Sensor Description

The sensors detect fast changes in the thermal background radiation over a small region (approximately a cone 2 meters in diameter on the floor). If there is continuous motion under a sensor (such as a large group of people walking by), then the sensor will fire repeatedly, about once every 2 to 3 seconds. If people are standing or sitting under a sensor the sensor will still activate, but less often. This is due people adjusting their stance, gesturing, or performing self-adjustment motions. Someone standing absolutely still under a sensor could theoretically avoid triggering it, but in reality it is very difficult to stand that still for very long. Larger stationary groups will tend to have higher activation rates, since the inter-arrival time of gestures and adjustments goes down.

The workday on Friday, August 4th, 2006 (time codes 1154692800000–692800000) is a good period to find examples of many kinds of gross behaviors. That is because SIGGRAPH was in Boston that week and many people came to MERL on Friday for a demo open house, so there are large groups of people walking around the space and loitering in lobbies and hallways. This is in addition to the more typical behavior we see in the space.

Moving Around the Lab

The first example shows people moving around the space. A very common thing to see is a single person, or possible a couple people walking side-by-side, walking down a hallway. Here we see one such behavioral trace in the North wing of the 8th floor. Shown on the left of the above piano roll, you can also explore the raw data by examining sensors 418, 419, 420, 421, 422, 423, 426, 427, and 428 from 09:11:13 to 09:11:26 (time codes 1154697071000–85000).

On the right side of the same plot it is possible to see a small group moving from Nitta lecture hall to the Belady conference room. The trace is similar to the single-parson case, but it appears wider because each sensor fires multiple times as the group moves underneath. The data includes sensors 318, 319, 320, 321, 379, 380, and 375 from 09:12:17 to 09:12:34 (time codes 1154697137000–54000).

A Small Meeting

There is a meeting in Belady on and off all day on the 4th. You can see one session at sensor 452 from 09:13:00 to 09:24:30 (time codes 1154697180000–870000). Even when people are sitting down and are subjectively &lq;still&rq;, the sensor will still occasionally register motion due to communicative gestures, shifts in posture, or self-adjustments. The inter-activation times during this session of the meeting range from from a couple seconds to a couple dozen seconds. This is why it is difficult to make reliable, accurate motion activated lights with a single sensor. you can, however see the meeting bracketed by arriving and leaving signatures on either end (sensor sequence 379, 380, 375 for arrival around 09:12:30 and sequence 375,374,376 for departure around 09:24:30).

Large Groups

The final set of examples shows the signature of large groups, which are fairly rare at MERL. Near the top of the above plot we see another session of the same small meeting in Belady (452). Compare that trace to the the large group demonstrations in the lunch room and the main reception lobby (the latter starting on the right side of the plot around 11:33:13 (time code 1154705593000) and encompassing the lobby sensors 298, 334, 336, 338, 339, 340, 345, and 346.

This group arrives at the lobby from Eight North while traversing the sequence of sensors 400, 401, 444, 403, 394, 396, 397, 398, 390, and 391 from 11:32:30 until 11:32:57 (time codes 1154705550000–77000). This group was probably comprised of about a dozen people. Notice that the movement trace is very wide, with each sensor firing maybe a half dozen times during the passing of the group.

When an individual moves through the space they create a structured pattern of activations that link parts of the space together in a meaningful way. Tracking is the process of recovering this structure from the raw observations. Here we see a set of activations linked together into a tracklet:

Thank you to those of you who inspired this tutorial by asking interesting questions about the nature of the data and the sensors.

Finding More Examples

Single people walking around MERL is by far the most common example behavior in the data, so finding examples should not be hard. The easiest way to find clean examples is to filter the tracklets database for tracklets that span a large distance in an appropriately short time. Tracklets originating at the Nitta Room (doors at sensor 299 and 379) are more likely to contain larger numbers of people, however the majority of all tracklets represent single people.

Look in the files belady.txt, 8th.txt and 7th.txt for examples of scheduled small meetings. Look in nitta.txt for meetings scheduled in the large lecture hall, some of these are small meetings that overflow from the small rooms, and some will be larger meetings. Scheduled meetings do not always happen, they may be delayed relative to the schedule, and there are many unscheduled meetings, but the above files are still a good place to start.

Large stationary crowds rarely occur at MERL. The August 4 above is probably the primary example. The large groups rotated around demo stations in the lobby, the lunch room, the office near sensor 442, and DiamondSpace demo space on the 7th floor, off-net near sensor 505. The file lunch.txt indicates meetings that were scheduled for the lunch room: these are typically all-hands socials intended to celebrate some event with food, and therefore also draw large crowds.

Days marked in fire.txt may contain significant evacuations that will include large, but probably diffuse crowd movements. The 11:53am evacuation on April 20, 2006 include a large, cohesive crowd moving from Nitta to the fire evacuation stairwell in the off-net hallway that connects sensors 445 and 222.