TRAJECTORY SEMANTIC VISUALIZATION

Stanimir Bakshev, Laura Spinsanti

Database Laboratory, Ecole Polytecnhique Federal de Lausanne, Lausanne, Switzerland

Jose Antonio Fernandes de Macedo, Creto Vidal

Department of Computing, Federal University of Ceara, Ceará, Brazil

Marco Antonio Casanova

Department of Computing, Pontifical Catholic University of Rio de Janeiro, Rio de Janeiro, Brazil

Keywords: Trajectory Data, Semantic Visualization.

Abstract: Thanks to current GPS technologies, the capture of evolving positions of individual moving objects has

become technically and economically feasible. This opens new perspectives for a large number of

applications (from transportation and logistics to ecology and anthropology) built on the knowledge of

objects’ movements. The goal of this work is to propose a framework that supports querying and

visualization of trajectory data. Trajectory data and its semantic context are modeled by the means of an

application ontology, which allows the user to elaborate semantic queries. Results are rendered using an

automatic matching procedure that allows the user to change the actual visualization of the data.

1 INTRODUCTION

Thanks to current GPS technologies, the capture of

evolving positions of individual moving objects has

become technically and economically feasible. This

opens new perspectives for a large number of

applications (from transportation and logistics to

ecology and anthropology) built on the knowledge

of objects’ movements. From the users' viewpoint,

the concept of trajectory is rooted in the evolving

position of some object, travelling in geographic

space during a given time interval (Spaccapietra et

al., 2008). During motion, each object interacts with

different objects, both static, such as commercial

areas or traffic junctions, and dynamic, such as sport

events or specific weather conditions. This

additional information can help people solve some

common tasks such as pattern identification and

explanation of social phenomena.

GPS recorded data, after some post-processing,

provides many of the physical attributes of the

movement – latitude, longitude, time-stamp,

velocity, direction, distance. However, GPS

trajectories lack semantic information. Research

addressed the task of deriving semantic information

from GPS trajectories using the trajectories

themselves as well as further background knowledge

(Baglioni et al., 2009), (Baglioni, et al., 2008),

(Alvares et al., 2007). The automatic annotation of

GPS trajectories can further be advanced by methods

from the field of data mining and machine learning.

Yet, more complicated patterns and information are

hard to extract from raw GPS data. One needs to use

a conversion routine that is closely related to the

specific application of the data, and that also allows

for direct human manipulation and reasoning.

There are many GIS tools that facilitate the

visualization of spatio-temporal data. On the other

hand, recent research efforts (Baglioni et al., 2009),

(Spaccapietra et al., 2008) aimed at supporting

trajectory-based applications with new conceptual

models, where the semantics of movement is

explicitly expressed via application-aware trajectory

modeling, using ontology management applications

or semantic extensions in popular RDBMS. The

purpose of this work is to integrate these approaches

in order to provide a framework that enables the

visual exploration and analysis of semantically

annotated trajectory data. The results from this

326

Bakshev S., Spinsanti L., Antonio Fernandes de Macedo J., Vidal C. and Antonio Casanova M..

TRAJECTORY SEMANTIC VISUALIZATION.

DOI: 10.5220/0003565603260332

In Proceedings of the 13th International Conference on Enterprise Information Systems (ICEIS-2011), pages 326-332

ISBN: 978-989-8425-53-9

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

analysis can be later used to contribute to the

structured knowledge people have about moving

objects application domains.

There are several important questions in

trajectory data visualization and analysis. How to

characterize cluster of trajectories according to

domain knowledge? Why do trajectories indicate

frequent stops at specific places? Are there any

clusters or trends in this distribution? These

questions determine the need for a framework that

enables users to visualize trajectory data taking into

account domain knowledge, which can be either

spatial – the names of points of interest; temporal –

working days, rush hours, etc; spatio-temporal –a car

passing by polluted areas; or neither spatial nor time

related - such as which trajectories indicate tourist

behavior.

