The city is constantly growing and changing. Information about public areas requires regular updates: a forest may have been cleared for a new residential area, and a former wasteland may have turned into a park that needs maintenance.
The city maintains this information in a register of public places, which is updated by comparing registry data with the latest aerial images. However, manually going through large areas and thousands of aerial images is slow, and insufficient time is available to do it adequately. As a result, registry data is not always up to date.
AI can now help identify objects in aerial images, as revealed in the AI experiment conducted by the City of Helsinki and Digia. The experiment was carried out as part of the city's Kokeilukiihdyttämö (“Experiment Accelerator”), which brings the city's employees’ digitalization ideas into agile experimentation.
AI Completed Weeks of Work in 10 Minutes
The primary purpose of the AI experiment was to test whether the manual image recognition process could be automated with AI, thereby speeding up the review of aerial images.
"Information about the areas has been collected since the early 2000s, and there is a lot of it, over 480,000 records. Distinguishing a forest from a lawn is still relatively easy, but a stream flowing at the edge of a park, grass under trees, or plantings require more attention," says Ritva Keko from the City of Helsinki.
The assumption was that AI should identify objects with 80% accuracy.
"The results surprised even the professionals. As expected, AI sped up the process: AI completed work that took humans two weeks in ten minutes. But it also exceeded expectations in accuracy, correctly identifying about 87% of the objects," says Hilla Tilhi, Technical Consultant at Digia, who conducted the experiment.
Goal: Even 100% Up-to-Date Registry Data?
When AI identifies potential erroneous objects and alerts about aerial images it cannot fully confidently recognize, humans can focus on correcting erroneous information and checking unclear images. The result could be even 100% up-to-date registry data. The improvement is significant, as it is currently estimated that 10-20% of the information in the public areas register is outdated or otherwise incorrect, depending on the area.
Up-to-date information about public areas is the basis for many different purposes: maintaining green areas in summer, planning snow removal in winter, and so on. The experiment sparked numerous new ideas for utilizing AI and up-to-date information.
"We learned a lot about the possibilities of AI. The results were very encouraging, and we want to continue developing the method for implementation. Suppose all the time spent on improving data accuracy can be used for correcting data instead of searching for errors. In that case, we can significantly improve the quality of our data assets," says Ritva Keko.
Read more in the City of Helsinki's Kokeilugalleria ("Experiment Gallery") (in Finnish) >>
Read more: Digia's AI Proof-of-Concept for Testing AI Ideas (in Finnish) >>
More information:
Minna Häkämies
Head of Data Insight, Digia
tel. 040 516 6108
minna.hakamies(a)digia.com
Questions about the AI experiment related to public areas:
Ritva Keko
Project Manager, City of Helsinki
tel. 050 364 9034
ritva.keko(a)hel.fi
Questions about Kokeilukiihdyttämö in general:
Ville Meloni
Project Manager, City of Helsinki
tel. 040 026 0000
ville.meloni(a)hel.fi