Hub Map: a new approach for visualizing traffic data sets ...dro.deakin.edu.au/eserv/DU:30084280/avazpour-hubmap-post-2015.pdf · {asimmons,iavazpour,hvu,rvasa}@swin.edu.au Abstract—Visualizing

DRO Deakin Research Online, Deakin University’s Research Repository Deakin University CRICOS Provider Code: 00113B

Hub Map: a new approach for visualizing traffic data sets with multi-attribute link data

Citation: Simmons, Andrew, Avazpour, Iman, Vu, Hai L. and Vasa, Venkata Rajesh 2015, Hub Map: a new approach for visualizing traffic data sets with multi-attribute link data, in VL/HCC 2015 : Proceedings of the IEEE Symposium on Visual Languages and Human-Centric Computing, IEEE, Piscataway, N.J., pp. 219-223.

DOI: 10.1109/VLHCC.2015.7357220

© 2015, IEEE

Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

Downloaded from DRO: http://hdl.handle.net/10536/DRO/DU:30084280

Hub Map: A new approach for visualizing trafficdata sets with multi-attribute link data

Andrew Simmons, Iman Avazpour, Hai L. Vu, Rajesh VasaFaculty of Science, Engineering and Technology

Swinburne University of TechnologyHawthorn, VIC, Australia

{asimmons,iavazpour,hvu,rvasa}@swin.edu.au

Abstract—Visualizing road traffic datasets involves representingjunctions, their links, and the attributes of those links. Currenttraffic visualization techniques are not sufficient for professionaltraffic engineers, as they are limited in the number of attributesthat can be represented. This paper proposes a new approach tovisualize multiple attributes on graph edges without compromis-ing their visibility. In particular, we introduce a parameterizedconnector symbol that increases the number of attributes that canbe displayed on graph edges. We demonstrate that our approachcan significantly increase the number of traffic parameters thatcan be displayed compared to existing traffic visualizations.

I. INTRODUCTION

Effective management of road traffic congestion requires de-tailed inspection of a traffic network. Professional traffic engi-neers must understand characteristics of congestion, includingnetwork effects experienced from congested roads in the sur-rounding vicinity. To gain this understanding, various attributesof traffic data such as geospatial, directional, temporal, andvolumetric attributes must be presented together.

Current tools used by professional traffic engineers providevery little support for traffic network inspection. They areoften designed similar to tools targeted for general public.Specifically, they are limited in their visual representationof traffic parameters, i.e. only one (non-geospatial) trafficparameter is being represented by any given visualization. Thisrepresentation is insufficient to understand the characteristicsor cause of traffic congestion. Due to the ineffectiveness ofcurrent tools, traffic engineers must often resort to manuallyinspecting raw sensor data in the form of spreadsheets. Thisis not only cumbersome, but also limits traffic engineers’perspective to a single traffic junction, preventing a full un-derstanding of its interaction with the rest of the network. Inthis paper we introduce a new approach and connector symbolfor traffic network visualization that enables visualization ofall traffic junction parameters. Our approach maps multipletraffic parameters to symbol parameters, whilst preserving thespatial dimensions of the traffic network.

The rest of this paper is organized as follows. We provide amotivating scenario in section II followed by our approach inSection III. An evaluation of our visualization approach usingprinciples of Physics of Notation is provided in section IV.Finally we conclude the paper in section V.

II. MOTIVATING SCENARIO

Consider Mary, a traffic engineer, who is tasked with ensuringthat the traffic network operates efficiently despite an increas-ing number of road users. Her first task is to identify con-gestion hotspots within the network. The traffic managementsystem she is using provides her with color-coded map of links,which is very similar to the traffic overlay layer of GoogleMaps (Figure 1(a)) designed for general public. However, un-like the general public, Mary needs access to additional trafficparameters that can be used to see potential congestion beforethe situation escalates, e.g. saturation. Saturation measureshow close a link is to its maximum capacity. Unfortunately,color-coded maps can only display one congestion parameter ata time. As such, Mary needs to toggle back and forth betweenmultiple traffic overlays. Mary would like to prioritize whichcongestion hotspots to investigate first, but due to the limitedcapacity of human color perception, she cannot distinguishbetween close colors of the scale. Indeed, the color scaleused by the Google Maps traffic overlay consists of only fourdistinguishable colors.

