HOG table

An indoor user pointing at a location on the table-top surface, which contains a virtual representation of the outdoor world. The table uses four video cameras to capture a 3D representation of objects located on the table.

The reconstruction table with the new apple 30 inch display

A closer shot of the 30 inch apple cinema display with the covers up. I finished mounting the display on the 26th of October 2007. It is significantly more useful than using a projector as I did previously. A top down projector casts imagery onto the users hands and props and therefore leaves a shadow on the surface too. A projector is also considerably more painful to mount accurately every time the table is moved the projector has to be remounted. Whist the 30 inch cinema display took me about a week to mount it produces a better image without the other problems described for projectors.

Example of the type of image that can be seen on the surface of the table.

Four mac minis (one attached to each camera) are used to perform image processing

Close up of one of the dragonfly camera from point grey and the custom mounts used to acquire images

The synchronization unit that we created to synchronize the dragonfly cameras

Close up of the sync unit attached to one of the dragonfly cameras

The second table created

The very first table I created - made from a cardboard box

It is necessary to separate users’ gestures from the rest of their bodies, and so I built a tabletop device resembling a miniature movie set, as can be seen on the right. The table has a blue-coloured wall perimeter, so that props placed onto the table are easily segmented from the background. Four Point Grey Dragonfly cameras are evenly spaced around the outside and all pointing towards the centre of the table. The cameras are mounted using standard optical camera mounts to support fine-grained calibration. A projector mounted above renders a top-down view of the remote users’ working area onto the table surface. The cameras are recessed into the surface of the table so that they are able to only see objects above the plane of the table surface and unable to see the projected image. The cameras are kept horizontal to simplify the calibration process.

We use four easily transportable Mac Mini computers, each dedicated to processing the image from a single camera. Each Mac Mini processes captured images to remove the background, and produce a smoothed alpha mask with small holes removed. The OpenCV library is used to calculate a vector-based contour that approximates the outline of the foreground object.

See also the page on dragonfly synchronization and more on the HOG table history