1. A Participatory Urban Traffic
Monitoring System:
The Power of Bus Riders
Zhidan Liu, Member, IEEE, Shiqi Jiang, Pengfei Zhou, Member, IEEE, and
Mo Li, Member, IEEE
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS,
OCTOBER 2017
Kang Yen
1
3. • This paper presents a participatory sensing based urban traffic monitoring
system, which takes the public buses as probes to sample the instant
road traffic conditions.
• The advantages of this approach are large service coverage and low
energy cost.
Introduction
3
4. • Our system relies on the help of bus
riders and crowdsources the traffic
sensing tasks to their mobile phones.
Introduction (cont’d)
The mobile phones anonymously
upload data to a backend server.
traffic sensing data
4
5. Motivation (I)
• People used to widely deploy
infrastructural devices like
inductive loop detectors and
traffic cameras at roadsides to
detect instant traffic conditions.
• These conventional approaches
incur maintenance costs, which
greatly limit the road coverage.
5
7. Motivation (III)
This figure show that the average errors are high as 41m and 68m when
the phone is at the bus stop and on a moving bus respectively.
GPS localization errors in downtown
Singapore
average
error
7
8. Problem
Precise bus
tracking problem
Bus status and bus
stop identification
among all possible
bus stops
participatory sensing data
Cellular signals for vehicle tracking
8
9. Related Work
We experiment with 5 bus
routes and measure the
cell tower signals at 86
stops
Similarity measurement of bus stop cellular
fingerprints(Measured bus routes)
9
11. Related Work (cont’d)
Similarity of the cellular
fingerprints collected at
the same bus stop.
score > 3 (90%)
score > 4 (50%)
The matching algorithm runs over all
bus stop candidates in the database,
and select one bus stops with the
highest similarity score
11
12. System design overview
step1
step2 step3 step4 step5
The system consists
of two major
components, i.e
online/offline data
collection and
trajectory mapping
for traffic estimation
12
13. Data collection
• A beep detection approach is
applied to detect whether the
user is on a bus or not.
• The card readers generate a
unique beep with specific
frequencies of 1KHz and
3KHz.
• The mobile phone terminates
current trip if no beep is
detected for △=10 mins, which
implies the user has got off the
bus.
13
14. Per sample matching (I)
ID:1
Bus stopA
ID:3
ID:5ID:7
Bus stopB
ID:3
ID:5
ID:2
next bus stop
14
15. • The matching algorithm runs
over all bus stop candidates in
the database, and select the
bus stop with the highest
similarity score.
• All cellular samples with low
highest similarity score are
discarded with no further
processing.
Per sample matching (II)
ID:1,
ID:3,
ID:5,
ID:7
ID:1,
ID:3,
ID:6,
ID:7
Bus stopAID:3,
ID:5,
ID:7
ID:3,
ID:7,
ID:9
The Bus stopA shows the cell tower IDs in
descending order of their Received Signal Strength
ID:1,
ID:3,
ID:5,
ID:7
15
16. Per bus stop clustering
• The cellular samples collected
at 3 different bus stops are
clustered into 3 groups in the
space.
• By applying the co-clustering
on all cellular samples, we
finally derive a sequence of n
clusters {C1,C2,...,Cn}.
16
17. Per trip mapping
• The maximum likelihood
estimation recovers the
trajectory of current trip
in best matched bus stop
sequence
• It also determines the
most likely bus stop for
each cellar sample
cluster on the trajectory.
17
20. • The detection ratio is above 90%
when the distance to card reader
is within 4m.
• The bus identification error is less
than 8%.
• In general, our bus identification
algorithm works well with rare
mis-identification cases.
Bus stop detection and
identification performance
20
21. Traffic estimation performance (I)
• The paper classifies the
travel speeds of automobiles
into 5 levels.
• We find that road segments
of the left and bottom area
have the best traffic
conditions with traffic speeds
higher than 50 km/h.
21
22. Traffic estimation performance (II)
• Only 8 bus routes can cover more than 50% major
roads of the studied area.
• Compared with Google traffic map for the area, our
system provides higher road coverage ratio.
22
23. Traffic estimation performance (III)
• When we use 70% bus stop references, the
estimated traffic conditions do not degrade much .
• With fewer bus stops as landmarks in 50% roads,
the roads are segmented in a more coarse-grained
manner.
23
24. Traffic estimation performance (IV)
• Fig12:Traffic estimation
compared with official
traffic data and Google
Maps’ indicator.
• Fig.13 suggests that VA is an effective
measure for traffic conditions for heavy
traffics which result in low traffic speeds.
24
25. System overhead
• The power consumption
is as high as ~450mW if
we replace the cellular
signal with GPS.
• In general, cellular save
more energy than GPS.
25
26. Conclusions and future work
• This paper presents the design,
implementation and evaluation
of a participatory urban traffic
monitoring system.
• They plan to derive the complete
traffic of a region.
• They would want to encourage
more bus riders’ participation for
better performance.
• I think they should improve the
user interface of App and make
it easily usable.
26