This photo shows 3 of 4 projections panels in the VAG cupola. Note that the image on the right has not been fully mapped to the panel shape (yet).

I have been working since January on The Glocal Project, a large-scale contributive art project being run out of the Surrey Art Gallery's TechLab. The  team as a whole has been developing a series of simple software applications designed to act as creative and pedagogical tools for both individuals and educators.

A couple of weeks back, we had the chance to bring some of our work out of the lab and into the Vancouver Art Gallery, where we presented a large four-screen projection in the Gallery's rotunda cupola.

A grid of images from a single camera feed mapped onto the cupola panels

Technically, this presented a challenge. Not only did we want to have four linked screens in which images could overlap - we also wanted the projections to be 'mapped' to the four curved panels of the cupola. All four panels together would be shown in 'flat' form on four plasma screens. To add to the confusion, we would be running four live camera feeds into the system - three webcams and one wireless DV camera. 4 cameras. 6 computers. 8 screens. As has been the case for much of the Glocal project, we turned to Processing for a solution.

Our first plan was to utilize Daniel Shiffman's Most Pixels Ever Library. Though it is very easy to set up and use, it is not made to transfer large amounts of data. We wanted to stream 4 camera feeds to 6 client machines, which meant there would be a lot of information moving over the network. So, we shifted gears. The system ended up being driven by a custom-written UDP muiticast server, which sent camera data along with events to trigger changes in the visual application.

The images from the camera were placed into a generative panel set, derived from the language of comic pages. Some images were displayed live - others were delayed by up to 30 seconds. Some panels were rendered in greyscale, and some were combined with other feeds to create dynamic multiple exposures. The grid patterns were occasionally very ordered, but often quite abstract. Sometimes unknowingly, the gallery visitors were building their own micro-narratives - temporal stories that played without a beginning or end in the circular rotunda. 

We ended up capturing 2,700 images from the 5 hours that the installation was alive. You can see just more of these micro-narratives in the Glocal Flickr Pool that we have created to document the event. Here are just a few sample images:

These images are flat - and show all 4 of the rotunda panels together. For the projections, each panel was rendered onto a custom 3D shape. Becuase the placement of each of the 4 projectors was different, we had to be able to tweak the size and dimensions of these 3D shapes. Again, this happened in Processing, where our OpenGL shapes were created and modified by a small custom applet. This way, we were able to fit each panel image to the architecture live - and make adjustments as needed.

A simple grid is rendered onto the 3D shape to aid in calibration

The 3D shape is then modified using a set of key commands (this can be teaked live)

Finally, the image feed is mapped onto the shape. Here I am using a straight webcam feed to demonstrate.

Glocal is an open-source project, so we will eventually be posting the code that we used for this installation for download. Stay tuned on our blog for details.