Second keynote speaker presentation
By Hans Skov-Petersen
BIKEABILITY & University of Copenhagen, Denmark
Topic: Application of GPS tracking in bicycle research
Introduction to Multilingual Retrieval Augmented Generation (RAG)
Application of gps tracking in bicycle research
1. Photos: Inger Grønkjær Ulrik, Andre Neves and Hans Skov-Petersen
www.bikeability.dk
Application of GPS tracking
in bicycle research
Hans Skov-Petersen
hsp@life.ku.dk
Geoscience and natural resources
University of Copenhagen
2. Overview of the presentation
• Basic considerations:...
• Sampling locations or sampling individuals
• A base-line framework for analysis of tracking
data
• Motivation and field of application for
investigations in cyclists’ route choice and and
way finding behaviour
• Scopes of spatial cognition and behaviour: a
‘Focal‘ vs ‘Global’ approach
• Data models and spatial domains in spatial
behaviour: Fields vs networks
4. GPS tracking:
Analytical framework
Description Inference
Locations only Additional layers
Local
Individual points
Where is (x, y)?
What is the PDOP of..?
Distance to paths’ and points of
interest.
Where do stops occur?
Focal
Spatial/temporal
‘window’
How fast?
Stop/go?
How steep? Speed/slope relations
Choice of ‘next point’
(relative to options)
Zonal
Single
track/tours/routs
How far?
Round trip?
Average speed?
Altitude difference?
Min/max altitude along a track
Land cover distribution
Choice of route (relative to
options)
Global
All tours, for an
individual or all
respondents
Data mining
Spatial/temporal clustering
Area of interest
Path pressure
Kernel distribution
OD distribution
Relation of congested
locations
= Map Algebra
(Dana Tomlin)
5. Path pressure
Description Inference
Locations only Additional layers
Local
Individual points
Where is (x, y)?
What is the PDOP of..?
Distance to paths’ and points of
interest.
Where do stops occur?
Focal
Spatial/temporal
How fast?
Stop/go?
How steep? Speed/slope relations
Choice of ‘next point’
(relative to options)
Zonal
Single
track/tours/routs
How far?
Round trip?
Average speed?
Altitude difference?
Min/max altitude along a track
Land cover distribution
Choice of route (relative to
options)
Global
All tours, for an
individual or all
respondents
Data mining
Spatial/temporal clustering
Area of interest
Path pressure
Kernel distribution
Relation of congested
locations
7. Zonal Statistics
Description Inference
Locations only Additional layers
Local
Individual points
Where is (x, y)?
What is the PDOP of..?
Distance to paths’ and points of
interest.
Where do stops occur?
Focal
Spatial/temporal
How fast?
Stop/go?
How steep? Speed/slope relations
Choice of ‘next point’
(relative to options)
Zonal
Single
track/tours/routs
How far?
Round trip?
Average speed?
Altitude difference?
Min/max altitude along a track
Land cover distribution
Choice of route (relative to
options)
Global
All tours, for an
individual or all
respondents
Data mining
Spatial/temporal clustering
Area of interest
Path pressure
Kernel distribution
Relation of congested
locations
9. Bikeability: GPS trip statistics
Number of respondents 179
Number of trips 1292
Avg. dist 5.4 km
Avg. time 22.4 min
Avg. speed 14.4 km/h
10. Speed/Slope
Description Inference
Locations only Additional layers
Local
Individual points
Where is (x, y)?
What is the PDOP of..?
Distance to paths’ and points of
interest.
Where do stops occur?
Focal
Spatial/temporal
How fast?
Stop/go?
How steep? Speed/slope relations
Choice of ‘next point’
(relative to options)
Zonal
Single
track/tours/routs
How far?
Round trip?
Average speed?
Altitude difference?
Min/max altitude along a track
Land cover distribution
Choice of route (relative to
options)
Global
All tours, for an
individual or all
respondents
Data mining
Spatial/temporal clustering
Area of interest
Path pressure
Kernel distribution
Relation of congested
locations
14. Studying wayfinding behaviour
Motivation and potential fields of application
Preference estimation and evaluation
• Investigation of the relative importance of
characteristics to the bicycle infrastructure
Route finding
• Preset impedance parameters
• Incremental, personalized parameters (web
2.0 style)
Accessibility modeling
• Assessment of anticipated effects of planned
infrastructures
Behavior simulation
• Agent-based modeling
15. Revealed Preference
Look at what people do!
