16 October 2020

Model 5621

Machine Learning Platform for Urban Planning

Model 5621 explores how the newest technology can help children vocalise their ideas and become part of the urban development process of their own neighbourhood. Woensel-West, a district in Eindhoven, is undergoing a transition with noticeable effects on the scenery and district layout. This brings an opportunity for novel urban planning to the area.

We, a collaborative design-duo of Daan Rongen and Jasper Bunschoten, believe that involving local residents in this process is a crucial factor in the pursuit of liveability. We aim to kickstart the discussion about future plans for this area by calling on to the childlike fantasy within the local residents, and in particular: children. This platform translates neighbourhood ideas, in the shape of LEGO scale models, into realistic-looking satellite images.

Satellite imagery generated by our model that was trained on satellite images of Eindhoven
Satellite imagery generated by our model that was trained on satellite images of Eindhoven

Computer Vision

The colours in this scale model resemble the following concepts: yellow stands for residential areas, blue for water, green for grass (parks), and grey for infrastructure (roads). The webcam analyses your scale model and interprets it using computer vision. This generates a color indexation as processed blueprint that our Machine Learning model recognises.

Your blueprint is ran through a convolutional GAN. This machine learning model has been trained with 1.000 satellite images from Eindhoven. The platform applies this model to your LEGO blueprint to generate a new realistic-looking satellite image. The output and result of this process is a satellite image of your ideal neighbourhood. Planning your dream neighbourhood becomes as easy as building a LEGO model.

The interface of the web-app, showing from left to right: the camera input, the computer vision analysis, a loading component to indicate machine prediction, and the final generated satellite image
The interface of the web-app, showing from left to right: the camera input, the computer vision analysis, a loading component to indicate machine prediction, and the final generated satellite image