Nearly everyone is familiar with the cadastral maps that depict property lots. However, few know that these maps are only an approximation of reality. This does not mean that the Dutch Cadastre (Kadaster) does not know exactly where the boundaries are. They have been measuring and recording boundaries via five million field operations for around one hundred and fifty years now. Sioux’s mathware has created a breakthrough in digitalising this massive volume of complex geo data.

‘Creating an accurate cadastral map of the Netherlands is advisable, even just in terms of public perception’, states Rik Ebbeling, Kadaster’s manager of product and process innovation. ‘On top of that, it will increase our efficiency. Figuring out a specific field operation and then performing the necessary measurements in the field takes around eight hours currently. If a one-to-one map is available, we can speed up that process considerably, thanks to automated interpreting and immediate access to viable GPS coordinates. However, manually vectorising our field operations would cost a fortune. That is why we put out a Request For Information, asking whether automation could provide a solution. Eleven companies responded to our request, each of which suggested a partial solution. That made us feel hopeful. We thought that it would not be possible, but now we can see that there are opportunities out there.’

Sioux & Kadaster

Machine learning

One of the companies that responded to the Kadaster’s request was Sioux. Jeroen Franken, Senior Mathware Engineer at Sioux: ‘As a high-tech company, Sioux has immense know-how regarding image recognition and big data. Our mathematicians saw opportunities right away. Their scientific ability is immense, but they also know when it’s necessary to test complex matters in a practical manner. The Kadaster took up our recommendation and asked us to create a Proof of Concept in four months’ time. The only way to do that was to divide the problem into manageable chunks, enabling us to focus properly and not get lost in the complexity. We spent a few weeks working hard on a theme, before moving on to the next task; like removing the jpeg artefacts in scanned maps. Luckily, we already had the solution for that lying around. We also needed an algorithm that would be able to recognise different types of lines. On top of that, we had to find a way to recognise and read both horizontal and angled numbers. Machine learning is able to do that. Another considerable challenge was understanding the relationships between all of these lines and numbers, knowing which are important, how they relate to one another, and whether the input and output is logically sound.’

‘Sioux’s agile approach worked really well for us, because we were able to contribute to a good result’, Ebbeling adds. ‘We do not have the required technical expertise, but we do know which data is valuable, how to interpret it, and what we need exactly. By working together, we are able to bring out the best in each other. As such, Sioux has come to truly comprehend our issues.’

Sioux Rik Ebbeling

Human algorithm

After finishing its feasibility study, Sioux came to the conclusion – barring a whole load of ifs, ands & buts – that it should be possible to digitalise a single map in 18 minutes. The lion’s share of that time is taken up by necessary manual work. The project has since entered a new stage. A multidisciplinary team featuring employees from various organisations, including eight Sioux employees, is working on site at Kadaster on optimising the methodology, mathware, and software.

Ebbeling: ‘Digitalising our field operations would never be possible without the dedi-cation of the people who are responsible for colouring in the blind spots. Some maps are too unclear or incomplete or contain errors. Moreover, not all maps utilise the standard methodology for marking out lines and noting down numbers. And I could keep going, listing causes for our troubles. But 18 minutes is already a lot better than the 90 we started with. Sioux’s recommendation to organise the manual correction work per element and not per map, resulted in this potential time gain. They dubbed this methodology the ‘human algorithm’. Our aim now is to go even fur-ther in terms of speed and clarity regard-ing what can actually be achieved. We also need to find a solution for automatically linking digitalised maps together. Kadaster will be making its decision in April 2019. Whenever our people use maps, they are always digitalised. Depending on the final price tag, we may start doing the same for certain other areas or for the Netherlands as a whole.’ 

Source Magazine

This Source Magazine is full of ambition. You can download the entire magazine by clicking the button.

Download Source Magazine