Transcript UbiActive Smartphone-Based Tool for Trip Detection and Travel

UbiActive
Smartphone-Based Tool for
Trip Detection and Travel-Related
Physical Activity Assessment
Yingling Fan, [email protected]
Qian Chen
Chen-Fu Liao
Frank Douma
Sensing – Survey – Assess & Report
User
Wear smartphone
on her right hip
Daily Assessment
% of active & happy travel;
% of energy expenditures
related to travel
Auto Sensing
Location & Speed (every 30
seconds); acceleration (1Hz)
After-trip survey
Trip mode, activity,
companionship &
experience
2
Raw sensing outputs
Timestamp
2011-11-01 15:41:47
2011-11-01 15:41:48
2011-11-01 15:41:49
2011-11-01 15:41:50
2011-11-01 15:41:51
2011-11-01 15:41:53
2011-11-01 15:41:53
lAx
-0.336726
0.035131
-0.295038
-0.086559
-0.022146
0.053333
0.079704
lAy
-0.046676
-0.005836
0.006505
-0.254355
0.079066
-0.013562
-0.013553
lAz
0.133635
0.104520
0.259984
0.191731
0.011211
-0.002895
-0.122060
Lat
44.971064
44.971064
44.971064
44.971064
44.971064
44.971064
44.971064
Lon
-93.244507
-93.244507
-93.244507
-93.244507
-93.244507
-93.244507
-93.244507
Speed
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
1.000000
3
How to detect a trip?
• Counter A is for judging the start of a trip
– Every 30seconds, if the detected movement is larger than 30 meters,
counter A would automatically add one.
– When counter A reaches 20 counts, indicating there is a 10-minute
continuous movement, a valid trip is considered to be happening.
• Counter B is for determining the end of a trip.
– Every 30 seconds, if no “location change” is updated, count B will
automatically add 1.
– When counter B reaches 10 counts, meaning there is no significant
movement for 5 consecutive minutes, the trip is considered.
• Note:
– Both counters A and B have default value at zero.
– Counter A will be reset to zero if location change is not detected before
reach 20 cts.
– Counter B will be reset to zero if location change is detected before
4
reach 10 cts.
After-Trip Survey
5
Evaluation and Testing
• Lab testing
–
Network usage: data size is less than 1 KB per day
–
Memory storage requirement: collect around 7Mb of raw sensor data and statistics per day
(150mb for 3 weeks)
–
Battery life: around 12-15 hours without additional voice/text/data usage
–
Trip Detection: almost 100%
• Testing among 17 real smartphone users recruited from the
University of Minnesota campus
–
Time: October-November, 2011
–
$100 cash reward upon completion of 3 weeks of compliance.
–
Initial background survey, exit survey, and requirement to fill out paper version diary.
–
23 Participants recruited
6
Participant summary statistics
• Of the 17, 12 males, 9 White, 5 Asian, average age 23.
• 8 undergraduate, 8 graduate students, 1 alumni
• 7 car owners.
7
A Case Study
Trip Information of a Participant on November 3, 2011 – Part I
8
Trip Information of a Participant on November 3, 2011 – Part II
9
Findings: What went well?
•
Phone based survey collected info on 509 trips occurred in 256 person-days with
valid data.
•
36% were made on foot, 1% by bike, 26% by private car, and 37% by transit
•
29% were back-to-home trips, 30% school-related, 10% work-related, 11% eatingrelated, and 9% were shopping/errands.
•
56% were made alone, 34% with friends, and the rest with family.
1 –most
negative
2
3
4
5 most
positive
How satisfied?
0.39%
4.63%
29.73%
36.10%
29.15%
How good feel?
0.58%
5.21%
33.59%
31.47%
29.15%
positive outweigh negative?
0.77%
2.90%
25.87%
31.08%
39.38%
how happy
0.19%
3.09%
30.12%
35.91%
30.69%
10
Participation Experience & Compliance
• 76% participants reported “satisfied”
• 88% reported increased travel behavior awareness.
• 98% at least “somewhat agree” that they felt comfortable
having smartphone detecting travel behavior
Compliance (always)
120%
Disruptiveness (very)
100%
80%
60%
40%
20%
re
po
r ts
da
ily
on
Re
ad
vib
ra
tio
n
on
Ke
ep
G
PS
Ke
ep
er
tra
ve
l
tri
gg
er
if
Fi
ll o
ut
pa
p
an
nu
al
ly
su
rv
ey
M
-tr
ip
ph
on
e
ut
af
te
r
W
ea
r
Fi
ll o
Ti
m
el
y
ba
tte
r
y
ch
ar
gi
ng
0%
11
Caveats: What went wrong?
• Some reported poor trip detection rates. Trip detection
rates (range 0-90%) depend on
–
–
phone brands & phone newness
GPS signal strength at trip origin, destination, route.
• Converting acceleration outputs to energy expenditure
estimates is much more complex than expected. Hardware
differences exist.
• Battery consumption issue is a key challenge.
• No behavioral differences between intervention and control
groups.
• Issues of missing data
12
Missing Data Plots
HTC EVO
Motorola Droid
Google Nexus
Samsung
Galaxy
13
Next step: UbiActive → SmartTrAC
•Sensing + survey → Sensing + data mining + survey
•After-trip survey → end-of-the-day activity or trip survey
walk
home
walk
car
school
shop
car home
eat
car home
walk
home
walk
car
school
shop
eat
14
This project and subsequent work are supported by
– the ITS Institute, and
– the Center for Transportation Studies at the University of
Minnesota.
[email protected]
15