A Smartphone Application
that Automatically Provides Arrival Time
based on Estimation of Train on Board
Haruki Inoue
Graduate School
of Business Administration
and Computer Science
Aichi Institute of Technology
Katsuhiko Kaji
Faculty of Information Science
Aichi Institute of Technology
Study Introduction
Launch
Application
A Sta. C Sta.
10:45Dep.Time
SEARCH
Searching
We consider 
about the Train
Automatically
When will I arrival
at My Destination Station?
• Departure Station
• Departure Time
• Arrival Station
Input Information
None
Propose Method
Transit App We don’t remember
the Departure Time
Hurry up!
Input Information
We don’t consider  
about the Train
Mobility as a Service (MaaS)
This is a concept that provides Various Transportation Services
through a Single Interface
Before
Ferry App Railway App Bus App
ServiceService Service
After
Mobility App
Service
Users get Information
from Each Application
Users get Information
from One Application
Relationship between MaaS and This Study
Users receives the Arrival Time through a Single Interface
Before
Ferry App Railway App Bus App
Arrival Information
After
Mobility App
Users considers
about Transportations
Users doesn’t considers
about Transportations
Arrival Information
Automatic
Estimation
I’m on the train
now
I want to know
the arrival time
If Users wants to know the Arrival Time
Related Work
An online localization method for a subway train utilizing
the barometer on a smartphone [Hyuga et al,. 2016.]
A Sta. B Sta. C Sta.
Subway Tracks
Between
A Sta.
and
B Sta.
Barometer 1010mb
Application analyzes Barometric Sensor
Log data and measured Values of Barometric Sensor
Log Data
Measured Values
*Sample
Related Work
An online localization method for a subway train utilizing
the barometer on a smartphone [Hyuga et al,. 2016.]
Unexpected Environment
High-Class Train
Rapid Train
Overtaking Train
Pass
Parallel Section
A Line
B Line
System Abstract
User Web Server
Database
Movement
History SQL
Result
Geographic
Calculation (PostGIS)
Rapid
ASta.
16:30
Dsta.
16:40
• Geographic data uses National Land Numerical Information
• Train Information is saved in GTFS format
Arrival
Information
{time,lon,lat},
{time,lon,lat},
{time,lon,lat},
• Line Geo Data
• Station Geo Data
• Time Schedule
• Train Information
1. Line Estimation
2. Direction Estimation
3. Tran Estimation
4. Fetch Arrival Time
Application can support Other Public Transportation in the future
System Abstract
User Web Server
Database
Movement
History SQL
Result
Geographic
Calculation (PostGIS)
Rapid
ASta.
16:30
Dsta.
16:40 Arrival
Information
{time,lon,lat},
{time,lon,lat},
{time,lon,lat},
• Line Geo Data
• Station Geo Data
• Time Schedule
• Train Information
1. Line Estimation
2. Direction Estimation
3. Tran Estimation
4. Fetch Arrival Time
• Geographic data uses National Land Numerical Information
• Train Information is saved in GTFS format
Application can support Other Public Transportation in the future
Calculate the likelihood of the Line
We calculate the Number that is in the Allowable Range ( )w
The number of inp w = x Total number of Geo Points = n
We set the to 140m from the Log Data of the Movement Historyw
It can cover the shift of the Movement History while stopped
Line Likelihood
l =
x
n(Range )0.0 ≤ l ≤ 1.0
Calculate this likelihood
for All Lines
Railway
p1 p2
p3
p4
p5
p6
w
Geo Point = p
Movement History by GPS of Smartphone
Processing in Parallel Sections
Parallel Section
A Line
B Line
If there are Multiple Lines with High Likelihood
Those Lines are also Provided
However, These cases need not be provided
• Branch from Another Lines to a the Single Line
• Merge from the Single Line to Another Lines
ld = lmax − la
• Highest Likelihood
• Likelihood of Any Line
lmax
la
When is greater than a Fixed Value
Those Lines are not also provided
ld
Therefore, we set a Threshold Value
Merge
Branch
Sta.
Sta.
Line A
LineB
System Abstract
User Web Server
Database
Movement
History SQL
Result
Geographic
Calculation (PostGIS)
Rapid
ASta.
