PROFILE DATA DEPLOYING STRATEGY FOR EFFICIENT COOPERATIVE REASONING ON ONTOLOGY

International Journal of Distributed and Parallel Systems (IJDPS) Vol.2, No.6, November 2011
DOI : 10.5121/ijdps.2011.2602 13
PROFILE DATA DEPLOYING STRATEGY
FOR EFFICIENT COOPERATIVE REASONING
ON ONTOLOGY
Soomi Yang
Department of Information Engineering, The University of Suwon, Gyeonggido, Korea
smyang@suwon.ac.kr
ABSTRACT
This paper introduces the profile data management scheme such as data structuring, interchanging and
data deployment. An overwhelming amount of multimedia data is generated by surveillance devices such
as smart cameras. Profile data contained in them if properly structured and is integrated, can induce a
useful context information. This work builds a hierarchical profile data deploy structure and import
related multimedia data to annotate rich data arriving from multiple sensor streams, in this case smart
cameras. The annotation process provides an impetus to the improvement of knowledge over time.
Proactive deploying provides the main concepts and properties to model a hierarchical area data
structure which can span a university campus or an apartment or a city. We also define management
policies to compare their performance for the wide area surveillance specifically.
Profile Data Deploying, Reasoning based on Ontology, Wide Area Surveillance System, Context
Inference, Smart Camera Network
1. INTRODUCTION
The popularity of smart cameras providing many intelligent functions grows with the advance
of electronic technologies and Internet applications. It explores the convergence of the caching
and streaming technologies for multimedia data. For the surveillance of the large area, agents
built in networked RFID sensors, CCTVs and smart cameras should collaborate through
integration of profile data. Profile data includes audio, video, recognized feature data, ontology
data and others. Most such transfers would take place with the streaming data passing through
one or more of the agents caching the streaming data as it passes through. Effective caching
techniques will be critical for the successful deployment of streaming multimedia data over the
surveillance network. This should be obvious, because of the huge latencies involved, and the
requirements of real time play out. This paper describes a framework for the working of such a
distributed caching and streaming system. Distributed agents receiving heterogeneous data from
various sources have autonomy, collaborate with each other, and do ontology reasoning based
on distributed knowledge bases. A framework for the integration of profile data supplied by a
set of agents is designed. The agents are interconnected through a peer network. And non-leaf
administrative agents in a geographic area hierarchy or in an administration hierarchy build a
hierarchical structure according to their right hierarchy. This framework guarantees the
consistency and expressivity of the ontology used in the data integration process. In the process
of reasoning each agent may process the consolidated data for the distributed and autonomous
reasoning, which is scalable and efficient[1], helps security persons by giving appropriate
decision or prediction based on huge ontology data about situation it gathers.
The distribution of demands for profile data items is often skewed, and the surveillance devices
have different capabilities and data formats. These can lead to poor data communication and
dropped messages. In this paper, we propose a proactive caching scheme for efficiency and