Usually, for each application domain, there is a

set of specific features of the trajectory. Even though

the semantic context may facilitate the

understanding of the data, it may vary among

different applications. Each application usually deals

with special concepts that may be structured and

described in various ways. Thus, an approach that

can provide correct visual cues while being able to

integrate with the different semantic data, regardless

of their structures and purposes, is needed.

Furthermore, by using highly interactive computer

interfaces, one may display visual cues that will

allow for further data analysis, knowledge extraction

and decision support, since the data may be

visualized from different perspectives.

Based on the above discussion, this work

proposes a framework that allows users interested in

trajectory data visualization and analysis to describe

the se-mantic context of the available trajectory data,

to identify and to specify interesting queries, and to

visualize and to analyze the results. The main

contributions of this work are the following:

• Trajectory Visualization – a computer

graphics library that allows for 3D visualization in

standard web browsers. It supports various visual

cues that are used to present application specific

concepts in spatio-temporal space;

• Trajectory Analysis – a set of tools and data

access routines that allow for trajectory data

analyses, such as spatial and temporal queries and

OLAP queries based on trajectory aggregations;

• Semantic Annotations – a set of ontological

modules that allow for the creation of platform

independent, application domain models that are

used as metadata description of the raw trajectory

data, allowing for more natural query formulation

and interpretation of the visualized results;

• Automated visual cue matching – a routine

for automatic matching between result datasets

(trajectories and their semantic context) and the

appropriate visual cues (markers, paths, areas, etc).

2 FRAMEWORK

2.1 Visualization Framework Overview

Trajectory datasets are usually huge with respect to

the number of records. In order to enable human

interpretation of trajectory data, it is necessary to

represent the data in a proper visual way (Andrienko

and Andrienko, 2008). The proposed framework

allows users to visualize structured trajectory data

through an interactive Web-based interface.

Trajectory visualization is explicitly done in the

context of trajectory semantics. The set of possible

semantic annotations is not predefined (explicitly

enumerated), as this can limit the use of the

framework in certain situations. For example, street

junctions are important for traffic monitoring

applications to analyze car flow through the crossing

of two streets, but they are not useful in applications

that analyze bird migration. So, only very generic

semantic annotations are predefined, and they can be

used as a basis for new annotations that are more

application-specific.

This work assumes that the semantic attributes of

a trajectory stem from its interaction with real-world

objects that fall into one of the following categories

– spatial, temporal, spatio-temporal and conceptual.

Spatial semantic attributes can be areas or points of

interest that have spatial interaction with the

trajectory. Temporal semantic attributes relate to the

time fraction during which the trajectory took place.

Spatio-temporal semantic attributes indicate the

evolving position of an object over time, or time-

stamped interaction with places of interest.

Conceptual semantic attributes are all attributes that

do not fall into the previous three categories. These

include any physical characteristic of a trajectory or

any externally annotated value.

The framework provides a set of visual cues and

tools that can be used in a variety of cases, without

further implementation efforts. The main plot area

used to visualize the data is a three-dimensional

structure called spatio-temporal cube (Gatalsky et

al., 2004). The spatio-temporal cube consists of

three axes that represent the x-y geographical

location in a given reference system and the time

axis. Using this approach one can represent all

spatial and temporal semantic attributes of a 2D

TRAJECTORY SEMANTIC VISUALIZATION

327

trajectory, allowing the user to selectively focus on

any of them via a rotation of the cube around the

three axes. Some visual cues are presented outside

the cube, such as components that are used to

facilitate selection of data aggregation levels.

A routine for automatic visual cue generation is

also available. Based on additional semantic

metadata, the framework infers the best visualization

technique. Furthermore, the current implementation

allows users to control this process and select which

attributes should be considered during visualization.

For example, the territorial division of a town will

be visualized, by default, as a set of semi-transparent

gray polygons over a geographic map. However

users are free to adjust this representation, simply by

indicating an attribute of those areas, such as

population or size. The representation will change to

a set of polygons whose colour will now represent

the colour-coded value of this attribute.

A general approach to handling large datasets

includes aggregation and summarization.

Aggregation means combining data items that are

close or similar. Summarization means deriving

characteristics of so-formed aggregates (i.e. groups

of data items) from the characteristics of their

members (Andrienko and Andrienko, 2008).