Once zoomed into the vicinity of a hotspot, Mary’s second taskis to identify bottlenecks. This requires some detective workon her part, and she must make the most of all informationavailable to her. Mary uses a Geospatial Information System(GIS) tool to create a Chart Map (Figure 1(b)) that encodesthe traffic parameters at each junction as symbols. It has beenshown that it is possible to encode a large number of attributeswithin a single symbol, for example, by encoding them ina Chernoff face [1]. Mary identifies a junction with a veryuneven phase split, i.e. one of the links entering the junctionrequires a much longer share of green time than the others.However, as her Chart Map aggregates information at thejunctions (e.g. as a pie chart) rather than on the links, shestruggles to determine which green time in the junction symbolcorresponds to which link.

To investigate further, Mary needs to obtain detailed informa-tion of the links. She first refers to a diagram of the junctionto find which detectors belong to each link. She then needs tolook up the detector numbers in her spreadsheet or databaseto find the data for those particular detectors. Mary uses thismethod to quantify abnormalities, but caught up in numbercrunching, she is unable to form a complete picture of thesituation.

As it is usually not sufficient to examine any one junctionor link in isolation, Mary needs to identify network effects

(a) Color-coded traffic overlay visualization used byGoogle MapsTM. Map data c�2015 Google.

(b) Chart Map encoding green times in a pie chartsymbol at each junction. Traffic data c�VicRoads.Base map c�OpenStreetMap contributors.

(c) Flow Map encoding truck flow as width oflinks. c�US Department of Transportation.

Fig. 1. Selection of visual representations used in state of the art traffic systems.

from roads in the surrounding vicinity. Mary consults trafficflow maps (Figure 1(c)) that visualize flow as line thickness.An interesting variant of these flow diagrams are Sankeydiagrams where thickness makes up a significant proportion ofthe diagram, thus allowing values to be read with reasonableprecision. However, Sankey diagrams usually sacrifice spatialdimensions of the data due to their layout mechanisms. Sim-ilar to color-coded maps, flow maps only visualize a singleattribute of the links, which limits Mary’s ability to reasonabout the nature of the interactions between junctions.

Mary’s story represents everyday tasks performed by trafficengineers. We have summarized these tasks in Table I. In thispaper we present a new traffic network visualization that isoptimized to meet the needs of domain experts - i.e. highlytrained traffic engineers. Specifically, we address the questionof whether it is possible to include multiple edge attributes ingeospatial graph visualization without exceeding the cognitivelimitations of expert end users.

TABLE I. TASKS PERFORMED BY TRAFFIC ENGINEER

Task Search Area Required level of DetailIdentify congestionhotspots

Large (entirenetwork)

Low (general area, presence of con-gestion)

Identify bottlenecks Medium(vicinity)

Medium (precise locations, charac-teristics of congestion)

Identify network effects Medium(vicinity)

Medium (precise locations, charac-teristics of congestion)

Obtain detailed informa-tion

Small(detector)

High (all traffic parameters)

III. APPROACH

In this section we provide our approach to traffic network vi-sualization. We first describe our approach to modeling trafficparameters. Then we describe our visualization design and howit represents the parameters of this traffic model.

A. ModelingTraffic management systems operate on a set of fundamentaltraffic parameters that are collected from detector loops attraffic stop lines. These parameters are outlined in Table II. Dueto the placement of detectors, link parameters are measurednear the junctions, and are for the incoming traffic directiononly, traffic leaving the junction is not monitored.

Network congestion is a complex phenomenon and oftencan only be described by a combination of multiple trafficparameters [2]. For example, a traffic engineer must first know

TABLE II. COLLECTED TRAFFIC PARAMETERS

Parameter Unit AppliesTo

Description

TrafficVolume

Vehicles Link The number of vehicles that pass overthe link during the green time.

GreenTime

Seconds Link The amount of time dedicated to the linkfor each traffic cycle.

Flow Rate Vehicles perSecond

Link The number of vehicles that pass over thelink per second during the green time.

