dieter pfoser, lbs workshop1 issues in the management of moving point objects dieter pfoser nykredit...

28
Dieter Pfoser, LBS Workshop 1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark [email protected]

Post on 21-Dec-2015

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 1

Issues in the Management of Moving Point Objects

Dieter PfoserNykredit Center for Database Research

Aalborg University, Denmark

[email protected]

Page 2: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 2

Talk Outline

Motivation Moving Point Objects, Data, and Queries Query Processing

– Access Methods– Infrastructure

Uncertainty Future Work, Trajectory Mining

Page 3: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 3

Motivation

Spatiotemporal applications deal with spatial phenomena changing over time

Emerging applications that handle moving point objects

– fleet management– traffic management– mobile communication– environmental monitoring system

New type of data that stems from recording the movement in time trajectories

Page 4: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 4

Data

Sampling the position of a moving object at time points and interpolating in between samples

Page 5: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 5

Data (2)

Geometrical representation in 3D space (2D spatial + 1D temporal)

The resulting line segments comprise a polyline in 3D, the trajectory of the moving point object.

Page 6: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 6

Generate_Spatio_Temporal_Data (GSTD)

Fast moving objects (heading)

Page 7: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 7

Data Particularities

Typical: Dataset Objects Now: Dataset Trajectories Segments By adding time, we can derive further

information, e.g., speed of the object

Page 8: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 8

Queries

Coordinate-based Queries– point,– range, and– nearest-neighbor queries

Trajectory-based Queries– topological queries: enter, leave, cross, and bypass

and – navigational queries using derived information, e.g.,

speed and heading Combined Queries Join

Page 9: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 9

Topological Queries

Trajectories enter, leave, cross, stay within, or bypass a given spatiotemporal range

Signature: range {trajectories} {trajectories}

leaveenter cross bypass

stays within

Page 10: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 10

Navigational Queries

Considering derived information in query processing, e.g., speed (top and average), heading, traveled distance, covered area, etc.

Signature: range {trajectories} int|real|bool

Page 11: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 11

… where are we?

Motivation Moving Point Objects, Data, and Queries Access Methods Movement Types Uncertainty Future Work, Trajectory Mining

Page 12: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 12

Access Methods

We store segments in the index R-tree

– approximating bounding box, orientation of the segment, and trajectory id

1 4 3

2

STR-tree (SpatioTemporal R-tree)– spatial discrimination, i.e., preserve

spatial proximity of segments in a leaf node and

– trajectory preservation, i.e., segments belonging to the same trajectory (proximity with respect to trajectories)

Page 13: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 13

Access Methods (2)

TB-tree (Trajectory Bundle)– strict trajectory preservation,

i.e., one leaf node contains segments of only one trajectory

– neglecting spatial discrimination with respect to the two spatial dimensions

– given temporal discriminationbased on transaction time properties of the data

Page 14: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 14

Performance Studies

Datasets– synthetic datasets generated using GSTD

Experiments– Index size– Range Queries– Combined Queries

Parameter– number of moving objects– time horizon– object speed

Page 15: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 15

Index Size

R-tree STR-tree TB-tree

Index Size~ 95 KB

per object~ 57 KB

per object~ 51 KB

per object

Space Utilization

55%-60% ~100% ~100%

STR and TB-tree: smaller index through higher space utilization (“packing” of nodes)

TB-tree vs. STR-tree: trajectory id stored per leaf node gives smaller node size

Page 16: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 16

… where are we?

Motivation Moving Point Objects, Data, and Queries Access Methods Movement Types Uncertainty Future Work, Trajectory Mining

Page 17: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 17

Query Processing and Infrastructure Three different movement types

– unconstrained movement (vessels at sea)– constrained movement, infrastructure (cars,

pedestrians) – movement in networks (trains and, in some cases,

cars)

Page 18: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 18

Constrained vs. Unconstrained Movement Objects can be obstructed in their movement

Infrastructure, e.g., buildings, pedestrian zones (cars), roads (pedestrians)

Page 19: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 19

Uncertainty

Errors related to trajectory representation– Measurement error: every measuring technique has

an associated error, expressed by a positional probability function

– Sampling error: recording a continuous movement at time points introduces uncertainty in between samples

Exploiting information on the movement reduces error– maximum speed– map matching, dead reckonin

Page 20: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 20

Future Work

A detailed view– too many angles for future work for all of them to be

mentioned here...– e.g., more complex queries, refining query processing

algorithms, different approximation techniques, etc. A global perspective

– considering different kinds of data, e.g., networks– exploring uncertainty further (real data?)– applying the presented approach in a real-world

application context

Page 21: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 21

Putting the Stuff to Use...

Traffic– increased number of vehicles– demand for online information

Know more about current traffic condition Prediction about future traffic conditions and

prediction on the fly

Page 22: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 22

Ontologies

Trajectory

Ontology

Traffic Sensor

Vehicle Tracking

Imagery (aerial

photographs)

Observations

Page 23: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 23

Objectives (1)

Build a tool (Kinesis-Miner)

1. Extract knowledge about moving vehicles– busy routes in Aalborg at 15:00– number of vehicles now heading towards Nytorv – roads in which cars speed up during Saturday nights

2. Predict troublesome situations– traffic jams

3. Provide options– alternative routes for 15:00 through Aalborg– suggested routes through Hamburg over the weekend

Page 24: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 24

Spatiotemporal Data Mining

Spatiotemporal Data Mining: knowledge extraction from large spatiotemporal repositories in order to recognize behavioural trends and spatial patterns for prediction purposes – What is the connection between traffic jams in Aalborg

and the time of the day?– What are the locations of accidents on highways and

accidents on minor roads?

Page 25: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 25

Mining Methods

Classification– roads with many traffic violations dangerous

roads Characterization

– trajectory heading to town and high speed in city within an hour

Clustering– many vehicles close to each other possible traffic

jam Association

– number of vehicles in town and the time of the day

Page 26: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 26

Deliverables (1)

Mapping of all sorts of data to trajectory ontology Kinesis-Miner for knowledge extraction Kinesis language for support and interface Algorithms to support the mining methods

Page 27: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 27

Deliverables (2)

Graphical User Interface Kinesis Mining Language

trajectory characterizer

trajectory cluster finder

trajectory associator

trajectory generalizer

trajectory pattern finder

Spatial Database Server

spatial data

non-spatial data

trajectories

Page 28: Dieter Pfoser, LBS Workshop1 Issues in the Management of Moving Point Objects Dieter Pfoser Nykredit Center for Database Research Aalborg University, Denmark

Dieter Pfoser, LBS Workshop 28

Application Areas

Everything that moves!! Traffic (cars, ships, ..) Users with mobile phones

– How many users in Aalborg over Saturday night– Which cells will be overloaded