We are happy to present you Urban Monitoring – new product from GeoAlert, partner of Geomatrix. Do you want to get all the statistics on the location you are interested in? Then get more details here.
After long journey of experiments and datasets making we are launching our first brand new product “Urban Monitoring” for buildings recognition and classification in satellite imagery.
The idea we’ve got focused on through the references to Mapbox-Google-Microsoft-etc. works is to recognize all buildings in the world to collect more accurate information regarding populated areas, city patterns and related changes.
The point of our current work is an implementation of the model as a workflow into Geoalert platform that allows anyone to try building recognition in his own area of choice.
So the simplified view of the workflow is presented in the scheme below:
The immediate application of the workflow we would figure out is to begin the processing of all populated areas around the Globe, as it had been released in “Miscrosoft building footprints” available for USA and Canada. But it would be very easy if it could be done in a single workflow. In life we have a lot of challenges. To mention very few of them:
- There is a lot of difference between the buildings patterns in USA, in the old towns in Europe or in high-density housing areas in the Middle East — still a lot need to be done to train our models to perform the same quality in any urban territory;
But now we are close to remove the technological barriers of digital mapping: 1) it’s too long — it takes 10–15 days of the cartographer’s work to prepare map for one small city about 100 sq.km
2) it costs a lot: in traditional workflow you need to purchase imagery, to hire cartographers to digitize maps. Is it worth if you need to know how many households are in the specific area (but the statistical data isn’t available or you are not going to trust it)? And you need to know it faster?
The Geoalert Urban Monitoring provides you with the single-house-accuracy statistics based on what satellite can see in your area of interest.
- You have to define area with the rectangle tool (in the demo version only this type of tool is available and the area size is limited to 10 sq.km)
- To run the processing task you have to login (it’s OK to be logged in with your Google account)
- Press button and wait (or browse other tasks) to download the report after the green light appears.
As simple as that. And we have a lot of ideas of applications we aim to try our platform for and a lot of features we’d like to add in its professional version. Give us a feedback!
Besides the recognition of buildings features we trained our models to classify buildings by type — this is the tricky thing for Machine learning algorithms. There only three classes are presented in Demo but we have more under research in progress. The second tricky thing is a superposition of the building roofs and the building footprint which are not in the same place due to the satellite view angle. Why Microsoft called their dataset “building footprints”? Because they can get the aerial photo for the entire territory of USA and the aerial photos are generally taken as vertical as possible. But in the same time the oblique images from the satellite provide us with the opportunity to calculate buildings heights to get the estimations of number of storeys and the total numbers of living areas in the multi-apartment buildings. All in one single satellite image!
We called it “mapping”and “heighting”. More details of this part of the work will be published in the scientific paper.
One more thing for now. The world population is projected to grow and more than 68% is projected to live in urban areas by 2050. Can we detect changes? Where are the places with the highest probability to change the urban landscape? To find out how to answer this question in the details we started from the detection of construction sites by comparison of the imagery for last two years. We make the first results for several Russian cities available through the platform.
Once again — we are happy to announce our first version of Geoalert platform and look forward to making it useful for people and businesses. Thank you for reading this!