Public-Transit Data Extraction From OpenStreetMap Data · OpenStreetMap Data Zhiwei Zhang February...

IntroductionBackground

ProcedureEvaluation

Further Research

Public-Transit Data Extraction FromOpenStreetMap Data

Zhiwei Zhang

February 13, 2017

Reviewer: Prof. Dr. Hannah Bast

Supervisor: Prof. Dr. Hannah Bast

Tutor: Patrick BrosiZhiwei Zhang Public-Transit Data Extraction From OpenStreetMap Data


ProcedureEvaluation

Further Research

Table of Contents

1 Introduction2 Background3 Procedure

Data ExtractionRepair System

Oder RepairGap RepairClassificationConnection RepairTopological Sort

Generating GTFS Feed4 Evaluation5 Further Research

Zhiwei Zhang Public-Transit Data Extraction From OpenStreetMap Data


ProcedureEvaluation

Further Research

Introduction

1 OpenStreetMap?

2 Meaning of topic?



ProcedureEvaluation

Further Research

Introduction

1 OpenStreetMap?an open-source project providing free geographic data

2 Meaning of topic?



ProcedureEvaluation

Further Research

Introduction

1 OpenStreetMap?an open-source project providing free geographic data

2 Meaning of topic?Provide data-support for route planing softwares



ProcedureEvaluation

Further Research

Table of Contents







ProcedureEvaluation

Further Research

Mapping Relation



ProcedureEvaluation

Further Research

Data ExtractionRepair SystemGenerating GTFS Feed

Table of Contents







ProcedureEvaluation

Further Research


Brief



ProcedureEvaluation

Further Research


Data Extraction

OSM Data

Node

Way

Relation



ProcedureEvaluation

Further Research


Data Extraction

OSM Data

Node

Way

Relation

route=railway/route=rail/

route=train/route=tracks



ProcedureEvaluation

Further Research


Data Extraction

OSM Data

Node

Way

Relation

Network

Vertex

Edge

Graph





ProcedureEvaluation

Further Research


Data Extraction

OSM Data

Node

Way

Relation

Network

Vertex

Edge

Graph



in Networkor

railway=rail/railway=

narrow gauge



ProcedureEvaluation

Further Research


Data Extraction

OSM Data

Node

Way

Relation

Network

Vertex

Edge

Graph



in Networkor

railway=rail/railway=

narrow gauge

in Network



ProcedureEvaluation

Further Research


Repair System

before: class: unknown

E1

E 2

E5E4E3

after: class: local

E1

E 3

E5 E6

E7E8

E 4

E2



ProcedureEvaluation

Further Research


Oder Repair - Introduction

: not contained in this graph

: contained in this graph

E4E 1

E2

E3

E5

Original Edge Order (containing fake gaps):

E1 → E2 → E3 ⇢ E4 ⇢ E5

⇢ : means a gap.



ProcedureEvaluation

Further Research


Order Repair - fixing gap between E4 and E5

E4 E5



ProcedureEvaluation

Further Research



bidirectional A*

search for path

E4 E5



ProcedureEvaluation

Further Research



bidirectional A*

R-Tree(local edges)

search for neighbors

search for path

E4 E5



ProcedureEvaluation

Further Research



bidirectional A*

R-Tree(local edges)

return neighborssearch for neighbors

search for path

E4 E5



ProcedureEvaluation

Further Research



bidirectional A*

R-Tree(local edges)


insert found pathsearch for path

E4

E1

E2 E3 E5



ProcedureEvaluation

Further Research



fake gap

real gap

bidirectional A*

R-Tree(local edges)



E4

E1

E2 E3 E5



ProcedureEvaluation

Further Research



fake gap

real gap

bidirectional A*

R-Tree(local edges)



E4

E1

E2 E3 E5

Edge Order (after using bidirectional A*):E1 → E2 → E3 → E2 → E1 → E4 → E1 → E2 → E3 ⇢ E5



ProcedureEvaluation

Further Research



fake gap

real gap

bidirectional A*

R-Tree(local edges)