Trajectories sharing a common semantic attribute

are joined together to form a group, such as

grouping all trajectories that took place in a specific

city area. The average speed of those trajectories can

be visualized as polygonal area in the spatio-

temporal cube (since the city area is a spatial

attribute) and the summarized average speed value

can be then color-coded or presented with a chart

over this area.

Once the current set of data is visualized, the

user is able to freely browse and navigate through it.

Suppose that there are trajectories that are located

inside the neighbourhoods of a town. The supported

operations include filtering the data (show only

congested or pedestrian trajectories), drill-down in

the data (show only the trajectories in a given

neighbourhood), roll-up in the data (show data for

the entire town). Once identified using the different

tools to manipulate the visualization, interesting data

can be exported and saved for future references.

Even though the data may be aggregated prior to

visualization, this can still lead to significant

amounts of data that should be transferred over the

network. The data should be efficiently stored and

manipulated with minimum latency. A set of

methods is developed to form an interface for data

access. This allows for remote access to data via

standard protocols and provides optimization

techniques for quick data access and transfer.

2.2 Framework Components

In order to be able to visualize and analyze trajectory

data and its semantic annotations, there is a set of

components that communicate among themselves to

provide a fully functional system, with the

requirements stated in Section 2.1. In this section,

these subsystems are identified, following a bottom

up approach – from raw trajectory data to its visual

representation.

Trajectory data usually comes in the form of

time-stamped coordinate pairs that identify the

current location and the time of each measurement.

This data can be further processed for removing

noise, for identifying gaps or irrelevant data. The

result is a set of identifiable, structured trajectories

that can be efficiently stored and accessed in any

database system, such as Oracle.

However, to be useful, that data should be

integrated with additional semantic data, which is

specific for each application domain. For instance,

this may include not only geographical data, such as

territorial division and points of interest, but also

animal territories, climate zones, pollution areas, etc.

For this purpose, experts in a particular domain may

first identify its main concepts, then enumerate the

relations among those concepts and note their

possible interactions with moving objects. The goal

is to construct an application ontology that explicitly

defines the semantic context of a trajectory in which

the trajectory data should be integrated.

As a last step, the semantic data is integrated

with the trajectory data. There are several

approaches that allow for such integration (Baglioni

et al., 2009), (Baglioni, et al., 2008), (Alvares et al.,

2007). Once both the trajectory data and the

description of its possible semantic attributes are

available, the proposed framework can query and

visualize the data. Even though data can be directly

queried from the database with standard query

languages, there is a possibility to make this process

more tightly coupled with the application domain,

thus making the meaning of the results clearer.

With an application ontology describing the

trajectory data and enumerating the set of

application-specific concepts and the possible

relations among them, it will be convenient to

exploit this information to allow for easier, more

natural formulation of data queries. Assume that a

person wants to visualize all congested trajectories

that passed by a city’s traffic zone. Using the

ontology, a procedure can infer what stands behind

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

328

the concepts of traffic zone and congested trajectory.

This will allow it to get only the relevant data and

know the structure of this data, thus allowing the

framework to provide appropriate visual

representation. The main benefits stemming from

this approach are the following. First, data queries

can be formulated in terms of the application

domain, which means that the same trajectory data

can be used in various applications, where only the

semantic schema will vary. For example traffic

applications can introduce types of trajectories such

as congested, high speed or pedestrian, which pass

through traffic zones, while another application

dealing with studies of people behavior in city areas

can refer to shopping, entertainment and sport

trajectories that reach shops or cinemas. Second, the

visualization of the data can be done directly by the

framework, without the need of additional coding

efforts or explicit specification as to what visual cue

should be used for trajectories or their semantic

attributes. During data identification, the user should

select a list of trajectory semantic annotations that

are of interest to him. When the framework

visualizes the trajectory data resulting from this

query, it will show the trajectories with the relevant

values of their semantic attributes.

