Analyzing cyclists’ behaviors and exploring the environments from cycling tracks

1
Analyzing cyclists’ behaviors and exploring the
environments from cycling tracks
Xingzhe Xie1
, Kevin Wong2
, Hamid Aghajan2
and Wilfried Philips1
1
TELIN – IPI – iMinds, Ghent University, Belgium
2
Electrical Engineering, Stanford University, US

Overview
•
Introduction
•
Approach Overview
– Isolating tracks from GPS data
– Extracting features at different levels
– Analyzing cycling behaviors
– Road associations using OSM data
•
Results
•
Conclusion
2

Overview
•
Introduction
•
Approach Overview
•
Results
•
Conclusion
3

Dataset
● All Raw GPS data for 79 cyclists
ID, PersonID, Time, Latitude, Longitude, Heightm, Speed,
and etc.
● Cycle trips details database
TripID, PersonID, TripMonth, TripDay, TripDate, TripDay,
OriginTime, DestTime, Purpose, Dist, TimeSpent, and
etc.
● Participants’ socio-demographic database
TripID, AgeGroup, HealthStatus, Employment, AHIncome,
Gender, Yearof cycle

Our goals
Isolate cycle tracks from GPS raw data,
Analyze cyclists’ behaviors(both personal and
shared behaviors)
Explore how the tracks’ features relate to
cyclists’ demographic characteristics

Overview
•
Introduction
•
Approach Overview
•
Results
•
Conclusion
6

Isolating tracks from GPS data
Hierarchical Database

Extracting features at different levels
Tracks' features: BeginTime, EndTime, BeginLocation,
EndLocation, Distance, Duration Time, Staying Segments,
Average Speed and Speed Standard Deviation
Persons' features: Average speed, Average time duration,
Average distance of all tracks with the same purpose
Overall features: Average speed of people in the same age
group, or with the same gender or health status, or cycling
years.

Analyzing cycling behaviors
Assumption: the two places where cyclists visit most
often are the home and workplace (because most of
the cyclists in our dataset bike to and from work)
{Starting locations, ending locations, staying locations}
Home and work place are inferred based on the
frequency of the cyclists visiting these places
Cyclist's activities are classified to staying and moving
according to speed. Staying locations are the place
where the cyclist performs staying activity for more
than 10 seconds

OSM Preliminaries
OpenStreetMap (OSM): A
collaborative mapping data
contain many types of GIS data
including:
● Road locations and names
● Points of Interest
● Natural Features
● Bodies of Water
● Political boundaries
Cycling GPS traces are
associated with the closest road
segment
OSM extract of a 25 square km
section near Newcastle upon Tyne.

OSM Geographic Processing
Cumulative Cyclists tracks covered 4
degrees of Latitude and Longitude,
about 600,000 OSM road segments
Developed a spatial partitioning
algorithm to search for nearest road
segments
● Used 300 longitudinal and latitudinal
bins, a total of 90,000 spatial bins
● Largest bin size was 600 road
segments, a worst case speed up of
1000X.
Spatial bins transversed by one
cyclist.

Road associations using OSM data
● Bin size chosen
to limit worst case
runtime
● Mean of 20
segments per bin.
● Median bin size of
4 segments
● Roughly 56000
out of 90000 bins
were empty

Overview
•
Introduction
•
Approach Overview
•
Results
•
Conclusion
13

Cycling behaviors analysis
Red lines: Places where this
cyclist visits often
The height of red lines:the
frequency of his visit to that place
Home
Work place

Cyclist ID Distance
(km)
Mean
Speed(kmh)
Speed standard
deviation
Duration
(m:s)
Purpose
103 5.42 15.9 10.3 14:55 1
103 5.48 16.9 8.86 11:44 1
103 5.49 15.9 8.04 12:20 1
103 5.78 13.8 10.1 13:11 1
103 5.72 14.6 9.76 12:17 1
103 1.47 7.40 6.42 9:43 1
Tracks from home to work
Abnormal routine from home to work: the last track:
This cyclist goes to another place for work

MO TU WE TH FR SA SU
To work 57 62 61 55 48 4 4
To visit (friends, etc) 2 1 4 7 10 6 7
To work related task 5 10 9 8 5 2 0
To Food shopping 5 11 5 4 10 7 8
To Non-food shopping 4 1 3 5 3 3 5
To School (Student) 3 5 5 4 4 2 1
To Entertainment 4 3 1 6 8 5 4
To Eat (Lunch, etc) 2 2 4 1 2 3 2
To Home 62 65 66 58 57 23 32
Other 13 11 14 4 10 5 9
the relation between purpose and day

Road associations using OSM data
Cyclist Trajectory Closest Road Segments

Finding areas with cycling anomalies
Our initial anomaly of interest is rapid starts and stops of the cyclist, show in
red.

Closest road segments to cycling anomalies
Alternative Anomaly visualization as a
heatmap.
● Hotter colors represent higher numbers of
anomalies.
● Portions of the heatmap with no anomalies
were left as white for readability.
● "Highbury" and "Jesmond Dene
Road",highlighted in red had the most
anomalies
● Site of 12 high acceleration/deceleration
events

Overview
•
Introduction
•
Approach Overview
•
Results
•
Conclusion
20

Conclusion
What we did:
Initial analysis about cycling behaviors and road structures
Future work:
Apply machine learning method to analyze the the relation
between age/gender/health status and cycling features
Explore how the environment affects cyclists’ behaviors
through studying the tracks and map data
Use other supplementary data could be incorporated as well,
such as historic weather data for the time and place of the
GPS traces along with historical traffic data

THANKS FOR YOUR ATTENTION!
Xingzhe.Xie@telin.ugent.be
kbw5@stanford.edu

Analyzing cyclists’ behaviors and exploring the environments from cycling tracks

More Related Content

Viewers also liked (20)

Similar to Analyzing cyclists’ behaviors and exploring the environments from cycling tracks (20)

Recently uploaded (20)

Analyzing cyclists’ behaviors and exploring the environments from cycling tracks