The Sixth International Conference on
Information Processing in Sensor Networks (IPSN '07) IPSN '07 will be
hosting an Extreme Sensing competition April 26, 2007.
Each team will need to build a sensor system using five sensors or less
that can count people walking through a 10ft x 10ft arena. In Phase 1,
the arena will be placed in a high traffic area and all teams will sense
normal foot traffic from conference attendees for a duration of 10
minutes. Points will be awarded for correct detections and subtracted for
false positives and false negatives. In Phase 2, each team must try to
walk through the arena 10 times without being detected by their opponents'
sensors, in order to increase their own score while decreasing the scores
of their opponents. Results will be broadcast live during Phase 1,
providing each team an opportunity to discern ho wthe other systems work
and how they might fail. For more details about the competition, and how
to enter, visit the URL listed above. The grand prize will be a
Each team must build a sensing system to
count the number of people that exit from a 10ft x 10ft arena,
which will be located in a high-traffic area such as a doorway. Each team
must setup a sensing system around the area, leaving two sides and the
center unobstructed for people to walk, as shown in the diagram below.
Unobtrusive sensors such as floor sensors or ceiling arrays will be
permitted in the main arena area. Teams may use passive infrared motion
sensors, active infrared break beams, ultrasound, dopplar radar, floor
sensors, etc. A system may also include combinations of sensors, with two
restrictions: 1) no system may have wires more than 12 inches in length* 2)
no system may use more than five pixels in total, eg. no cameras
with more than five pixels, and no more than five single-pixel**
sensors. Actuators will be permitted, as long as they do not interfere
with other teams' sensors and as long as they do not help the system
emulate more than five pixels (see below**).
During Phase 1, the sensing systems will
count normal foot traffic from conference attendees for 10 minutes. The
arena will be empty when this phase begins, and conference attendees will
be free to walk through in any way that they like. Ground truth will be
obtained from a series of judges who will count the number of people
who exit the arena by pushing a button on a mote. During this
phase, contestants will not be allowed to walk through the arena.
Phase 2 will consist of 10 rounds in which
each team is allowed to walk 1 player through the arena. The goal of each
team is to score points for themselves without scoring points for
competing teams. Each person may walk in any manner they choose, as long
as it is monotonically forward (ie. the person cannot walk back and forth
inside the arena). Teams will be assigned a random order for each round.
A team may opt to skip a round, in which case they may walk multiple
people during subsequent rounds (possibly simultaneously).
If there are 3 other teams, then in Phase 2 you will
need to cross an arena covered by up to 15 different sensors. Unlike
Mission Impossible, it is unlikely that you will get through
completely undetected. Instead, you will need to choose your walking
strategies carefully to avoid the teams that are most likely to beat you,
given the scores after Phase 1. The counts for each team will be shown
live on a video screen during Phase 1, as shown in the Figure to the
right. This is an opportunity for each team to discern how the other
systems work, and how they might fail. For example, radar and PIR sensors
may not detect slow moving objects; infrared break beams may count
multiple adjacent people as a single person; floor sensors may not detect
people with long strides; even multi-sensor systems may be fooled by
multiple people walking in different directions, etc. This knowledge can
then be used during Phase 2 to score points for your own team but not for
the other teams. No competitive strategies are off-limits, except that
all players must walk monotonically forward. This means that running,
jumping, etc. are acceptable, sprays that illuminate laser beams are
acceptable, etc. Be sure to focus on avoiding the sensors of the team
whose score is closest to yours.
To enter, a team must send email to Kamin
Whitehouse containing the team's name, institutional affiliation, the
players' names, and a rough description of the hardware to be used.
Entries must be received on or before April 10, 2007. The team
should expect a response with the username and password to be used on the
competition day. Email should be sent to:
On Game Day, each system must
immediately*** and automatically (ie. no human intervention) report scores to
a scoring server. This can be done through the following XML-RPC
bool report(int detections)
This function can be used to add DETECTIONS
to the total number of detections by a particular team.
In the case of any dispute, decisions about
scoring and eligibility will be made by judges on Game Day and will be
final. Contestants who would like to do something that might be the cause
of dispute should ask for clarification of the rules and/or pre-approval
of the hardware and techniques before the April 10 deadline. All rules of
this contest are subject to clarification or change at any time.
Congratulations to the "Pixelators" from
MIT who won the competition by count 203 out of 204 people with their
RSSI based, single-pixel radio sensor! We had several excellent competing
teams that used ultrasound sensors, infrared sensors, lasers, weights
sensors, and radio signal strength sensors. Thanks to "Super Counting
Bros" for the video
of Phase I, and to the Pixelators for the
following video of the complete competition!
The results of the three top-scoring teams
are shown in the graph below and compared with the ground-truth results
obtained by the judges. In Phase I, when the systems were used to measure
normal foot traffic from general conference attendees, the results of all
three teams were quite. In Phase II, when the teams could use adversarial
strategies to trick other teams' sensors, the behavior of the three teams
was quite different. The "Pixelators" score actually improved, indicating
that it was difficult to defeat using adversarial techniques. Towards the
end of the contest, the "Hawkers" score suddenly increased by almost 100
counts due to a single malfunctioning sensor.
* Wireless networks may be used. One
long cable (e.g. a serial or USB cable) may be used to connect your sensor
system to a computer.
** A single-pixel sensor can be
defined in many ways. As a general guideline, a "pixel" is a sensor input
that is directly mapped to exactly one sensor output, with no logic in
between. In some cases, a single AND or OR gate on binary sensors may be reasonable (in
order to make the sensor bigger, for example). Single-pixel sensors that
emulate multi-pixel sensors through movement
of the sensor, the use of moving mirrors,
etc, will not be acceptable. Similarly, using actuators that help the
sensor emulate a multi-pixel sensor (such as a time synchronized array of
infrared transmitters) will also not be acceptable. Borderline and
questionable hardware should be discussed before the April 10 deadline,
and all hardware is subject to approval by the judging panel on Game
*** The reports must be immediate
enough for an outside observer to discern the causal relationship between
the total score and what is happening in the arena.