16:30
Dsta.
16:40 Arrival
Information
{time,lon,lat},
{time,lon,lat},
{time,lon,lat},
• Line Geo Data
• Station Geo Data
• Time Schedule
• Train Information
1. Line Estimation
2. Direction Estimation
3. Tran Estimation
4. Fetch Arrival Time
• Geographic data uses National Land Numerical Information
• Train Information is saved in GTFS format
Application can support Other Public Transportation in the future
First
Sta.
p1
p3
p4
p2 Last
Sta.
First Half r1
Average Degree of Progress
r2
r1 < r2 For Last Station⋯ r1 > r2 For First Station⋯
r1 = r2 ⋯ We redo the Estimation
Direction Estimation
Second Half
Average Degree of Progress
r2
r1
r3
r4
Geo Point
We calculate the Degree of Progress ( ) for the Line using PostGISr
(0.0 ≤ r ≤ 1.0)… Percentage of the Train running from the First Station to the Liner
System Abstract
User Web Server
Database
Movement
History SQL
Result
Geographic
Calculation (PostGIS)
Rapid
ASta.
16:30
Dsta.
16:40 Arrival
Information
{time,lon,lat},
{time,lon,lat},
{time,lon,lat},
• Line Geo Data
• Station Geo Data
• Time Schedule
• Train Information
1. Line Estimation
2. Direction Estimation
3. Tran Estimation
4. Fetch Arrival Time
• Geographic data uses National Land Numerical Information
• Train Information is saved in GTFS format
Application can support Other Public Transportation in the future
Train Estimation
Sta Sta Sta
Railway
Estimated Position Estimated Position
Sta Sta Sta
Railway
Estimated Position Estimated Position
Currently
Position
d2
We calculate the Estimated Position of the Running Train on the Line
We calculate the Distance between the Estimated Position and the Current Position
These calculations are performed in Time Series on the Movement History
We calculate the RMS of Time Series of Distance We convert RMS into Likelihood
Boarding Train = Highest Likelihood Train
d1
How to calculate the Estimated Position?
B Sta. C Sta.
Railway
Accelerating DeceleratingConstant Speed
We propose a Model that estimates the Position of a Train
in consideration of Acceleration and Deceleration
A Sta. B Sta. C Sta.
Railway
I’m running between B sta. and C sta.
10:40 10:45
Time : 10:42
10:35
We estimate the Position between Stations based on Time
We consider the Movement of a Train
Total Mileage
vmax(a + 2r + d)
2
s =
Train Movement Model
Speed
Time
vmax
0
Departure
Time
as ae ds de
p1 p2a dr
AcceleratingStop Time Constant Speed Decelerating Stop Time
Arrival
TimeStart End Start End
Calculate the Estimated Position
vmax
v0
as ae ds de
p1 p2a dr
AcceleratingStop Time Constant Speed Decelerating Stop Time
Speed
Time
Departure
Time
Arrival
Time
t
Estimated
Position
m =
asecond(t − as)2
2
(as < t < ae)
vmaxa
2
+ vmax(t − ae) (ar ≤ t ≤ ds)
s −
esecond(de − t)2
2
(ds < t < de)
Distance from
Departure Station
Evaluation Experiment
We calculate Train Likelihood Ranking (in the Line)
We calculate the Percentage of Boarding Train within the 1st and 2nd
Implemented Section
We randomly generated a Movement History for 180 seconds
60 Samples (Up and Down Line) Only High-Class Cars
Nagoya Railway
Nagoya Line
Kintetsu Railway
Kyoto Line
Kintetsu Railway
Osaka Line
Calculate Percentage in Time Series
100%
90.0%
We calculated the Percentage every second
Because we want to observe the Time Series Change of the Percentage
Time Series Likelihood Ranking Graph
96.2%
Consideration
We confirmed that the Percentage
of being within 2nd was 100%
The Application provides a Train within 2nd 
(in the Line)
We confirmed that the Percentage increased in Time Series
After a while, Accuracy Increases
Summary
We propose An Application that Automatically Provides
Transportation Arrival Information
Future Tasks
• We examine Estimation Methods in the Underground Section
• We examine Estimation Methods for Other Vehicles
We confirmed that the Percentage of being within 2nd was 100%
We confirmed that the Percentage increased in Time Series
We estimate the Train on Board from the Movement History

A Smartphone Application that Automatically Provides Arrival Time based on Estimation of Train on Board

  • 1.