E4

E1

E2 E3 E5

Edge Order (after Order Repair):E4 → E1 → E2 → E3 ⇢ E5



ProcedureEvaluation

Further Research


Oder Repair - Procedure

Algorithm 1: Order RepairInput: a network contains graphs, edges and vertexs

1 foreach graph ∈ network do2 gap list ← GapCheck(graph );3 r-tree ← BuildRTreeUsingLocalEdges(graph );4 foreach gap ∈ gap list do5 GapRepairUsingBiAStar(r-tree, graph, gap );

/* find the first non-repetitive edge and delete edges front

it

make sure the front end-edge is in position */

6 LocateFirstEdge(graph );



ProcedureEvaluation

Further Research


Oder Repair - Explanation

Questions:1 repetitive edges?

key to keep the local order rightremovable through DFS

2 theoretical basis of LocateFirstEdge?the front end-edge intersects with its neighbors at oneidentical conjoint point

3 why bidirectional A* algorithm?fast and efficientif a complete path doesn’t exist, highly possible to find apartial path to transform a fake gap into a real one



ProcedureEvaluation

Further Research


Gap Repair - Introduction

: not contained in this graph

: contained in this graph

E4E 1

E2

E3

E5

Original Edge Order (containing real gaps):

E4 → E1 → E2 → E3 ⇢ E5

⇢ : means a gap.



ProcedureEvaluation

Further Research



E4E 1

E2

E3

E5

path finding

search for path



ProcedureEvaluation

Further Research



E4

E 1

E2

E3

E5

path finding

R-Tree(global edges)

search for neighbors

search for path



ProcedureEvaluation

Further Research



E4

E 1

E2

E3

E5

path finding



search for path



ProcedureEvaluation

Further Research



E4

E 1

E2

E3

E6E5

path finding






ProcedureEvaluation

Further Research



E4E 1

E2

E3

E6E5

Edge Order (after Gap Repair):

E4 → E1 → E2 → E3 → E6 → E5



ProcedureEvaluation

Further Research


Gap Repair - Procedure

Algorithm 2: Gap RepairInput: a network contains graphs, edges and vertexs

1 gap list dic ← empty dictionary; // key is graph , value is a gap list2 foreach graph ∈ network do3 gap list ← GapCheck(graph );4 gap list dic .Add(graph, gap list );5 r-tree ← BuildRTreeUsingGlobalEdges(network ); // all the edges in network6 foreach (graph, gap list) ∈ gap list dic do7 foreach gap ∈ gap list do

/* GapRepair returns true/false to indicate if a gap is successfully fixed

pathFindingAlgo could be A* Algorithm or Dijkstra’s Algorithm */

8 IsFixed ← GapRepair(r-tree, graph, gap, pathFindingAlgo );9 if ¬IsFixed then

10 break;



ProcedureEvaluation

Further Research


Classification - Introduction

Graph

service : expressref : ICE255

Graph

service : regionalref : RB1101

Graph

service : nullref : ICE255

Graph

service : nullref : null



ProcedureEvaluation

Further Research



Graph


Graph


Graph


Graph


long-distancelocal unknown



ProcedureEvaluation

Further Research



Graph


Graph


Graph


Graph


Dic{ice:long-distance,rb:local }



ProcedureEvaluation

Further Research



Graph


Graph


Graph


Graph


SVMs-training model

length=500 km length=100 km



ProcedureEvaluation

Further Research



Graph


Graph


Graph


Graph

service : regionalref : null

SVMs-predict model

length=90 km



ProcedureEvaluation

Further Research


Classification - ProcedureAlgorithm 3: ClassificationInput: a network contains graphs, edges and vertexs

1 locals , long distances , unknows ← empty list;2 foreach graph ∈ network do3 if IsLocal(graph ) then4 locals .Add(graph );5 else if IsLongDistance(graph ) then6 long distances .Add(graph );7 else8 unknows .Add(graph );

/* get a dictionary, in which one prefix points to only one class label local or

long-distance */

9 ref predix dic ← SummarizeUniqueRefPredix(locals, long distances );10 PredictUsingDic(ref predix dic, locals, long distances, graph );11 svms ← SVMs();12 svms .Training(locals, long distances ); // train SVMs with the lengths

