Pedestriandynamics:ExperimentandSimulation22October 2015|Dr.MaikBoltes

ForschungszentrumJülich,Germany

DivisionCivil Securityand Traffic§ ForschungszentrumJülich(ResearchCentre)Researchonenergyandenvironment,informationandbrain~5500staff member

§ Institut„JülichSupercomputing Centre“(JSC)Operatingsupercomputerofthehighestperformanceclass;simulatingcomplexproblems~200staff member

§ Division„Civil Securityand Traffic“18staff member (including 5Ph.D.students,3graduate students)http://www.fzj.de/ias/jsc/cst

3

DivisionCivil Securityand Traffic§ Interdisciplinaryteamfocusingonmodelsandsimulationsofcomplexsystemswithapplicationsincivilsecurityandtrafficplanning

Pedestrian Dynamics,Experiments and AnalysisPerforming experiments forquantitative description of pedestrian dynamics

Pedestrian Dynamics andTraffic SimulationModelling of pedestrian dynamics, large scale evacuation, intermodal transport and mixed traffic

Fire SimulationModelling of fire and smoke spread in buildings or in common carriers using CFD methods

Overview

§ Motivation§ Laboratoryexperiments§ Automaticextractionoftrajectories§ Analysisresultsbyexample

§ Simulation§ Validationandverification§ Openaccess

Motivation§ Exemplarydisasters duetocrowdcrushes:

§ Mainreason:highpedestriandensitiesduetoinsufficientpedestrianfacilitiesorinadequatecrowdmanagement

Year Place Occasion Deaths

1896 Moscow, Chodynka Feld (Russia) Coronation of Nicholas II 1389

1989 Sheffield, Hilborough-Stadion (England) Soccer game 96

1990 Mecca, Al-Ma'aisim tunnel (Saudi Arabia) Pilgrimage, Hajj 1426

2006 Mecca, Jamarat-Brücke (Saudi Arabia) Pilgrimage, Hajj 364

2010 Duisburg (Germany) Festival, Love Parade 21

2010 Phnom Penh (Cambodia) Festival, Khmer Water 378

2015 Shanghai (China) New Year 36

2015 Mecca, Mina (Saudi Arabia) Pilgrimage, Hajj 2217

[wikipedia.org/wiki/List_of_accidents_and_disasters_by_death_toll]

Motivation§ Exemplarydisasters duetocrowdcrushes:

§ Mainreason:highpedestriandensitiesduetoinsufficientpedestrianfacilitiesorinadequatecrowdmanagement

Motivation§ Aims:Increaseof

§ Safety,e.g.Designofescaperoutes(stadiums,theaters,schools,…)

§ Comfort,e.g.Designoftransportinfrastructures(stations,airports,trains,…)

§ Toolsandmethods:§ Legalregulations(prescriptivemethod)§ Guidelinesandhandbooks(macroscopicmodels)§ Computersimulations(microscopicmodels)

MotivationKnowledgeofpedestriandynamicsisinsufficient

→ Largeuncertainties ofquantitativeresultsfromsimulationsoftware

Evacuationtimeforacorridor:

[C. Rogsch]

MotivationKnowledgeofpedestriandynamicsisinsufficient

→ Non-negligibledifferencesinguidelines

[A. Seyfried]

ContradictionofdataMayresultfrom:§ Directionofflow(e.g.uni- andbidirectional)

§ Measurementmethod§ Culture§ Populationdemographics§ Psychologicalfactors(e.g.motivation)§ …

[A. Seyfried]

Motivation

Reliableempiricaldata ofpedestrianmovementarenecessary

→analyze pedestriandynamicsfor

→modeldesign→modelcalibration→modelvalidation

Microscopicmodelsneedmicroscopicdata[M. Chraibi][A. Portz]

