You are on page 1of 1

Coordinating Measurements for Air Pollution

Monitoring in Participatory Sensing Settings
Alexandros Zenonos, Sebastian Stein and Nicholas R. Jennings
Agents, Interaction and Complexity Research Group
School of Electronics and Computer Science
University of Southampton

Introduction

Coordination Algorithm

Participatory sensing is becoming an effective and cheap tool for
monitoring environmental phenomena.
Participatory sensing is about crowdsourcing
sensory information via sensors carried by
ordinary people (i.e. non-experts).

Local Greedy Search (LGS) Algorithm
A1

A2

t1
U=10

U=17

U=20

U=15

U=20

U=19

t2

Sensor information include air quality, noise,
radiation levels, light, humidity and gas
concentration.

U=0

Example:
• 2 participants.
• 2 time steps.
• 4 iterations (k)

U=21

U=9

U=11

Problems in current participatory sensing applications:
• Partial coverage of the areas of interest,
which results in poor situational
awareness.
• Duplicate work, which results in energy
loss.

• Search greedily in space and time for the best
observation by making small/local changes in
policies.
• Keep going until no further improvement is possible.

Experiments
Simulation

Benchmarks
• Greedy algorithm: Greedily
select the best observations at
each timestep
• Patrol: Take all measurements at
every timestep.
• Random: Take measurements
randomly throughout time.
• Exhaustive Search: Optimal
policy.




Real sensor data used.
Real mobility patterns.
1-5 days
1-250 participants

System Architecture
Key results

Aim

LGS is 33.4% better than the state-of-the-art greedy algorithm
and significantly faster than the optimal algorithm.

Provide algorithms for coordinating
measurements in the participatory
sensing setting for environmental
monitoring in order to achieve better
situational awareness and avoid
duplicate work.

We use D-optimality Criterion
to value the information at
spatio-temporal locations. 𝐼

𝑋𝐵 ; 𝑋𝐴 = 𝐻 𝑋𝐵 − 𝐻(𝑋𝐵 |𝑋𝐴 )

Use Gaussian Processes to
model correlations over space
and time. 𝑓

~𝐺𝑃(0, 𝐾(𝒙, 𝒙′))