This work uses several techniques to display

large numbers of related visual cues even in

relatively small computer displays. These techniques

include three-dimensional visualization that allows

to integrate both spatial and temporal attributes of

trajectories within the spatio-temporal cube;

mapping layers that allow users to get instant, visual

geographic reference of the trajectory data; a multi-

layered perspective view that enables the perception

of many spatial layers at the same time, without

overlapping; freely changing the viewpoint so one

can focus on certain data and notice patterns that are

not easily identifiable from a fixed angle (see Figure

3).

A comparison with the data warehouse

abstraction is possible here. For example, in the

dimensional approach in data warehousing, data is

partitioned into either "facts", which is generally

numeric data, or "dimensions", which are the

reference information that gives context to the facts.

With respect to trajectory data, it can be broken up

into facts, such as movement speed or duration, and

into dimensions, such as points of interest, areas and

dates. Also, the retrieval of data from the data

warehouse tends to be very quick. So, as already

mentioned, by changing the viewpoint in the

visualization, one can focus on a specific dimension

– time (with side view) or points of interest (with top

view).

If the resulting dataset is too large, additional

grouping of the results can be performed using the

relationships among the concepts. For example,

suppose that a traffic analyst identified that

maintenance work (a concept) affected a set of

streets, and that he/she wants to visually analyze the

average speed and the number of trajectories that

passed through those streets during the period when

only part of the streets were closed. Obviously, this

aggregated number will provide some additional

semantic information about the trajectory, meaning –

did this maintenance work cause more congestion

than usual?, or did it make people use alternative

roads instead? Visualizing this type of aggregated

trajectory data and providing the necessary

analytical tools can further enrich the understanding

of the path generated by a moving entity in a given

context.

After a query has been sent for execution, the

results need to be visualized in a proper way. In

general, each query selects a subset of the

dimensions and measures in a predefined trajectory

cube. Thus, the structure of the results can be

inferred directly from this cube, and then can be

further reduced to some basic components needed to

provide proper visualization. Since spatio-temporal

cube is used as primary display to visualize the data

on a two-dimensional map with additional third

dimension for the time, the results need to be

decomposed into basic elements like time instants or

intervals, location points, and scalar values.

This decomposition is done via the ontological

module Visual Cues, represented on Figure 1. The

main map overlays are enumerated and linked to the

Geometry concept from the Movement ontology,

which includes trajectory conceptualization (more

details about Movement ontology can be found in

(Zhixian et al., 2008), via the has Shape property.

Using the fact that each Geometry instance can be

reduced to a set of Points, this provides the

necessary location points on the map. Apart from

that, each Point is time-stamped with Time instance.

As a result it is possible to construct a pattern for the

data that can be visualized by each map overlay. On

the other hand, the data that is returned from a query

to a given trajectory cube can also be reduced to sets

of Points which will form the pattern of the data.

Matching this pattern with the pattern of all overlays

generates the possible visualization technique that is

being applied in the specific case.

This paragraph shortly describes the

correspondences between data patterns and the

available overlays, and also what approach should

TRAJECTORY SEMANTIC VISUALIZATION

329

Figure

1: The visual cues ontology.

Figure

2: Structure for the Traffic Management Ontology.

be followed when creating new dimensions. The

Marker overlay visualizes data that represents

location on the map. It can be used to point events,

points of interest or geographical objects. For a

single Marker the data pattern comprises of a Point

instance and a scalar value that can be used to select

a distinct visual hint such as color or icon. The Chart

overlay can be used to present statistical information

that is referred to a single location of area. This can

be the weekly distribution of the number of

trajectories passing through a point of interest or any

other scalar value distribution over a discrete time

interval. The data pattern here requires a Point

instance and a set or multiset of scalar values that

are presented with bar, pie or line chart. The Arrow

overlay is intended to show the characteristics of a

movement between two distinct regions or points in

a geographical region of interest (ROI). These

characteristics can be the main directions of

movements within a city or indicate time intervals

for certain events, by orienting the arrow to be

collinear with the time axis. The pattern required for

this overlay consists of an OrientedLine and a scalar

value that can be mapped to different weights or the

colors of the arrow.