To reveal preferences,
behaviour has to be
investigated in terms of
possibilities
... So it is all about choices
made among alternative
options
16. 1: Do we have a perfect, ’mental
map’ to base out choices on?
2: Do we apply knowledge that
can be perceived from our
immediate surroundings?
? ?
... A ’focal’ or ’node’ scope (locomotion)
... A ’global’ or ’route’ scope (wayfinding)
?
How do bicyclists navigate
– investigation strategies
18. Strategies for generation of
alternative routes
Based on OSM (with moderations) chosen route
was compared to …
• Shortest path
• A random selection of alternatives
• Based on a modified labeling algorithm
• Max overhead distance over chosen route: 25%
• Max distance from chosen route: 1000 m
• Max 20 alternatives
• Two approaches:
• Including Path Size (a measure of internal
overlap)
• Max25: Allowing only member with less
overlap than 25% with any other alternative
in the set
?
20. The Route models
Parameter Path Size Max25%
Shortest
path
Length -0.00433*** -0.00254*** 0.06005 ***
Number of left turns -0.18323 *** -0.21797 *** 0.33594 ***
Number of right turns -0.0738 ** -0.12617** 4.22133 ***
% of route with Cycle track 4.66329 *** 4.68672 *** 141.683 ***
Cycle lane 5.86333 *** 7.82885 *** 66.2823 ***
Designated cycle track 6.20337 *** 8.72841 *** 182.773 ***
Shared track 2.17781 *** 2.79813 *** 270.278 ***
% of route with Artery road -2.34578 ** -4.04166 ** -88.9948 ***
Minor road 0.80765 0.43654 52.9103 ***
Other road (road type not
specified) 0.43071 -0.63882 84.0751 ***
Road with multy story housing -0.78488 -1.42363 108.121 ***
Shopping street -9.5252 -16.99 204.692 ***
Log Likelihood
Function -852 -309 -917
Routes are compared to a standard situation with no
bicycle facilities on a main road
21. The Focal model
Quite early results….
Parameter Coefficient
Significans
level
Directions Relative angle to destination 0.0005317***
Left turn -1.181548***
Right turn -1.48063***
Uturn -1.747325***
Bicycle facilities Track 0.7465979***
Lane 0.9427726***
Designated track 1.140476***
Shared track 0.5860569***
Green Environment -0.0608228***
Road type Artery road 0.7905237***
Minor road 0.8029141***
Other road (road type not specified) 0.5247407***
Road with multi storyed housing 0.0150856*
Shopping street 0.2214325***
22. The route model vs the focal model:
The known vs unknown areaal
A route is stated to be in a ‘unknown area’ if
more than 50% of its points where more than
250m from poinst on any other route taken by
the same respondent
Pseudo R2 Route model
Max25%
Focal model
All
n=1291
0.7523 0.3679
Known area
n=1092
0.7811 0.3672
Unknown area
n=199
0.6046 0.3743
23. The Focal model
The first and the last 25% of of a trip was
defined as ‘start’ and ‘end’
Pseudo R2 Focal model
All (unfortunately not ‘Middle’)
n=192,370
0.3679
Start
n=50,953
0.3130
End
n=42,424
0.3618
24. Way forward..
• Further analysis has to be performed
• Focal vs Global scope for different cyclist types and
different cycling situations
• Refinements of parameters
• Reassessment of the estimates to support probabilistic
locomotion in Agent Based Models
• We are aiming at four papers from the study:
• Cyclists’ wayfinding and route choice (GPS/RP)
• Cyclists’ wayfinding and route choice (SP)
• A typology of Danish cyclists, based on mobility styles
• Cyclist types applied to wayfinding
28. That’s it
Thanks for now
Hans Skov-Petersen – hsp@life.ku.dk
Jette Bredahl Jacobsen
Bernhard Snizek
Suzanne Elisabeth Vedel
Skov & Landskab, LIFE/KU
Bernhard Barkow, creativeyes.at
Bikeability
– cities for zero-emission cities and public health