14
interoperability. Implementation is also going on into our distributed surveillance network
environment.
The rest of the paper is organized as follows, Section 2 surveys related work of cooperative
inference schemes. Our proposed modelling framework is explained in Section 3. Section 4
describes the adaptive profile data management technique. In Section 5, simulation results are
presented and the performance is evaluated. We also show our implementation results. Finally,
Section 6 concludes with an outline of our future work.
2. RELATED WORK
There have been a lot of cooperative data deploying schemes proposed with analogical peers.
Numerous studies have been carried out on the caching and replication techniques in distributed
environments. Especially, [2] dealt with multimedia data which is the most interested data
format for the surveillance environment. In [3], a cooperative caching framework is introduced
and claimed to be effective to data availability. In [4], a replica allocation method and clustering
in distributed networks are introduced to improve data accessibility in a mobile communication
environment. Another cooperative data management scheme for similar peers is described in
[5], which combines the P2P communication technology with a conventional mobile system. [6]
tries to allocate fair buffer to regulate a varying traffic, and [7] suggests a proactive network
management solution. However, previous works were mostly based on the ground that the
distributed nodes in the network have the same characteristics, and the data transmitted is
standardized and intermittent. Our distributed surveillance environment consisted of various
data source devices, multimedia data is generated continuously and analysis should be done in
real time.
To ease merging of heterogeneous data, effort for the standardization for physical security is
done by ONVIF(Open Network Video Interface Forum)[8] and PSIA(Physical Security
Interoperability Alliance)[9]. They define, recommend, and promote standards for IP-based
security products. Our implemented system tried to meet the requirements of industry by
providing functions recommended by standard organizations. There are few of surveillance
systems adopting ontology-driven technologies. [10] introduces artificial intelligence techniques
only for the interpretation of objects. [11] uses ontology but does not build agents for web of
data. Furthermore existing wide area surveillance system are closed system and do not provide
surveillance information as a public web services which is suggested in the standard.
To address these challenges, we advocate an adaptive hierarchical profile data deploy
framework in which heterogeneous devices with different transmission ranges, latency and other
factors co-exist in the distributed surveillance network with a proactive data management
scheme to provide efficient public web services.
3. PROPOSED SURVEILLANCE FRAMEWORK
3.1. Hierarchical Infrastructure
The architecture consists of a hierarchy of agents, which contains a number of non-leaf node
administrative agents and leaf node agents with sensors, as shown in Figure 1. It consists of
national agents at the top, some regional agents at the middle level, and many local agents at the
second lowest level. At the lowest level, data source agents such as agents built in smart camera
make leaf nodes. Though data source agents can be directly connected to higher level agents,
most of data source agents are connected to the local agents. Meanwhile, the users are directly
connected to any level of agents. Each agent will be allowed to source the profile data into the
caches ahead of any client requests. Thus, the agent may not be on-line all the time but its
content can always be assumed available and hosted in one or more of the caching agents for the
integrated reasoning. This proactive caching also considers load balancing aspect. The
rectangles mean region servers and the ovals mean data source servers. The data source server

15
can be a single camera node or sensor node. Possibly it can be a back end server for several
cameras or sensor nodes. The lowest level server is called ‘level 1 server’. Level 1 region
servers receive data from data source servers in their domain and manage them. Rarely upper
level region servers can have data source server of their own.
Figure 1. Hierarchical infrastructure of region servers
The servers form a graph structure. They can communicate each other freely within access
control permission to perform their own intelligent distributed inference based on their own
ontology knowledge base.
3.2. System Framework
We are designing and deploying a system for the wide area surveillance system covering a
province area containing about 10 cities. The design of architecture for the surveillance of the
area consists of an integrated framework of networked RFID sensors, CCTVs, and smart
cameras. The surveillance device’s purpose is not only to take pictures, record videos and log
location data, but also to analyze a scene and report items and activities of interest to the user.
Also it is required to take actions based on prediction by ontology reasoning.
Various data, such as video, feature data including biometrics, event alarms, originate from
many kinds of input devices. The input devices can be deployed as fixed devices or on moving
robots for dangerous situations like floods, earthquake, etc. Although, agents are distributed by
geographic area hierarchy, they have their own knowledge base including ontology data about
situation, inference engine and communication protocol between input devices for queries.
Deploying profile data in distributed smart camera surveillance network has usually a
hierarchical architecture on top of the hierarchical administration organization. That is, each
hierarchical level of administrative agents is configured to operate in the same level, which is
usually placed at the access point connecting to the higher level of agents. These data, often
referred to as institutional, regional, and national data depending on the hierarchy of the
administrative agents operating them, is integrated with each other so as to infer and deliver
efficiently the requested result to the users.
With a profile data distribution, even small agents, with the help of higher level agents, can
achieve very high hit rate locally, thus reducing significantly the bandwidth requirements of the
network connection and the latency perceived by the users. To pursue this performance gain, the
deploying of profile data in administrative agents and proper distribution structure is needed for
the better hit rates achievable by the keeping profile data only in local data store with peer
network.
Despite this potential gain of the administrative agent distribution, there still remain many
technical and managerial problems to be solved before it could fulfill the gain in full. The
level n
. . .
level 2
level 1

