Press Enter to Search


April 1st, 2012

The landmark exhibition at Harvard’s GSD culls from 75 years of teaching and practice, showing seminal artifacts in plexi covered cases with vertical annotations done in journalistic style. The challenge put to the design team at INVIVIA was to develop an interactive timeline placed at the entrance to the exhibit that would allow visitors to gain a quick temporal context or an in-depth view of a specific period or set of artifacts. The interactive table had to be intuitive to use, show graphic and textual detail in high resolution, support multiple users and be quite robust. Oh, and it had to be finished and installed in 2 months!

We looked at several options. Using a large touch sensitive table was attractive in that it would allow many users and provide a natural interface that would make it easy to browse subjects in depth.  But the cost and complexity of the large interactive surface seemed prohibitive.  A simpler approach, which quickly became the favorite, used top projection onto a matte white surface (plywood covered with adhesive backed matte vinyl) and multi-blob detection to notice where the users hands were. This approach doesn’t allow you to notice whether the user it touching the surface or above the surface, so selecting/deselecting an object becomes difficult.  So we came up with an interaction scheme using different horizontal zones that had different interaction meanings.  The computer vision challenge is to notice only the user’s hands even though there is a continuously changing projected image on top of them.  This gets solved by flooding the table with IR light and filtering out the white light to the webcam with thick Neutral Density filters.

IR image

The IR image with multi-Blob detection.

White-light view

White-light view of table with projection.

The multi-Blob tracker was a repurposing of Python code used for many touch table applications, based on Markus Gritsch’s VideoCapture python module ( The vision algorithm starts with capturing a base frame and then subtracting each subsequent frame from it to find what moved on the table. In this difference frame we define a Region of Interest and look for candidate blobs.  These candidate blobs are flooded and culled to remove too small blobs generated by noise in the camera signal. Each real blob is measured and these measures are written to a file. Then this process is repeated for the next frame.

We decided on full HD 1920 by 1080 PDLP projectors because of the high visual quality of the DLP image, the fact that there are many competing projectors in this market segment which keeps the price down, and the relatively high resolution.

We built a crude but effective prototyping area in INVIVIA’s recently renovated top floor space by supporting a 16ft, 2×12” board between the top of the AC control room and a 10ft ladder mounted on tables. This allowed us to quickly build projector housings out of MDF with a bandsaw and drywall screws and attach them to the board with drywall screws and eventually with sliding supports for fine position adjustments.

These projectors must be operated nearly horizontally or they will overheat so each projector housing had a 45 degree mirror attached to angle the image toward the table below.

Initially we hoped to use three projectors arranged horizontally, giving us a 5760 by 1080 total image size, but realized after initial testing this that stretching the 1080pixel image across the 32in dimension of the table would not handle the small type we hoped to annotate the images.   We wound up buying an additional projector, turning each projector 90 degrees so that the 1920 pixels stretched across the 32in table and shrunk the table length to 4320 pixels for a ratio of 2.25:1.

Installation view

Projector sled

The final aspect of the projection setup that needed to be worked out was the IR emitters.  We chose rectangular array emitters normally used outdoors to illuminate parking areas for nighttime surveillance.

The final installation used a well reinforced plywood projector/computer tray hung below a false ceiling.  Openings were cut for the projector light, the IR emitters, and a webcam. Each projector was mounted on a slide that held the 45 degree mirror and allowed easy adjustment of the critical space between projectors that almost gave a sense of one continuous image.

There are no comments yet, add one below.