    A Smartphone Application thatAutomatically Provides Arrival Time based on Estimation of Train on Board Haruki Inoue Graduate School of Business Administration and Computer Science Aichi Institute of Technology Katsuhiko Kaji Faculty of Information Science Aichi Institute of Technology
  • 2.
    Study Introduction Launch Application A Sta.C Sta. 10:45Dep.Time SEARCH Searching We consider  about the Train Automatically When will I arrival at My Destination Station? • Departure Station • Departure Time • Arrival Station Input Information None Propose Method Transit App We don’t remember the Departure Time Hurry up! Input Information We don’t consider   about the Train
  • 3.
    Mobility as aService (MaaS) This is a concept that provides Various Transportation Services through a Single Interface Before Ferry App Railway App Bus App ServiceService Service After Mobility App Service Users get Information from Each Application Users get Information from One Application
  • 4.
    Relationship between MaaSand This Study Users receives the Arrival Time through a Single Interface Before Ferry App Railway App Bus App Arrival Information After Mobility App Users considers about Transportations Users doesn’t considers about Transportations Arrival Information Automatic Estimation I’m on the train now I want to know the arrival time If Users wants to know the Arrival Time
  • 5.
    Related Work An onlinelocalization method for a subway train utilizing the barometer on a smartphone [Hyuga et al,. 2016.] A Sta. B Sta. C Sta. Subway Tracks Between A Sta. and B Sta. Barometer 1010mb Application analyzes Barometric Sensor Log data and measured Values of Barometric Sensor Log Data Measured Values *Sample
  • 6.
    Related Work An onlinelocalization method for a subway train utilizing the barometer on a smartphone [Hyuga et al,. 2016.] Unexpected Environment High-Class Train Rapid Train Overtaking Train Pass Parallel Section A Line B Line
  • 7.
    System Abstract User WebServer Database Movement History SQL Result Geographic Calculation (PostGIS) Rapid ASta. 16:30 Dsta. 16:40 • Geographic data uses National Land Numerical Information • Train Information is saved in GTFS format Arrival Information {time,lon,lat}, {time,lon,lat}, {time,lon,lat}, • Line Geo Data • Station Geo Data • Time Schedule • Train Information 1. Line Estimation 2. Direction Estimation 3. Tran Estimation 4. Fetch Arrival Time Application can support Other Public Transportation in the future
  • 8.
    System Abstract User WebServer Database Movement History SQL Result Geographic Calculation (PostGIS) Rapid ASta. 16:30 Dsta. 16:40 Arrival Information {time,lon,lat}, {time,lon,lat}, {time,lon,lat}, • Line Geo Data • Station Geo Data • Time Schedule • Train Information 1. Line Estimation 2. Direction Estimation 3. Tran Estimation 4. Fetch Arrival Time • Geographic data uses National Land Numerical Information • Train Information is saved in GTFS format Application can support Other Public Transportation in the future
  • 9.
    Calculate the likelihoodof the Line We calculate the Number that is in the Allowable Range ( )w The number of inp w = x Total number of Geo Points = n We set the to 140m from the Log Data of the Movement Historyw It can cover the shift of the Movement History while stopped Line Likelihood l = x n(Range )0.0 ≤ l ≤ 1.0 Calculate this likelihood for All Lines Railway p1 p2 p3 p4 p5 p6 w Geo Point = p Movement History by GPS of Smartphone
  • 10.
    Processing in ParallelSections Parallel Section A Line B Line If there are Multiple Lines with High Likelihood Those Lines are also Provided However, These cases need not be provided • Branch from Another Lines to a the Single Line • Merge from the Single Line to Another Lines ld = lmax − la • Highest Likelihood • Likelihood of Any Line lmax la When is greater than a Fixed Value Those Lines are not also provided ld Therefore, we set a Threshold Value Merge Branch Sta. Sta. Line A LineB
  • 11.
