Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Download
Standard view
Full view
of .
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
NWU-EECS-08-12

NWU-EECS-08-12

Ratings: (0)|Views: 9|Likes:

More info:

Categories:Types, Research
Published by: eecs.northwestern.edu on Jun 02, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

06/02/2010

pdf

text

original

 
 
Electrical Engineering and Computer Science Department
Technical ReportNWU-EECS-08-12December 10, 2008
 
Moving Convolutions and Continuous ProbabilisticNearest-Neighbor Queries for Uncertain Trajectories
 
Goce Trajcevski Roberto Tamassia Hui Ding
Dept. of EECS Dept. of CS Dept. of EECSNorthwestern University Brown University Northwestern Universitygoce@eecs.northwestern.edu rt@cs.brown.edu hdi117@eecs.northwestern.edu 
Peter Scheuermann Isabel Cruz
 Dept. of EECS Dept. of CSNorthwestern University University of Illinois at Chicagopeters@eecs.northwestern.edu ifc@cs.uic.edu 
Abstract
This report presents our solution to the problem of processing continuous Nearest Neighbor(NN) queries for moving objects trajectories when the exact position of a given object at aparticular time instant is not known, but is bounded by an uncertainty region. As has alreadybeen observed in the literature, the answers to continuous NN-queries in spatiotemporalsettings are time parameterized in the sense that the objects constituting the answer varyover time. Incorporating uncertainty in the model yields additional attributes that affect thesemantics of the answer to this type of queries. In this report, we firstly formalize the impactof uncertainty on the answers to the continuous probabilistic NN-queries (i.e., the semanticsof the answer to such queries), and we provide a compact structure for its representation.Then, we propose efficient algorithms for constructing that structure. For practical purposes,it is essential that the results can be incorporated on top of an existing Moving ObjectsDatabase, for which we identify syntactic constructs for several qualitative variants of continuous probabilistic NN-queries for uncertain trajectories, and address the problemefficient algorithms for their processing
Research supported by the NSF
: IIS–0324846, IIS–0713403, OCI–0724806 , ITR IIS-0326284,IIS-0513553, IIS-0812258, IIS- 0325144/003
 