16
administrative agent’s data store might be more heavily loaded as the number of data increases
faster and their update intervals go shorter and shorter. The situation becomes worse as more
data are replaced by multimedia data. This then cause the middle level agents to suffer a long
period of time before they contain most of the data of agents in the surveillance camera
network, degrading the whole system performance significantly.
This paper assumes a proactive, rather than reactive, deploying and attempts to explore an
optimal profile data management strategy and service policy for surveillance networks. By
proactive deploying, it means that the middle level agents cooperate with the master in
exchanging necessary information on the data size, the request rate and so on enabling the
master to fetch and broadcast the data before a lower level agents requests it. To achieve this
goal, the surveillance network is modeled as a multi-layered distribution network in the next
section. And the system performances like hit rate, disk space, bandwidth gain of the agents and
the leaf node source servers, bandwidth for the surveillance network, and latency experienced
by the users are derived.
4. ADAPTIVE PROFILE DATA MANAGEMENT
As shown in Figure 1, the profile data including multimedia data and feature data will be
distributed over the region servers and data source servers. The data source server will need to
carry out the indexing and retrieval of the information distributed across the servers in an
efficient manner. To aid the task, we could use a data structure containing only the information
related to the requested objects.
In the hierarchical profile data structure, agents cooperate with each other in such a way that the
request not satisfied at the lower level is forwarded to a higher level in the structure hierarchy
until it satisfied. If the request is not satisfied at any level of agents, it is forwarded to the
original data source server. As contrasted to this hierarchical request chaining, forwarding of
request at the same level operates quite a different scheme on top of the surveillance network
architecture. It forms another overlay network.
With the proactive deploying, which we assume in this paper, the master agent fetches data
from the data source server based on the exchanged information and broadcasts them to the
subscribing agents before any institutional agent requests them. This service has no hierarchy
among the agents and all agents maintain the same set of data for some designated suspect. With
this service, any missed request at the local agent is directly forwarded to the original data
source server.
In this section, we specify some features of the forwarding profile data and present analytical
models for evaluating the system performances.
Define D={1,2,…,|D|} as the whole set of profile data, and assume that the data in D are ranked
in the order of their suspiciousness or importance. From the well-known results that the
probability of the i-th data is collected is Zipf distributed[12], we have the probability function,
D
i
i
i
f ,...,
2
,
1
,
)
( =
= α
σ
(1)
where
∑
∈
−
=
D
i
i 1
)
/
1
( α
σ
and α is a constant close to 1.

17
Let k
n be the total number of connected agents and k
λ be the data request rate, respectively. It
is natural to assume that k
λ is proportional to k
n .
If agents do not differ from each other in their preferences for data, the request rate of agent k
for the data i, denoted by ki
λ , can be written as
α
σ
β
λ
λ
i
n
i
f k
k
ki ⋅
=
= )
(
(2)
Let F
D be the set of profile data required. Also, Let i
S be the size of the i-th data. Then, any
subscribing agent k needs to have an institutional storage with the required disk space k
δ ,
which amounts to,
∑
∈
=
F
D
i
i
k S
δ
(3)
Note that every agent operates the same set of profile data for some designated suspect, the
expression for k
δ being the same and independent k.
With the same preference distributions and the same set of profile data, all the subscribing
agents have the same hit rate as a distributed surveillance environment, multimedia data and
feature data are produced continuously. To maintain the freshness and effectiveness of the data,
we should profile data adaptively. We use the term ‘weight’ of data to describe its relative
importance as compared to the other data as proposed in [13]. The higher the weight, the lower
is the probability of the clip being replaced. We also use a policy based on the size of the
objects, in which the weight is proportional to the inverse of the size of the data for the network
bandwidth usage efficiency. Therefore the weight w is computed as following, where F is the
number of times the data is accessed, S is the size of the data and R is the time since the last
access for the data. The three exponents f , r and s are weighting factor.
( ) ( )
F
f s r
f s r
ki k i
f s r
k i i
i D
w F S R
n S
iα
λ δ µ
σ
β µ
∈
=
=
= ⋅ ⋅ ⋅
∑
(4)
where i
µ is the update interval. The value for f should be a positive number, meaning that
more frequently accessed data is more likely to be found. The value of s can be should be a
negative number for the efficient use of network bandwidth, such that more small data is more
likely to be stored. The value of r should be a negative number, meaning that more recent data is