3 EXPERIMENTAL RESULTS

This section presents the visualization results

generated by a case study in the traffic management

domain. Information about trajectories is recorded

and stored in a relational database. First, an

ontological description of this application domain is

presented. This includes identifying some of the

main concepts in traffic management, and

enumerating their properties and relations. Then, it is

shown how to create two distinct data cubes that

help users analyze different aspects of the same

trajectory data. Finally, visual results for several

interesting queries are presented along with some

means that enable users to easily change the visual

representation of the data with simple manipulations

in the ontological description.

Geometry

Point

Polygon

Line

Simple Geometry

Is-a

Point Bag

Polygon Bag

Line Bag

Complex Geometry

Is-a

Oriented Line

Oriented Line Bag

Is-a

Overlay

Complex overlay Simple overlay

HeatmapNetwork

AreaPath ArrowChart

Marker

Grid

Is-a Is-a

Is-aIs-a

Is-a

Is-a Is-a

Is-a

hasShape

Polyline

Is-a

hasShape

ROI

Day

Trajectory

Move

ofTrajectory

passedBy

during

Pedestrian Trajectory

Car Trajectory

Congested Trajectory

Highspeed Trajectory

Is-a

Week Day

Weekend Day

Is-a

Street

TwoLaneStreet

House

Trafic area

Neighbourhood

Dense Neighbourhood

Is-a

locatedIn

Is-a

intersectsWith

Is-a

movedAlong

Year

Month

during

Region

Environment

Area

locatedIn

Summer Month Is-a

Winter Month

Is-a

locatedIn

Maintain activity

includesStreet

ObservationPoint

Is-a

OneLaneStreet

Is-a

Person

hasTrajectory

Town

Is-a

Municipality

Is-a

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

330

Figure 3: Semantic cube for car trajectories affected by maintenance activities.

Figure 4: Trajectories in city center.

Figure 2 shows the concepts of the traffic

management application. Each Move is assumed to

happen during a particular day and to be located in

one of the predefined areas resulting from the

division of the Town into a set of Neighbourhoods.

The concept of Traffic areas serves as an additional

division of the town into a set of areas that contain

Observation points monitoring the current traffic

conditions. The concept of Maintenance activities is

added to model actions that affect one or several

Streets, which can be One-lane or Two-lane streets.

Time is assumed to be composed of Weekdays or

Weekend days, while the months are either Summer

or Winter months.

In general, each Trajectory in this ontology is

considered to have the following semantic attributes:

it is located in one or several neighborhoods; it took

place during a specific day; and possibly passed by

some observation points located in various traffic

areas. Additionally, a trajectory was generated

during the movement of a specific person, either by

car or by walking, and the person’s movement took

place along streets that might have been affected by

ongoing maintenance activities.

One important point to highlight is the mapping

between trajectory data and domain ontology. It is

assumed that it is feasible to identify all structured

trajectories, with their respective sets of moves and

stops, and the physical characteristics of the move,

such as velocity, distance, location and time. This

Move

Month Year

CarTrajectory

TIME

Day

TRAJECTORY

Street

Town

Municipality

Neighbourhood

SPACE

MaintainActivity

TRAJECTORY SEMANTIC VISUALIZATION

331

data is loaded into the table Moves and is then

enhanced with semantic information about its

spatial, and temporal dimensions and about the ROI

it passed by. The result is a relational schema

populated with trajectory data on top of which it is

now possible not only to add a domain ontology that

describes the data and their relationships, but also to

describe interesting subsets of the data, which can be

automatically visualized and analyzed using the

tools the framework provides.

There are two semantically enriched subsets of

data that can be of interest in this scenario. The first

one helps users identify trajectory (either car or

pedestrian) characteristics in the context of their

temporal distribution among the different traffic

areas in the city. The second will help users follow

the impact that different maintenance activities in

the city had on the movement of cars. The generic

structure for this subset is identified in Figure 3.

Based on those needs, the user can define

trajectory cubes that relate trajectories to the

important semantic dimensions that were indentified

or, in other words, he/she can describe the semantic

annotations that are relevant to each case. Then the

user can arrange them in a multidimensional

structure where each cell contains some physical

characteristic of the trajectories, and each dimension

is related to trajectory interaction with different

traffic domain concepts.