Max Flow Vehicles perSecond

Link The maximum flow rate physically pos-sible when the link is running at 100%efficiency.

Saturated True/False Link Whether the link is experiencing demandbeyond its capacity. Determined by theamount of time that there is a vehicleover the detector.

Link Ori-entation

Degrees Link Angle link approaches junction

CycleTime

Seconds Junction The amount of time spent performingone traffic cycle. This time will be splitbetween the links. Due to pedestriancrossings, and a short delay betweengreen times, the green times of the linksdo not necessarily add up to the cycletime of the junction.

Latitude Deg. North Junction Physical location of junction.Longitude Deg. East Junction Physical location of junction.

if a link is in a saturated state to determine whether low flow oftraffic indicates congestion, or simply lack of demand.

Some traffic parameters are interrelated, specifically,Traffic Volume = Flow Rate × Green Time. This interrelationaffects the visual representation and we seek to designthe visualization in such a way that the geometricalrelationships are consistent with the relationships betweentraffic parameters.

B. Visualization Design in Hub Map

An ideal visualization would aim to maximize the amountof information conveyed without overloading the cognitivecapabilities of the traffic engineer. Our approach is built withthis goal in mind. Achieving this requires taking full advantageof available visual design space. We spread our design acrossposition variables, orientation, color, shape, and size.

Our approach utilizes a novel connector symbol to representattributes for links, and a circular symbol to represent attributesfor junctions. Predefined connector symbols are common ingraph based visual languages and indicate distinct roles. Forexample, composition and generalization symbols used inUnified Modeling Language (UML). We extend this con-cept by proposing a parameterized connector symbol capableof visualizing continuous values of attributes. This symbol

utilizes a rectangular shape with values represented by itslengths, width, and color. Connector symbols are connected tojunctions with their connection angle representing the link’sorientation.

We have named this visualization approach “Hub Map” due tothe visualization’s portrayal of junctions as spoked connectionhubs. Table III provides our selected mapping between symbolparameters and traffic parameters. This mapping is depicted ingraphical form in Figure 2.

TABLE III. LIST OF TRAFFIC PARAMETER ASSIGNMENTS TO SYMBOLPARAMETERS

Domain Parameter Symbol ParameterTraffic Volume Connector Area1

Green Time Connector WidthFlow Rate Connector Length1Max Flow Connector Length2Saturation Connector Color

Link Orientation Connector OrientationCycle Time Circle Radius

Latitude Vertical PositionLongitude Horizontal Position

green time

flow

cycle time

latitude

longitude

max flow

saturation

Fig. 2. Visual mapping of traffic parameters using junction and connectorsymbols.

The geometric relationship between area, length and width(Connector Area1 = Connector Length1 × Connector Width)maps perfectly to the relationship between Traffic Volume,Flow Rate and Green Time (Traffic Volume = Flow Rate ×Green Time). Flow is a key attribute in any traffic analysisand thus needs to be represented with great precision. As aresult, Flow Rate and Max Flow are mapped to the lengthattributes of the connector. Green Time has been mapped tothe width of the connector. Saturation can then be mapped tothe color of the connector. The junction cycle time has beenmapped to the circle radius. The geospatial attributes of thejunction, as well as the directional attribute of each link arepreserved with this visual representation. A proof-of-conceptvisualization has been implemented using the D3 visualizationlibrary [3] to handle the mapping of data attributes to symbolproperties1. The resulting visualization is displayed in Figure3.

IV. EVALUATION AND DISCUSSION

To evaluate the Hub Map approach, we use principles ofPhysics of Notations (PoN) [4]. Table IV summarizes PoNprinciples for Hub Map as well as four other most usedapproaches for traffic data analysis. We discuss the key issues

1http://austrafficwatch.ict.swin.edu.au/traffic vis/

Fig. 3. Sample of resulting traffic visualization, showing traffic at 8:00AM.Traffic data c�VicRoads. Base map c�OpenStreetMap contributors.

identified from this evaluation below. A more comprehensivediscussion of these principles is available online2.

