1 | Concept

What makes everyday weather so interesting that it’s one of the top small talk topics - is it’s nature of changing the atmosphere of our surrounding. It creates a cities and a moments atmosphere. That’s what led to this experimental project which was developed during the semester course [Deconstructing Clouds](https://fhp.incom.org/workspace/7239 „Deconstructing Clouds“).

Inspiration

• The project [Most Blue Skies](http://www.autogena.org/mbs.html „Most Blue Skies“) from the artists group [autogena](http://www.autogena.org/ „Autogena Portfolio“): The approach was to calculate the location of the most blue piece of sky in real time.

• [Sky Over Berlin](http://skyoverberlin.com/ „Sky Over Berlin“): As an homage to the beauty of the sky the artist [Ewa Tuteja](www.tuteja.info „Ewa Tuteja“) made a picture of the sky every day for one year and uploaded it on an [Instagram feed](https://www.instagram.com/skyoverberlin16/ „Instagram – Sky Over Berlin“).

Why looking out of the window isn’t an option

Sure, you can go outside or monitor and sense the environment yourself with the help of tools and applications like the movement [»Citizen Sensing«](https://citizensense.net/citizen-sensing-smart-city/ „Citizen Sensing“) is doing it.

I wanted my attempt to sense the sky to be more automated and experimental at the same time. How to map the sky over course of a day? And how can you compare the mood of different locations?

2 | Get the sky

Access to webcams all around the world

[Webcams.travel](https://de.webcams.travel/ „Webcams.travel“) is a website where you can checkout webcams all around the world. Besides that they’re providing an API for each of these cams so you can fetch their images, embed them on your own site or use them otherwise. Perfect for this project. Suddenly you have eyes all around the world. 64.499 at the moment of writing this documentation to be exact. So we have the webcam images of cities. But how do we extract only the sky and it’s »mood« from these images?

Extract the sky

Superpixels, wow … that does sound impressive! But rather than being a better or improved version of a pixel, superpixels are a collection of pixel clustered by similarity of certain parameters. There are different algorithms to segement an image in superpixels. An implemantation of the SLIC Superpixel algorithm ([scikit-image](https://github.com/scikit-image/scikit-image/blob/master/skimage/segmentation/slic_superpixels.py#L13 „scikit-image“)) was used in this project.

Determine the color

The next step would be to get as few segments as possible. The input is always the original image. In the end we basically just want the sky and everything else separated from each other. With an average of 5 segments (image on the right) I was able to get the best result for the most images I was trying out. More or less superpixel led to less exact outcomes. The middle image has 100 segments.

Finally we calculate the average color out of the color value for each pixel in the segmented areas.

Map the sky

For the working prototype the calculated colors of the sky where updated every hour and a gradient between the current and previous one was created.

3 | What’s next

At the moment »Fetching Sky« works as an art installation to look at. The next step would be the function to dynamically add cities to the stream. In this way it would become interactive and customizable for the recipient.

Problems

Problems such as the like the quality of the images and the incorrect color reproduction of the webcam images. Without correcting these, the color of the sky can be distorted. For example if the sun is shining directly in the camera or if rain is covering one of the camera lenses.