Suppose we want to answer the following query,

“Give me all trajectories that passed by the city

center area”. Figure 4 presents the result of this

query, which is a set of trajectories, whose semantic

type is color-coded. Dark grey lines represent car

trajectories, and light grey lines refer to pedestrian

trajectories. The time period selected is April 3rd.

The user may inspect the relationships among these

trajectories, and the different observation points. The

user may also observe that the pedestrian trajectory

in this case has a longer duration.

4 CONCLUSIONS

This paper presented a framework that allows

semantic visualization of trajectories taking into

account users’ domain knowledge. Using the

expressive power of custom icons, visual styles,

charts and direction indicators, layers and 3D

visualization it is possible to provide meaningful

representations of trajectories and navigate through

different aggregation levels. The domain knowledge

is explicitly modelled with an ontology that

facilitates the understanding of the data and is used

internally for automatic detection of the appropriate

visual representation.

REFERENCES

Alvares L., Bogorny V., Kuijpers B., Macedo J. A. F., B.

Moelans, Vaisman A.: A model for enriching

trajectories with semantic geographical information.

Proc. 15th annual ACM International symposium on

Advances in geographic information systems (2007)

Andrienko N., Andrienko G.: Spatio-temporal aggregation

for visual analysis of movements Visual Analytics

Science and Technology - VAST'08 (2008)

Baglioni, M.; Macêdo, J. A. F.; Renso, C.; Trasarti, R.;

Wachowitz, M. Towards Semantic Interpretation of

Movement Behavior. In: 12th AGILE Int. Conf. on

Geographic Information Science, 2009, Hannover/

Germany. Advances in GIScience – Proc. 12th AGILE

Conference, 2009. p. 271-288.

Baglioni, M.; Macêdo, J. A. F.; Renso, C.; Wachowitz, M.

An Ontology-Based Approach for the Semantic

Modelling and Reasoning on Trajectories. In: 27th Int.

Conf. on Conceptual Modeling - ER2008 Workshops:

Semantic and Conceptual Issues in Geographic

Information Systems (SeCoGIS), 2008, Barcelona,

Espanha. Advances in Conceptual Modeling

Challenges and Opportunities. Heidelberg: Springer

Berlin, 2008. v. 5232. p. 344-353.

Andrienko N., Andrienko G.: EDA: Tasks, Tools,

Principles, Fraunhofer Inst.A IS (2004)

Campora S.: Trajectory Data Warehousing, Master thesis

in LDB, EPFL (2010)

Gatalsky P., Andrienko N., Andrienko G. Interactive

Analysis of Event Data Using Space-Time Cube.

Proc. 8th Int. Conf. on Information Visualization, pp.

145 – 152 (2004)

Güting R. H., Schneider M.: Moving Objects Databases

(2005)

Kuijpers B., Othman W.: Trajectory databases: Data

models, uncertainty and complete query languages In

Thomas Schwentick and Dan Suciu, editors, ICDT,

volume 4353 of Lecture Notes in Computer Science,

pages 224--238. Springer (2007)

Parent C., Spaccapietra S., Zimanyi E., Donini P., Plazent

C., Vangenot C.: Modeling Spatial Data in the MADS

Conceptual Model In Proceedings of the 8th

International Symposium on Spatial Data Handling,

SDH'98, p. 138-150 (1998)

Poggi A., Lembo D., Calvanese D., Giacomo G.,

Lenzerini M., Rosati R. Linking Data to Ontologies. J.

Data Semantics 10: 133-173 (2008)

Spaccapietra S., Parent C., Damiani M. L., Macedo J. A.

F., Porto F, Vangenot C: A conceptual view on

trajectories Data & Knowledge Engineering (DKE)

(2008)

Zhixian Y., Macedo J. A. F., Parent C., Spaccapietra S.:

Trajectory Ontologies and Queries Transactions in

GIS, vol. 12, num. s1, 2008, p. 75-91 (2008).

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

332