 
Moving Convolutions and Continuous ProbabilisticNearest-Neighbor Queries for Uncertain Trajectories
GoceTrajcevski
DepartmentofEECSNorthwesternUniversity
goce@eecs.northwestern.eduRobertoTamassia
DepartmentofCSBrownUniversity
rt@cs.brown.eduHuiDing
DepartmentofEECSNorthwesternUniversity
hdi117@eecs.northwestern.eduPeterScheuermann
DepartmentofEECSNorthwesternUniversity
peters@eecs.northwestern.eduIsabelF.Cruz
DepartmentofCSUniversityofIllinoisatChicago
ifc@cs.uic.edu
ABSTRACT
This report presents our solution to the problem of process-ing continuous Nearest Neighbor (NN) queries for movingobjects trajectories when the exact position of a given objectat a particular time instant is not known, but is bounded byan uncertainty region. As has already been observed in theliterature, the answers to continuous NN-queries in spatio-temporal settings are time parameterized in the sense thatthe objects constituting the answer vary over time. Incorpo-rating uncertainty in the model yields additional attributesthat affect the semantics of the answer to this type of queries.In this report, we firstly formalize the impact of uncertaintyon the answers to the continuous probabilistic NN-queries(i.e., the semantics of the answer to such queries), and weprovide a compact structure for its representation. Then, wepropose efficient algorithms for constructing that structure.For practical purposes, it is essential that the results can beincorporated on top of an existing Moving Objects Database,for which we identify syntactic constructs for several qual-itative variants of continuous probabilistic NN-queries foruncertain trajectories, and address the problem efficient al-gorithms for their processing.
1. INTRODUCTION
Moving Objects Databases (MODs) [8] constitute a fun-
Research supported in part by NSF Awards IIS–0324846,IIS–0713403 and OCI–0724806.
Research supported in part by NSF Award IIS-0325144/003.
Research supported in part by NSF Awards ITR IIS-0326284, IIS-0513553, and IIS-0812258.
Permission to copy without fee all or part of this material is granted pro-vided that the copies are not made or distributed for direct commercial ad-vantage, the ACM copyright notice and the title of the publication and itsdate appear, and notice is given that copying is by permission of the ACM.To copy otherwise, or to republish, to post on servers or to redistribute tolists, requires a fee and/or special permissions from the publisher, ACM.
 EDBT 2009
, March 24–26, 2009, Saint Petersburg, Russia.Copyright 2009 ACM 978-1-60558-422-5/09/0003 ...$5.00
damental technology for a wide variety of applications thatmay require some type of Location Based Services (LBS) [27]for mobile entities. The main tasks associated with MODsare: (1) the efficient management of the location-in-time in-formation associated with mobile entities; (2) the efficientprocessing of various queries of interest, such as range ornearest neighbor (NN) queries. However, as has alreadybeen observed in the literature [5, 23], due to the imprecisionof positioning technologies (e.g., roadside sensors, GPS), itis not always possible to ascertain the exact location of aparticular moving object. Hence,
uncertainty 
must be takeninto account in the
data models
, in the
linguistic constructs
of the queries, and in the
processing algorithms
. The impactof various sources of imprecision in the context of probabilis-tic and uncertain data management has received consider-able attention recently (e.g., [32, 22]), including spatial andspatio-temporal settings (e.g., [5, 23, 36, 37]).Contrary to what happens in pure spatial settings [10, 25],the answer to a
continuous
NN-query in a spatio-temporalsetting is
time parameterized 
[35, 34] in the sense that theactual nearest neighbor of a given object need not be thesame throughout the time interval of interest. As an ex-ample, assume that we have a MOD which consists of aset of trajectories:
=
{
Tr
1
,Tr
2
,...,Tr
}
, and a query
Q nn(q)
:
“Retrieve the nearest neighbor of the moving ob- ject whose trajectory is
Tr
q
between 
t
b
and 
t
e
. The an-swer to the query is represented as a sequence
A nn(q)
:[(
Tr
i
1
,
[
t
b
,t
1
])
,
(
Tr
i
2
,
[
t
1
,t
2
])
,...,
(
Tr
im
,
[
t
m
1
,t
e
])], express-ing the fact that
Tr
i
1
is the nearest neighbor of 
Tr
q
initially and up to time
t
1
. However, the nearest neighborof 
Tr
q
during the time interval [
t
k
1
,t
k
]
[
t
b
,t
e
] (
k >
1) isthe trajectory
ik
∈ S 
.At the heart of the motivation for this work is the obser-vation that incorporating
uncertainty 
in the representationof the trajectories must be properly reflected in the syntaxof both NN-queries and of their respective answers. For ex-ample, consider a simple extension to
Q nn(q)
, in a mannerthat includes some uncertainty awareness,
UQ nn(q)
:
“Re-trieve all the objects that have a non-zero probability of being a nearest neighbor to the moving object 
Tr
q
, between 
t
b
and 
t
e
. In this case, in addition to a trajectory, e.g.,
Tr
i
1
be-ing the nearest neighbor of 
Tr
q
during [
t
b
,t
1
], it may well
 
