The Building Segmentation step employs a neural network trained to create segmentation masks for buildings in a satellite image. For our purposes, we need multiple / different segmentations for the rooftop, the building footprint, and sides of the building. The proposed neural network chosen for this task was a ResNet101 based Unet model pre-trained on the ImageNet dataset. 

The Shadow Segmentation would use another neural network to create segmentation masks for building shadows. For our purpose, we would require off-nadir images of satellites in which shadows are visible in the satellite images.

After we have both of these segmentation masks, we can use satellite’s metadata and our mathematical formulae, to estimate the height.

F. Qi, J. Z. Zhai, and G. Dang, “Building height estimation using Google Earth,” Energy and Buildings, 03-Mar-2016. [Online]. Available: https://www.sciencedirect.com/science/article/abs/pii/S0378778816301025. [Accessed: 29-Apr-2022].

G. Liasis and S. Stavrou, “Satellite images analysis for Shadow Detection and building height estimation,” ISPRS Journal of Photogrammetry and Remote Sensing, 11-Aug-2016. [Online]. Available: https://www.sciencedirect.com/science/article/abs/pii/S0924271616301939. [Accessed: 29-Apr-2022].

H. A. Amirkolaee and H. Arefi, “Height estimation from single aerial images using a deep convolutional encoder-decoder network,” ISPRS Journal of Photogrammetry and Remote Sensing, 24-Jan-2019. [Online]. Available: https://www.sciencedirect.com/science/article/abs/pii/S0924271619300139. [Accessed: 29-Apr-2022].

H. Hao, S. Baireddy, E. Bartusiak, M. Gupta, K. LaTourette, L. Konz, M. Chan, M. L. Comer, and E. J. Delp, “Building height estimation via satellite metadata and Shadow Instance Detection,” SPIE Digital Library, 12-Apr-2021. [Online]. Available: https://www.spiedigitallibrary.org/conference-proceedings-of-spie/11729/117290L/Building-height-estimation-via-satellite-metadata-and-shadow-instance-detection/10.1117/12.2585012.full?SSO=1. [Accessed: 29-Apr-2022].

Benefits of the Project

From this project, we can estimate the number of buildings in an area and all of their respective heights. From the height, we can estimate the number of floors in one building. This can be helpful for the government as they can automate their process for determining the tax bracket applied to the building, instead of manually calculating the amount of area covered by one building. Manual labor won’t be required anymore since there would be no need for sending people to manually calculate the height of every building thus saving up money.

We can determine the amount of residential buildings and number of stories in each residential building. That can help in town planning as it will help in dividing up the land resources efficiently.

Once we have determined the height of a building and/or the number of stories of a building, we can then estimate the number of people residing in that particular building allowing automatic population census. 

Technical Details of Final Deliverable

End to end deep learning method for generating heightmap from a satellite image. 

Interactive software for determining height using manual user input, (non automated)

Provide an algorithm, software application which can calculate height of buildings

Provide and interactive software that can take user input, and automatically provide the relevant information as outlined in previous sections (building height, number of stories etc) 

Final Deliverable of the Project

Software System

Core Industry

IT

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure, Sustainable Cities and Communities

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com