Urban green infrastructures are often inventoried in field, with huge economic costs. Nowadays, data sources such as LiDAR, satellite or Google Street View allow us to automatize these urban tree inventories, when technologies such as Computational Vision and Artificial Intelligence are used.
This experiment is based on an already functional tool that has been used for some cities in Spain. Nevertheless, the help from several EUHubs4Data partners will boost arböria by testing new geo-positioning methodologies or new techniques for image annotation clustering, among others.
The experiment is first focused on methodology. This way, new ways to improve tree positioning and avoiding false positives are tested, both using LiDAR and Google Street View.
Also, data bases of annotated tree images are trained to enhance the detection algorithms. With the methodology improvements, a proof-of-concepts will be developed, and the market readiness boosted. Finally, föra’s staff will upgrade its CV and AI skills.
To improve the current functionalities of arböria through AI and computational vision in terms of both methodologies and skills, together with improving its market readiness.
The main innovations are the creation of a dataset of 360-degree images from GSV where the trees are positioned and identified. The improvement of specific algorithms for the positioning and identification of trees based on LiDAR information and GSV images using computer vision will allow to develop a proof-of-concept to be carried out with a high degree of accuracy, improving arböria from experimental to a fully marketable solution.
arböria has clear social and environmental impacts. An automatic and reliable urban tree inventory is the first step to optimal management of those trees. Thus, better management of trees will lead to better health for both trees and citizens, since healthy trees create shadowed urban environments that prevent heat wave related deaths. The economic indirect impact associated to the environmental and societal benefits is also clear.
Our KER is a fully updated version of arbörIA, by improving its core algorithms to enhance its geopositioning skills and to be able to identify tree species using computer vision, as well as developing a proof-of concept that boosts its first steps into the market and to design an optimal commercial roadmap. Also, föra’s staff will upskill on AI-based image processing techniques.
- Improvement in results of geo-positioning of urban trees: reduction of false positives (2% in Month 5 and 5% in Month 10)
- Image datasets of tree species from GSV: number of annotated images from tree species: 2000 images of 1 species (Month 5) and 20000 images of 10 species (Month 10)
- Proof-of-concept results report
- Training/upskilling material