8A_2_A containment-first search algorithm for higher-order analysis of urban ...
7A_4_Gps data collection setting for pedestrian activity modelling
1. GPS Data Collection Setting
For Pedestrian Activity Modelling
Adel Bolbol Fernández
Tao Cheng
Department of Civil, Environmental & Geomatic Engineering
UCL
2. Outline
• Background
– Research Area
– Positioning Sensor Devices
– Scenario
• Why
– Epoch Rate Detection?
– Pedestrians?
• Test
– Data
– Method
– Limitations
• Observations
– Positional Error
– Route Length Error
– Average Speed Error
– Start & End Error
• Conclusions
Background Why? Test Analysis Conclusions
3. Background: Research Area
Location Information Activity Recognition
“Sensor Data” •Diaries/logs
GPS
GSM Wi-Fi RFID •Algorithms
Inferring
Trip Trip Start Transport
Trip Route
Purpose & End Mode
Modelling
•Spatio-Temporal Human Activity
•Travel Behaviour
Background Why? Test Analysis Conclusions
4. Background: Positioning Sensor Devices
Travel Data
Wi-Fi, GSM GPS
& RFID
Indoor Wide High Long Life
Small Size
Coverage Coverage Accuracy Battery
Non-
Logged Data
Real Time Interactivity
Desirable Properties Interactive
Background Why? Test Analysis Conclusions
5. Why Pedestrians?
1. Almost every journey starts and ends in the
“Pedestrian” Mode (Walking)
2. Unlike vehicles, No Traffic Routing rules are
followed
Background Why? Test Analysis Conclusions
6. Why Find Best Epoch Rate?
“Epoch Rate”
The frequency at which GPS
measurements are taken “Data Collection
Rate”
(e.g. every 30 seconds)
“Capture Rate”
Is the more data the better?
When is data too much?
& Where is the threshold?
Practicability of battery requirements?
Best epoch rate for a realistic,
accurate, battery efficient
representation of human behaviour
Background Why? Test Analysis Conclusions
7. Scenario
1. Route Detection Walk
5
2. Detection of MoT
3. Start & End Locations 2 Work
Bus
4. Trip Purpose
GPS
points
User GPS
Walk 3
Home – Work
Home
Walk Trip 4
Tube
1
Background Why? Test Analysis Conclusions
8. Why Find Best Epoch Rate Detection? 10S
1. For Route Detection
Epoch rate =
10 Seconds
Background Why? Test Analysis Conclusions
9. Why Find Best Epoch Rate Detection? 20S
1. For Route Detection
Epoch rate =
20
10 Seconds
Background Why? Test Analysis Conclusions
10. Why Find Best Epoch Rate Detection? 60S
1. For Route Detection
Epoch rate =
60
20 Seconds
Background Why? Test Analysis Conclusions
11. Why Find Best Epoch Rate Detection? 4
1. For Route Detection
2. For Mode Detection
3. For Detecting Start-End Work
Epoch rate =
60 Seconds
Walk
Bus
Home
Background Why? Test Analysis Conclusions
12. Test: Data
• 7-8 minute walking journey
• Data was collected every 1 second
• Data was thinned to the following epoch rates:
(1, 10, 20, 30, 60, 120 and 300 seconds)
• 11 datasets were collected for the exact same
journey
Background Why? Test Analysis Conclusions
13. Test: Method
1. Route taken 1. Positional Errors
(Map matching)
2. Distance Travelled 2. Route Length Errors
3. Trajectory Speed 3. Average Speed Errors
4. Identifying Trip Start 4. Distances from First
& End & last points to Trips’
Start & End
Background Why? Test Analysis Conclusions
14. Analysis: 1. Positional Errors
From Averages of All Thinned Data Groups
From Thinned Data Group 1 From Thinned Data Group 2
50
Positional Error (m)
40
Bearing in mind;
100 100
Positional Error (m)
Positional Error (m)
•Road links in London (278,691 links): 80
80
30
•Average Length of Road links = 80 m 60 SD = 98.47
60
20
•Average Speed of Pedestrians = 1.2 m/s40
40 Average
20 20
= 15 m
10
Therefore;
0 0
0
•To select the correct Road Link We need1 at least 2 points 120 each
-20
1 10 20 30 60 120 300
-20
10 20 30 60
on 300
•Most AppropriateMin Outlier Max Rate = Approx. 66s
-10 Epoch Outlier Min Outlier Max Outlier
1 10 20 30 60 120 300
Epoch Rate (s)
Epoch Rate (s)
Epoch Rate (s)
Min Outlier Max Outlier
Results from 11 Datasets
Background Why? Test Analysis Conclusions
15. Analysis: 2. Route Length Errors
1100
4200
1000
900
Total Route Length (m)
800
Actual Route
700 Length =
600
667m
500
400
300
200
100
0
1 10 20 30 60 120 300
Epoch Rate (s)
Min Outlier Max Outlier
Background Why? Test Analysis Conclusions
16. Analysis: 3. Average Speed Errors
2.5
2.3
2.0
1.8
Average Route Speed (m/s)
1.5
Actual Average
1.3
Speed =
1.0 Approx. 1.2 m/s
0.8
0.5
0.3
0.0
1 10 20 30 60 120 300
Epoch Rate (s)
Min Outlier Max Outlier
Background Why? Test Analysis Conclusions
17. Analysis: 4. Distances from last points to Trips
Starts & Ends
250
200
Distance from End Destination (m)
150
100
50
0
1 10 20 30 60 120 300
Epoch Rate (s)
•Pedestrians Average Speed = 1.2 m/s Min Outlier Max Outlier
•Worst Case:
10s = 12m
20s = 24m
30s = 36m
+ Positional Errors
60s = 72m •Do we require the exact location?
120s = 144m •If not ---- Land Use/Mix data
300s = 360m
Background Why? Test Analysis Conclusions
18. Conclusions
1. Positional Accuracy = Approx. 15-20m
2. Length Calculation: 20 sec was most accurate
3. Speed Calculation: 30 & 60 sec
4. Start & End: 10 & 20 sec – 1 sec good accuracy &
high uncertainty
5. Around 20 or 30 sec is most appropriate for
pedestrian route modelling
Background Why? Test Analysis Conclusions