UIC Innovation Lab, 2012
A Kinect stood in a 3D model of a floor: a demo built for a Lab,
replayed from a recording made in an office.
This study is anchored to a place: the sensor
stays put, and what is in its view changes.
“People, places, and things serve as observational anchors; activity is common to all three.”
Of the three, the same paper says that “attending to a place foregrounds changing presence and use”.
In July 2012 the Iota team built a viewer
that stood a live Kinect inside a model of a floor.
It drew what the sensor saw, as points, and whoever it tracked, as moving dots, at the place the sensor stood.
The sketch and its model are named for UIC, and TJ recalls an event at the UIC Innovation Lab. No plan or hypothesis for the demo has been found, so what it did has to be read from its code.
The instrument was one Kinect, saving
a frame of points or a stream of positions.
A depth frame
A point for each pixel of 640 × 480 with a reading, in millimeters: x to the Kinect's right, y up, z forward.
A tracker id
For each thing the tracker takes for a person, an id and one point, its center of mass.
A recording
Rows of id, x, y, z and the time. The file asks for one every 250 ms; they came about every 300.
The viewer drew three layers in one world,
on a black ground.
| Layer | Drawn as |
|---|---|
| The model | In its file's colors, lit half from all round and half from straight above |
| The points | Blue, one pixel each; keys 2 and 3 give the stored color or a gray by depth |
| The paths | By id, 4-pixel dots and thin half-transparent lines, a whole file at once |
The model and the sensor reach that world
by different routes, and one mirror shows it.
The model
The loader flips y as it reads the file, and the viewer scales it by 12.
The sensor
In 2012, stored points went through a saved alignment: a scale of 0.42, a turn and a place.
The screen
The whole world is mirrored once, as Processing's screen showed it, so the Kinect's right is on the right.
In the viewer's model, the frame goes where TJ
places it: its corner on the corner of the bay.
“The corner in the point cloud is the corner in the room containing the Kinect. So rotate 45 clockwise? Ish?”
- The turn from the 2012 alignment
- 50.7°
- The scale, where 2012 drew 11% small
- True
- The Kinect moves, and stays in the bay
- 62 in
What survives of the Lab is its model:
no frame or recording can be placed there.
| 16 July | A depth frame at 12:36, and 8 min 40 s of tracked positions from 13:15 |
|---|---|
| 17, 20 July | Exports of the model, the last, at 14:27 on the 20th, the one the viewer loads |
| 24 July | The viewer as it survives, copied at 10:40 |
| 3 August | The viewer's saved alignment, at 11:31 |
The one frame and recording were made
in the Iota office, at an office meeting.
“I think 16 July 2012 is the Iota office seen in the open house. This is the office at the north east corner. The clear corner in the image is the north west corner of the room. It was an office meeting.”
A note beside the sketches, dated 19 July 2012, calls them “the fixed data gathered from the meeting”.
Stood in the open house's model at true scale,
the frame fits that room.
Its floor goes on the floor, its clear corner on the room's north-west corner and its walls on the walls, with nothing stretched.
- The walls, against the model's
- Within 1.5 in
- Points 6 in past them, of 146,055
- 0
- Positions in the room, of 4,731
- 4,730
A step the fit did not use
falls where the model has a brick pier.
Two things are off, and depth that read long
could account for both.
- The corner measures 87.2° where the model has 90°.
- The Kinect comes out at 72.4, 317.0, about 10 in past the partition and 7 in past the window wall, just behind the model's brick pier.
- If far depth read long by about 1% per meter of distance, the corner would be square, the step would land at 107.3 in, and the Kinect would move to 79.4, 310.3, inside that pier, about 13 in short of the room.
The study stops at position:
an observation, not yet a model of activity.
“An observation is not an activity model.”
What makes this a meeting is a note and TJ's memory. The positions show where six ids were over 8 min 40 s, and not what anyone was doing or why.
What it can say is limited by one sensor,
one room and one point for each person.
- An id is the tracker's label for as long as it holds on to something. Six ids need not be six people.
- TJ's placement shows the frame in the Lab's model, and the frame was made in the office. The alignment the viewer saved in 2012 cannot be tested: nothing here was recorded with the Kinect where it puts it.
- No file records the event at the Lab, or says when it was.
Every claim here traces to the sketches,
their files, or TJ.
- The viewer, uic_kinectDemo, its loader's source, its saved alignment and its model, export_10.
- The sketches before it, the note beside them, and the frame and recording of 16 July 2012.
- The open house's model, on this site. TJ, 4 November 2026.
- McLeish, Computational Design and Modeling Activity, 2026.