Semiotic Clarity: Semiotic Clarity requires a 1:1 mappingbetween semantic constructs and graphical symbols. Whilstnone of the traffic visualizations contain redundant, over-loaded, or excess graphical symbols, many suffer from symboldeficit, i.e. some domain parameters listed in Table II haveno corresponding symbol parameter. Color-Coded Maps andFlow Maps represent the three geospatial parameters (Latitude,Longitude, Link Orientation), leaving one symbol parameter(color or width respectively) for displaying a traffic parameter;representing four traffic parameters out of nine (4/9). ChartMaps encodes multiple link parameters in a chart symbolat the junctions. However, it is not possible to tell howparameters at the junction correspond to link orientations. ThusChart Maps represent 8/9 symbol parameters. In contrast toexisting visualizations, Hub Map defines symbols for bothjunctions and links; representing all traffic parameters withtheir corresponding visual parameters (9/9).

Perceptual Discriminability: For perceptual popout, i.e. theability to pre-attentively identify symbols that match somecriteria, the symbol parameter of interest must be mapped tounique values for one or more visual variables. If the values ofa visual variable are reused for different symbol parameters,then pre-attentive selection of that symbol parameter is notpossible. Accordingly, due to the small number of symbolsused in Color-Coded Maps and Flow Maps, they define uniquevisual variables for all 4/4 symbol parameters. If attemptingto represent all parameters using a Chart Map, there wouldnot be enough unique visual variables to represent each, thusonly the 3 geospatial parameters (Latitude, Longitude, LinkOrientation) are guaranteed to have unique visual variables(3/8). Hub Map defines unique visual variables for geospatialvariables as well as using color for saturation (4/9). How-ever, as congestion requires considering multiple parameterstogether (conjunctions), it will be necessary for the end userto perform serial scanning of the map.

Semantic Transparency: The mapping of the 3 geospatialdomain parameters to positional and orientation variables issemantically transparent as it preserves the spatial layout. Incontrast, the orientation of the links are not considered withChart Map’s pie-chart segments. As a result, only Latitude andLongitude have obvious meanings (2/8). Color in Color-coded

2http://austrafficwatch.ict.swin.edu.au/traffic vis/notes/pon detailed.html

TABLE IV. EVALUATION OF TRAFFIC VISUALIZATION SYSTEMS AGAINST PRINCIPLES OF PHYSICS OF NOTATIONS

SemioticClarity(fractionof domainparametersmappedto uniquesymbolparameters)

PerceptualDiscriminabil-ity (fractionof symbolparameterswith uniquevisualvariables)

SemanticTrans-parency(fractionof symbolparameterswith obviousmeanings)

ComplexityManage-ment (canit visualizethe entirenetwork?)

CognitiveIntegra-tion (canthe usernavigatewithoutgettinglost?)

VisualExpres-siveness(fractionof visualvariablesused)

Dual Coding(fractionof symbolparameterswith multipleunique visualvariables)

GraphicEconomy(total symbolparameters,less is betteras it reducescognitiveload)

CognitiveFit (is thenotationsuited forexperts?)

Color-Coded Map 4/9 4/4 4/4 Yes Yes 4/8 0/4 4 NoFlow Map 4/9 4/4 4/4 Yes Yes 4/8 0/4 4 NoChart Map 8/9 3/8 2/8 No Yes 4/8 0/8 8 Yes

Spreadsheets 9/9 0/9 0/9 No No 0/8 0/9 0 YesHub Map (ours) 9/9 4/9 3/9 No Yes 6/8 0/9 9 Yes

Maps is analogous to traffic lights therefore 4/4 parametershave obvious meanings (geospatial parameters and color).Similarly, size in Flow Maps represents traffic flow and there-fore 4/4 parameters are covered. The non-geosptatial trafficparameters in Hub Map are all semantically opaque. Whilstthis would cause difficulty for novice users, semanticallyopaque mappings are common in notations designed for expertusers. As a result, although Table IV represents 3/9 parametersfor Semantic Transparency, we expect that with continuoususage experts can distinguish their meanings clearly.

Complexity Management: Color-Coded Maps and Flow Mapsuse clustering and removal of minor roads to present the userwith a legible visualization even from country-scale zoomlevels. Hub Map is defined as a graph and therefore, graphsimplification methods should be used for its approximation.Chart Maps suffers from a similar problem. Spreadsheetshave no intrinsic way to split the complexity into manageableparts, although technically savvy users may be able to createoverviews of the data using database queries.