    System Abstract User WebServer Database Movement History SQL Result Geographic Calculation (PostGIS) Rapid ASta. 16:30 Dsta. 16:40 Arrival Information {time,lon,lat}, {time,lon,lat}, {time,lon,lat}, • Line Geo Data • Station Geo Data • Time Schedule • Train Information 1. Line Estimation 2. Direction Estimation 3. Tran Estimation 4. Fetch Arrival Time • Geographic data uses National Land Numerical Information • Train Information is saved in GTFS format Application can support Other Public Transportation in the future
  • 12.
    First Sta. p1 p3 p4 p2 Last Sta. First Halfr1 Average Degree of Progress r2 r1 < r2 For Last Station⋯ r1 > r2 For First Station⋯ r1 = r2 ⋯ We redo the Estimation Direction Estimation Second Half Average Degree of Progress r2 r1 r3 r4 Geo Point We calculate the Degree of Progress ( ) for the Line using PostGISr (0.0 ≤ r ≤ 1.0)… Percentage of the Train running from the First Station to the Liner
  • 13.
    System Abstract User WebServer Database Movement History SQL Result Geographic Calculation (PostGIS) Rapid ASta. 16:30 Dsta. 16:40 Arrival Information {time,lon,lat}, {time,lon,lat}, {time,lon,lat}, • Line Geo Data • Station Geo Data • Time Schedule • Train Information 1. Line Estimation 2. Direction Estimation 3. Tran Estimation 4. Fetch Arrival Time • Geographic data uses National Land Numerical Information • Train Information is saved in GTFS format Application can support Other Public Transportation in the future
  • 14.
    Train Estimation Sta StaSta Railway Estimated Position Estimated Position Sta Sta Sta Railway Estimated Position Estimated Position Currently Position d2 We calculate the Estimated Position of the Running Train on the Line We calculate the Distance between the Estimated Position and the Current Position These calculations are performed in Time Series on the Movement History We calculate the RMS of Time Series of Distance We convert RMS into Likelihood Boarding Train = Highest Likelihood Train d1
  • 15.
    How to calculatethe Estimated Position? B Sta. C Sta. Railway Accelerating DeceleratingConstant Speed We propose a Model that estimates the Position of a Train in consideration of Acceleration and Deceleration A Sta. B Sta. C Sta. Railway I’m running between B sta. and C sta. 10:40 10:45 Time : 10:42 10:35 We estimate the Position between Stations based on Time We consider the Movement of a Train
  • 16.
    Total Mileage vmax(a +2r + d) 2 s = Train Movement Model Speed Time vmax 0 Departure Time as ae ds de p1 p2a dr AcceleratingStop Time Constant Speed Decelerating Stop Time Arrival TimeStart End Start End
  • 17.
    Calculate the EstimatedPosition vmax v0 as ae ds de p1 p2a dr AcceleratingStop Time Constant Speed Decelerating Stop Time Speed Time Departure Time Arrival Time t Estimated Position m = asecond(t − as)2 2 (as < t < ae) vmaxa 2 + vmax(t − ae) (ar ≤ t ≤ ds) s − esecond(de − t)2 2 (ds < t < de) Distance from Departure Station
  • 18.
    Evaluation Experiment We calculateTrain Likelihood Ranking (in the Line) We calculate the Percentage of Boarding Train within the 1st and 2nd Implemented Section We randomly generated a Movement History for 180 seconds 60 Samples (Up and Down Line) Only High-Class Cars Nagoya Railway Nagoya Line Kintetsu Railway Kyoto Line Kintetsu Railway Osaka Line
  • 19.
    Calculate Percentage inTime Series 100% 90.0% We calculated the Percentage every second Because we want to observe the Time Series Change of the Percentage Time Series Likelihood Ranking Graph 96.2%
  • 20.
    Consideration We confirmed thatthe Percentage of being within 2nd was 100% The Application provides a Train within 2nd  (in the Line) We confirmed that the Percentage increased in Time Series After a while, Accuracy Increases
  • 21.
    Summary We propose AnApplication that Automatically Provides Transportation Arrival Information Future Tasks • We examine Estimation Methods in the Underground Section • We examine Estimation Methods for Other Vehicles We confirmed that the Percentage of being within 2nd was 100% We confirmed that the Percentage increased in Time Series We estimate the Train on Board from the Movement History