Motivation

Laboratoryexperiments

LaboratoryexperimentsYear Place #Pers. #Runs Extraction Marker

2005 G– Jülich 34 6 manually

2005 G– Jülich 60 18 semi-automatically

2006 G– Düsseldorf 200 99 automatically

2008 India 64 6 manually

2008 G– Wuppertal 50 14 semi-automatically

2009 G– Düsseldorf 350 170 automatically

2012 G– Wuppertal 33 10 automatically

2013 Japan 28 260 automatically

2013 G– Düsseldorf 926 190 automatically2014&2015 G– Wuppertal 51 132 automatically

2015 Australia 200 110 automatically

2015 China 71 36 automatically

In artificial environments

Laboratoryexperiments

In real environments

Laboratoryexperiments

LaboratoryexperimentsAdvantages:§ Selectiveanalysis ofparameterswithoutundesiredinfluences

§ Variability oftheexperimentalprocedure

§ Generationofhighdensitiesseldomseeninfieldstudies

§ Possibilitytoextractmicroscopictrajectoriesathighqualityinspaceandtime

Goalofextraction§ Individualtrajectories§ Lowspatialerror§ Nearlynodetectionerror

§ Calculabilityofallquantitiesdescribingthepedestriandynamicsatanytimeandanyplace

→

Automaticextractionofindividualtrajectories

Opticalflow /Movementof crowds

[B. Krausz] [S. Ali]

Motioncapturing /Movementof body parts

[S. Schmidt] [M. Becker]

Detection of people

TemplatematchingHistogramDeformablepart-basedoforientedgradientsmodel

[D. M. Gavrila] [E. P. Fotiadis] [Q. Delamarre]

Missrate

[P. Dollar]

Missrate–Influence ofdetection errors

[P. Dollar]

Markersenables robustextraction

[S. P. Hoogendoorn] [W. Daamen] [S. Saadat]

[X. Liu] [W. Song][S. Wong]

Markersenables robustextraction

[M. Bukacek]

[M. Bukacek][D. Stuart]

1.Calibration 2.Detection

3.Tracking 4.HeightDetection

Extraction of Trajectories

Calibration

Undistortion due to distorting objectives:§ Model of pinhole camera with distortion maps

homogenous real world coordinate to pixel coordinate :§ Intrinsic param.: focal length , distortion center§ Extrinsic param.: position , orientation of the

camera§ Distortion coefficients: tangential , radial

[ D. C. Brown, Close-range camera calibration ]

[ Z. Zhang, A flexible new technique for camera calibration ]

Calibration (intrinsic)

2D/3D point correspondences / calibration points§ At least 4 point correspondences needed§ 3D points not all in one plane§ Iterative algorithm calculates the translation and the rotation of the camera

Calibration (extrinsic)

Field of view influenced by § Focal length of lens§ Pitch of camera§ Height of ceiling

Calibration (multiple views)

§ Largerobservationspace§ Smallerfieldofview→lessocclusion§ Complexcalibration

Calibration (overlapping camera grid)

Detection§ Orientedisolinesof samebrightness

§ Approxima-tion withellipsis

§ Identificationby size,po-sition,shape

Detection§ Colored hats§ HSVcolor space§ Closing,opening

DetectionApplication depending marker:Marker Robustness

DetectionAccuracyPosition

HighDensity

AdditinalInformation

NeededResolution

ExpenseProduction

++ ++ o + o o

o ++ + - + +

+ ++ - o + o

+ o - o ++ ++

o ++ + ++ - -

TrackingTranslationof of pixelregion insuccessive frame

Determine positionHeadcamera distance

Hue

Saturatio

n

Determine positionPositionerror caused byheight difference

Disparity betweenpoint of views of stereo camerafor measuring head distance

Determine position

Disparity betweenpoint of views of stereo camerafor measuring head distance

Determine position

Analysisresults by example

Smoothing§ Withstepdetection,frequencyanalysis,heightwaves,convexhulls,…

§ Foreliminationoflocalpositionerrororswaying

x [cm]

Movement direction x [cm]

Persons’ height

vertical (size)lateral (swaying)

Smoothing§ Withstepdetection,frequencyanalysis,heightwaves,convexhulls,…

§ Foreliminationoflocalpositionerrororswaying

x [cm]

y [cm]