be that some other objects may have a non-zero probabilityof being a nearest neighbor of 
Tr
q
in some sub-intervals of [
t
b
,t
1
].
Example 1.
Consider the scenario depicted in Figure 1. It illustrates 4 trajectories:
Tr
1
,
Tr
2
,
Tr
3
, and 
Tr
q
, shown as3D line segments; and possible bounds of the uncertaintiesof their locations, shown as sheared cylinders. Ignoring theuncertainty, the nearest neighbor of 
Tr
q
is
Tr
1
in 
[
t
b
,t
1
]
,and 
Tr
2
in 
[
t
1
,t
e
]
. However, if location uncertainty is taken into consideration, we see that not only 
Tr
1
, but also
Tr
3
has a non-zero probability of being the nearest neighbor to
Tr
q
at 
t
=
t
b
1
. Similarly, a
t
=
t
11
all three trajectorieshave non-zero probabilities.
Clearly, this needs to be considered continuously throughoutthe entire duration of [
t
b
,t
e
]. However, it is even more im-portant that we properly reflect it into all the sub-intervals,at a level of granularity dictated by the particular problemsetting.
T = t
e
XYT = t
b
T = t
1
TrqTr1Tr2Tr3T = t
b1
T = t
11
T
Figure 1: Continuous nearest neighbor for uncertaintrajectories.
We postulate that the structure of the answer,
UA nn(q)
,needs to be organized in a way that:
It identifies the trajectories
Tr
i
1
,
Tr
i
2
, ..., which havethe
highest probability 
of being the nearest neighbor to
Tr
q
,and the corresponding time intervals [
t
b
,t
1
], [
t
1
,t
2
], ....
Itidentifies
sub-intervals
within each [
t
k
1
,t
k
] duringwhicha particular trajectory would have been ranked as the onewith highest probability nearest neighbor of 
Tr
q
, had it notbeen for
Tr
ik
.
The structure is recursively refined for each sub-interval of time, until no lower granularity exists containing trajectorieswith non-zero probability of being a nearest neighbor to
Tr
q
.Each component of the answer may be augmented by anextra
descriptor 
of the properties of the probability val-ues of the trajectory associated with the particular timeinterval. For instance, such descriptors may contain: co-efficients/functions of an analytical expression (if possible),
min 
/
max 
values, plus a discrete sequence of values of theprobability at time instants within the given interval, etc.To represent the structure of 
UA nn(q)
, we propose an
interval tree
in which:
The root consists of the parameters of the query (i.e., querytrajectory
Tr
q
and the time interval [
t
b
,t
e
]).
The children of each internal node are the nodes that, withthe exclusion of their parents, have the highest probability of being the nearest neighbors of 
Tr
q
, within the time intervalbounded by the parent.The structure of each internal or leaf node consists of thefollowing attributes:1. time-interval [
t
i
,t
i
+1
] of relevance;2. unique
ID
, say,
Tr
i
, of the trajectory corresponding tothe answer during the time-interval [
t
i
,t
i
+1
];3.
descriptor 
D
i
of the properties of the probability of 
Tr
i
being the nearest neighbor to
Tr
q
within [
t
i
,t
i
+1
]; and4. pointers to the children-trajectories that have the next-highest probability of being the nearest neighbor within thedisjoint sub-intervals of [
t
i
,t
i
+1
].Clearly, this type of tree need not be balanced in termsof the height and number of children for each internal node,but we note that the leaf nodes correspond to the trajec-tories that have the smallest probability of being an uncer-tain nearest neighbor of 
Tr
q
within the corresponding time-intervals (i.e., no other trajectory has a smaller non-zeroprobability). We call this tree
IPAC-NN 
(Interval-basedProbabilistic Answer to a Continuous NN query) and weillustrate it in Figure 2. We note that if the root of thetree is removed, in effect we have a Directed Acyclic Graph(DAG), which represents the answer. Given this declarativedescription of the semantics of the answer to a continuousprobabilistic NN-query, the focus of the rest of this work ison the procedural counterpart: constructing the
IPAC-NN 
tree for a given query. We note that we do not address the is-sue of calculating the descriptors
D
i
of the individual nodes.Instead, we concentrate on
ranking 
[31]. In addition to for-malizing the semantics of the structure of the answers tocontinuous probabilistic NN-queries for uncertain trajecto-ries, our main contributions can be summarized as follows:
We identify a simple transformation of a view over theuncertain trajectories, which enables a construction of the
relative ranking 
of the probabilistic values for instantaneousuncertain NN-queries.
We demonstrate that our transformation is applicable toa large class of probability density functions (
pdf 
s) that de-scribe the uncertainty associated with the location.
We develop efficient algorithms to construct a geometricdual of a IPAC-NN tree.
We identify several syntactic variants for systematic incor-poration of uncertainty in the statement of the continuousNN-queries; for each variant we present an efficient algo-rithm for its processing, based on the dual of the IPAC-NNtree.
We present experimental observations demonstrating thebenefits of our approach.The rest of this paper is structured as follows. In Section2, we gather the necessary background. Section 3 presentsthe main contribution of our work in terms of the trans-formation of the uncertain trajectories and its implicationtowards algorithmic construction of the IPAC-NN tree, aswell as identifying the class of (instantaneous) location
pdf 
sfor which the transformation is applicable. In Section 4, wepresent the different variants of the continuous probabilis-tic NN-queries and their processing. Section 5 presents ourexperimental observations and Section 6 positions our workwith respect to the related literature. Finally, in Section 7,we give some concluding remarks and outline directions forfuture work.
2. PRELIMINARIES
In this section, we introduce the background necessary forthe development of our main results. First, we define themodel of uncertain trajectories used throughout this work.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->