Cognitive Integration: Cognitive integration is usually de-scribed in abstract terms of navigating a cognitive map. Forgeographic maps, we interpret this as allowing the user to zoomand pan the map. When presenting on a base map, graphicalvisualization techniques usually support navigability. Spread-sheets are not navigable, even with sorting. Users will need tojump between rows and tables to navigate the data.

Visual Expressiveness: From eight visual variables defined byPoN, Hub Map uses 6/8 variables (Vertical Position, HorizontalPosition, Orientation, Shape, Size, Color). Hub Map usesneither brightness nor texture (2 free variables), as these are re-served for overlaying the visualization on a base map to allownavigation. Other visualizations only use 4/8 variables.

Dual Coding: Despite having plenty of free visual variablesthat could have been utilized, none of the existing trafficvisualizations dual code information. For example, color andsize could be used together in order to increase the dis-criminability of values. As Hub Map already makes use ofmost visual variables, it does not dual encode either. Thedependence of Color-Coded Map and Hub Map on color maypresent problems for color-blind users, as they do not definean alternative.

Graphic Economy: While having less symbols is desirablefrom the graphic economy perspective to reduce cognitive load,it also leads to symbol deficit. Color-Coded Maps and FlowMaps have a low graphic economy (4 symbols). This makesit easy for novice users to quickly learn their visual schema.

Due to the need to present more symbols, Hub Map has ahigh graphic economy (9 symbols). This is likely to exceedthe working memory of users new to the notation. However,expert users hold the visual schema in long term memory, andthus this should not be an issue once the visual schema hasbeen learned.

Cognitive Fit: Color-Coded Maps and Flow Maps are designedfor novices. Visualizations designed for novice users are notgenerally suited for expert users. In contrast, Hub Map isdesigned for expert users who are sufficiently trained inboth the domain and the interpretation of the visualizationitself.

Hub map visualizations are limited with regards to perceptualpopout and complexity management. However, they can beextended to provide network-wide overview by integration ofmultiple zoom levels and multi-resolution approaches. Sucha multi-resolution visualization can provide Hub Map as thezoomed-in view for junction vicinity analysis, and used color-coded clustering for network overview. This allows the trafficengineer to immediately recognize hotspots using the overviewand zoom in to Hub Map to identify bottlenecks and studynetwork effects.

V. CONCLUSION AND FUTURE WORK

We have introduced Hub Map for graph-based visualizationswith multi-attribute links. We have demonstrated applicabilityof Hub Map with geospatial graphs such as traffic data. HubMaps can be utilized in other graph based data visualizationsas well. For example, software profiling tools can use thisapproach to show software component interactions, or callsgraphs between components.

Our future work consists of empirical user studies to determinethe amount of cognitive load our visualization creates for endusers. Other avenues of further work includes identification ofadditional domains and graph based data sets that could benefitfrom our visualization approach.

ACKNOWLEDGEMENTS

The authors would like to thank Paul Bennett and Dr. IanEspada from the Australian Road Research Board for theirinputs into the design of the visualization, and Nathan Landand Chris Harper from VicRoads for help with traffic data.This work was partially supported by the Australian ResearchCouncil (ARC) Future Fellowships grant (FT120100723),and Swinburne University Early/Interrupted Research CareerScheme.

REFERENCES

[1] H. Chernoff, “The use of faces to represent points in k-dimensional spacegraphically,” Journal of the American Statistical Association, vol. 68, no.342, pp. 361–368, 1973.

[2] R. Troutbeck, M. Su, and J. Luk, National performance indicators fornetwork operations. Austroads Inc., 2007, no. AP-R305/07.

[3] M. Bostock, V. Ogievetsky, and J. Heer, “D3; data-driven documents,”IEEE Transactions on Visualization and Computer Graphics, vol. 17,no. 12, pp. 2301–2309, Dec 2011.

[4] D. Moody, “The physics of notations: Toward a scientific basis for con-structing visual notations in software engineering,” Software Engineering,IEEE Transactions on, vol. 35, no. 6, pp. 756–779, Nov 2009.