Voronoi density ( )( ) iA iVoronoi

p x dx : x AAp xA : else

⎧ ∈⎪ρ = = ⎨

⎪⎩

∫r r r

r 1 mit

0

§ Velocity

§ Density

§ Specificflow

Profileof quantities

yx

dxdyxyv Δ⋅Δ=>< ∫∫ρρ

yx

dxdyvv xy

v Δ⋅Δ=>< ∫∫

Influence of culture to the velocity

[U. Chattaraj]

Influence of motivation to the velocity

[J. Lukowski]

Flowas afunction ofbottleneck width

Density profile§ Bidirectional

§ Unidirectional

§ Unidirectionalwith bottleneck

[J. Zhang]

Density infrontofbottleneck

[J. Liddle]

Specific flow profiles

l = 6 cm

l = 400 cm

Flowas afunction of bottleneck length

l = 400 cm

l = 0.06

l = 6 cm

UnidirectionalBidirectional

Flowas afunction of density /flow direction

T-junction

T-junctionExperimentandSimulation

[S. Burghardt]

Simulation

Pedestrian dynamics simulationOpensourceframeworkforpedestriandynamicssimulations“Jülich PedestrianSimulator”www.jupedsim.org

Simulationlevel

Strategic level

Tactical level

Operational level

Determination of the final destination and strategy to reach the goalParking, train stationQuickest path, known path

Determination of intermediate destinationsWalk, do not push, take left exit, lift,…

Locomotion systemGo, stop, slow down, swerve…

Tactical level

Goal and Strategy

Move to next destination

New Location /

Jam

Evaluate the situation based on the strategy

Change trip if necessary

Tactical level:Quickestpath based onvisibility

[Kemloh et al. “Modelling dynamic route choice of pedestrians to assess the criticality of building evacuation “, 2012, ACS.]

Tactical level:Informationpropagation

Tactical level:Reaction incase of fire

Operationallevel:Forcebasedmodelsincontinuousspace§ Equation of motion

§ Parallel update

Operationallevel:GeneralizedCentrifugalForceModel§ Velocity dependent space requirement

( )d v a b v= + ⋅

[Chraibi et al, ”Generalized centrifugal force model for pedestrian dynamics”, Physical Review E 82, 046111 (2010)]

Operationallevel:Gradientmodelfield

*Arne Graf, Automated Routing in Pedestrian Dynamics, Master thesis 2015.*J. A. Sethian. Fast Marching Methods. SIAM Rev. 41, 2, 199-235, 1999.*Dietrich et al. Gradient navigation model for pedestrian dynamics. Phys. Rev. E, 89, 062801, 2014.

Simulation

Hybridsimulation – coupling with macroscopic simulation

Verificationandvalidationofmodels

Verification and validation of models§ Verification

§ Makesuretheimplementationiscorrect§ Testiftheimplementationoftheunderlyingmodeliscorrect§ (Unittesting)

§ Validation§ Testtherealismofthemodel§ Comparisonwithempiricaldata§ Fundamentaldiagram,flow,fieldobservations,…

groups

Nesttest with PEDFLOW(PED14)

routing

bi-direct

const. velocity

corner

pre-movement time

[Michelle Isenhour, US Military Academy]

RiMEA (14tests)

Exit choice

Stairs

Route choice

Jülichbenchmarks

OpenAccess

§ Name:PeTrack(PEdestrian TRACKing)

§ URL(download,help):http://ped.fz-juelich.de/petrack

§ Language:English§ License:free§ OperatingSystems:Windows,Linux,…§ ProgrammingLanguage:C++§ UsedSoftwareLibraries:

§ UserInterface:Qt,Qwt§ ImageProcessing:OpenCV,Triclops (PointGrey)

CVSoftware

Database§ Afterfinishing thesis about experiments alldata (videos,trajectories,documentation)published publicly available

§ http://ped.fz-juelich.de/database

Pedestrian dynamics simulationOpensourceframeworkforpedestriandynamicssimulations“Jülich PedestrianSimulator”www.jupedsim.org

Maik Boltes Experiments, Technique, Extraction of trajectories

Thankyou!

Collaboration

[Company SL-RASCH, Saudi Arabia]

[University Tokio, Japan]

[Monash University, Melbourne, Australia]

Chowpatty beach, Mumbai, India [J. Sorabjee, Urban Design Res. Inst.]