A breakthrough: automatic wall extraction from satellite imagery

We developed a new method for automatic wall extraction from satellite imagery. The process provides professionals in multiple sectors with valuable intelligence.

At a glance:

• LuxCarta has developed an effective new process for automatic wall extraction from satellite imagery
• It efficiently identifies walls and fences and turns them into vector layers
• Learn how our process works 
• Find out how this breakthrough benefits multiple sectors

In recent years, many advances have been made in extracting and mapping information about urban environments from satellite images. Using various forms of artificial intelligence, you can now generate detailed maps with separate vector layers for things like building footprints, roads, vegetation, and other classes of information. 

But there’s one extremely common feature of urban environments that, so far, have been missed out: walls and fences. These boundaries are an integral part of any city, town or village. Yet there has been no way to automatically extract walls and fences from satellite images - until now. 

In July 2024, LuxCarta presented a paper at the IGARSS conference describing our new process for automatic wall extraction from satellite imagery. How did we do it? And how does it help you? 

A new process for automatic wall extraction from satellite imagery

In the paper we presented at the IGARSS conference, we described our process in detail - you can read more about it here. 

As far as we are aware, there has only been one previous attempt to automate wall extraction from satellite imagery. And that was a project for a single city in one country - whereas our approach is intended to be universal. 

So, how did we do it? 

• Imagery: We collected very high-resolution images (30cm/pixel) from the Pleiades Neo satellite array.

• Training locations: We selected 54 cities from 26 countries to train the algorithm on different wall/fence types in multiple locations. 

• Building and wall extraction: We used our deep learning models to identify and extract buildings and walls.

• Vectorization: We then transformed the deep learning results into separate vectors (polygons for buildings and lines for walls) so they can be readily integrated into maps 

• Regularisation: We developed computational geometry techniques to automatically regularise the shape and location of walls/fences. The algorithm ‘snaps’ boundaries to buildings they come into contact with, closes small gaps, and ‘cleans’ dangling edges to ensure coherence between the buildings and walls layers.

We tested our wall extraction on 1 km² areas in three cities in Yemen, Saudi Arabia and South Africa. We compared our algorithm against a ‘manual’ ground truth analysis of wall location. 

The results were encouraging:  the automated solution achieved a precision of 80.31% and recall of 86.32%.

Our building extraction was tested on 30 areas around 1 km² each, spanning 21 cities and 13 countries. It achieved a precision score of 87.67% and recall of 88.44%.

How does automatic wall extraction from satellite imagery benefit you?

We believe that our new technique for automatic wall extraction from satellite imagery is unique on the market today. Multiple sectors could benefit from this kind of detailed information about boundaries in urban areas. Here are just some ways it could help cities, governments, citizens, and private companies.

Illegal construction prevention

Illegal land grabs and construction in and around cities - particularly in developing countries - is a major challenge. Our technology can help city governments to monitor the appearance of walls and buildings and identify illegal construction.

Tax authorities 

Individuals and businesses in many countries are taxed on the land they own. Automatic wall extraction from satellite imagery allows tax authorities to accurately identify the land people are claiming as theirs (since you wouldn’t put a wall around something you don’t own). This can help with revenue collection and crack down on tax avoidance. 

Military intelligence

Knowing the precise location of walls and fences can be very valuable for military intelligence. This information can support mission planning, with detailed insights into possible obstacles or hideout. It can also help with monitoring the layout and development of enemy positions to help discover possible weaknesses. 

Market research

An understanding of the layout of walls in urban areas is insightful for market research purposes for a wide variety of leisure, sports and outdoors businesses. For example, by knowing that a new housing estate has many properties with large gardens, a gardening center business might choose to open a retail outlet close to this market. 

Property development firms

Similarly, property development firms can gather insights by using automatic wall extraction from satellite imagery. They can, for example, discover large plots of land with the potential for building apartments, ‘garden offices’, outhouses and similar small scale developments. 

Telecommunications companies

As telecoms firms roll out infrastructure for various kinds of IoT (Internet of Things), it is valuable to know about the location of all possible obstacles, including walls and fences. These obstacles may reflect or disrupt some types of 5G signals - knowing their location can be very helpful when planning base station and beacon locations. 

Simulations

For any firm that creates simulations of real world environments, automatic wall extraction from satellite imagery can be very helpful. Using maps with walls and fences geolocated on them can add crucial detail to simulations. Knowing there are walls or obstacles in a place can have a significant effect on the look and feel of a simulated environment. 

Digital twins

Any organization that requires a digital twin of an urban area will, of course, benefit from automatic wall extraction from satellite imagery. A digital twin that includes all walls, fences and barriers will provide a great level of depth and understanding. 

Agricultural research and intelligence

Automatic wall extraction from satellite images can also help with agricultural research and market intelligence in parts of the world where field boundaries are marked with walls or fences. Identify new animal pens on farms (e.g. for pork or chickens), estimate potential yield from arable fields based on size, assess claims for government grants based on land usage - and more. 

Try it for yourself in BrightEarth

You can soon experience our process for automatic wall extraction from satellite imagery for yourself in BrightEarth. Sign up to our Early Adopter program, and you will be able to select any region on the planet’s surface, choose the information you want BrightEarth to extract, and then generate vector layers for download.  

Beyond automatic wall extraction from satellite imagery

We believe our approach to automatic wall extraction from satellite imagery is genuinely unique on the market today. No other solution can automatically (and accurately) plot and extract walls and fences from satellite images, and turn them into a separate map layer. 

And we’re only just getting started. The LuxCarta team are working on even more advanced techniques that can help differentiate between fences and walls. We’re also hoping to include wall/fence height estimates in future. 

Want to try it for yourself? Sign up to BrightEarth’s Early Adopter program today.