13 PredictUsingSVMs(svms, graph );



ProcedureEvaluation

Further Research


Connection Repair - Introduction

E1

E 2 E3 E4

E5

Edge Order:E1 → E2 → E3 → E4 ⇢ E5

Vertex Order:E1: v1 → v2 → v3

E2: v3 → v4 → v5 → v6 → v7

E3: v5 → v8 → v9

E4: v11 → v10 → v9

E5: v13 → v14 → v15



ProcedureEvaluation

Further Research


Connection Repair - Introduction

E1

E 7

E3 E4

E 6

Edge Order:E1 → E7 → E6 → E3 ⇢ E4

Vertex Order:E1: v1 → v2 → v3

E7: v3 → v4 → v5

E3: v5 → v8 → v9

E4: v9 → v10 → v11

E6: v5 → v6 → v7



ProcedureEvaluation

Further Research


Connection Repair - Procedure

Algorithm 4: Connection RepairInput: a network contains graphs, edges and vertexs

1 foreach graph ∈ network do2 foreach edge front, edge back ∈ graph do

/* edge front and edge back are index-adjacent in graph */

3 if IsConjoint(edge front, edge back ) then4 if ¬IsHeadTailType(edge front, edge back ) then

/* make sure they are conjoint in head-tail type. If

necessary, separate one of them to achieve it */

5 SepatateEdge(edge front, graph ) orSepatateEdge(edge back, graph );

6 else7 foreach edge ∈ graph ⋀ edge behind edge back do8 graph .DeleteEdge(edge );



ProcedureEvaluation

Further Research


Topological Sort - Introduction

Problems:repetitive edges exist in graph

edges in graph are poor organized

Aim: each graph has a topologically sorted edge-list.

Algorithm 5: Topological SortInput: a network contains graphs, edges and vertexs

1 foreach graph ∈ network do2 adjacency list ← CreateAdjacencyLists(graph );3 DFS(adjacency list, graph );



ProcedureEvaluation

Further Research


Generating GTFS Feed - Introduction



ProcedureEvaluation

Further Research


Generating GTFS Feed - Procedure

Algorithm 6: Generating GTFS FeedInput: a network contains graphs, edges and vertexsOutput: a GTFS feed

1 GTFSSystem← GTFSSystem();2 r-tree ← BuildRTreeUsingGlobalStations(network );3 foreach graph ∈ network do4 route ←Route(graph ); // store the attributes of graph5 GTFSSystem .Add(route );6 railway list← SeparateIntoRailways(graph ); // create several

railways, each of them could store the information of a branch of graph7 foreach railway ∈ railway list do

/* separate the information in a railway into stop s, trip s and shape*/

8 ProcessRailway(GTFSSystem, railway, r-tree );

9 OutputGTFSFeed(GTFSSystem ); // output a GTFS feed using a CSV writer



ProcedureEvaluation

Further Research

Table of Contents







ProcedureEvaluation

Further Research

Evaluation - Part I

Conclusions:

using A* ⇐⇒ usingDijkstraOrder Repair isefficient in removinggaps.



ProcedureEvaluation

Further Research

Evaluation - TRAVIC I



ProcedureEvaluation

Further Research

Evaluation - TRAVIC II



ProcedureEvaluation

Further Research

Evaluation - OTP I



ProcedureEvaluation

Further Research

Evaluation - OTP II



ProcedureEvaluation

Further Research

Evaluation - Part II

Conclusions:

result in TRAVIC is very good.result in OTP is ok. For some routes is fine.

Analysis:

We take the condition into account, that some stops belongto a big station. So the GTFS feed contains this information.OTP doesn’t recognize this information in the GTFS feed. Itleads to a not good result.



ProcedureEvaluation

Further Research

Table of Contents







ProcedureEvaluation

Further Research

Further Research

1 Extending network by collecting all the ways near to at leastone of the railways

2 Improving the efficiency of classification by precomputingtraining data

3 Identifying neighbors with regard to the existence of a real gap



ProcedureEvaluation

Further Research

Thank you for your attention!



ProcedureEvaluation

Further Research

Question?


Public-Transit Data Extraction From OpenStreetMap Data · OpenStreetMap Data Zhiwei Zhang February...

Documents

Transcript of Public-Transit Data Extraction From OpenStreetMap Data · OpenStreetMap Data Zhiwei Zhang February...