18
more likely to be stored. If the recentness is more important than the frequency, the absolute
value of the exponent r should be greater than that of the exponent f.
5. PERFORMANCE EVALUATION
5.1. Simulation
For each given request rate β , we can inspect the effects to the data weight diversity. Figure 2
shows the expected data weight by data rank i and data update interval i
µ . With high ranked
data and small update interval, the data weight w grows high. Furthermore we can see the more
high data weight for the higher request rate β .
5 6 7 8 9 10 11 12 13 14 15
0
0.2
0.4
0.6
0.8
1
0
1
2
3
4
5
6
µ
i
w
β=1
β=1.5
β=2
Figure 2. The data weight diversity by request rate
Figure 3 shows the data miss rate diversity with different data weight. The difference between
the two graphs comes from the difference of the data weight that was seen in Fig. 2. To evaluate
the effect to the proactive data caching, we compare the data miss rate for the different weight
threshold for the data replacement using similar method in [14]. The data miss rate is calculated
relative to the case without proactive data caching. For the adaptive deploying technique, we
can raise the threshold value for the proactive caching. When we increase the threshold value,
the data miss rate Q decreases as shown in Figure 3.

19
0
5
10
15
0
0.5
1
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
i
µ
Q
threshold=1
threshold=3
Figure 3. The data miss rate diversity with different weight threshold
5.2. Implementation
We implemented our cooperative framework for integrating ontologies with an adaptive
deploying technique into our distributed surveillance system. Figure 4 and 5 show user
interfaces. Web-based user interface shown in Figure 4 lists biometric data recognized on the
right frame. On the left frame includes google map and marks for peculiar events. The event
lists pop up is the results of the integrated inference. When we click each of the event items, it
pops up related information. The subject tracking through the communication between the
independent data source servers is possible as shown in Fig. 5. They request and inquire profile
data for a combined inference through defined queries on the peer to peer computing
framework. In Figure 5, we show the grids used for the cooperative inference. Mobile user
interface is also implemented to see the real time or stored video and for smart camera control
Figure 4. Web interface with google map

20
Figure 5. Subject tracking between two data source servers
6. CONCLUSION
We describe the profile data deploying guidelines for better context inference based on
distributed ontology framework. Data source servers communicate each other freely within
access control permission to perform their own intelligent distributed inference based on their
own ontology knowledge base. Context inference including distributed multimedia data is
widely used in distributed surveillance environment. In such a distributed surveillance
environment, surveillance devices such as smart cameras may carry heterogeneous video data
with different transmission ranges, latency, and formats. These devices not only can get services
from a region server, but also they can form and generate a P2P network to provide services to
each other. For such a P2P network, an effective profile framework that can handle
heterogeneous devices is required. In this paper, we propose a flexible deploying scheme which
is adaptive to the actual device demands and that of its neighbors. Our scheme uses conformity
to update and share data in a cooperative way. Simulation studies are conducted to evaluate the
effectiveness of our flexible profile data deploying scheme. Implementation is also going on
into our distributed surveillance network environment. Based on mathematical derivations, the
proactive deploying of profile data maximizes the reasoning engine’s potential for making
decisions from operating the hierarchical structure distribution. Profile data deploying structure,
where data are deployed in the order of hierarchy and in the direction of suspect respectively,
are formulated and optimal service policy is pursued in view of which middle agents to include
in the system. Computational experiments are performed to investigate how the optimal
deploying scheme and service policy responds to system parameter changes. The system
parameters of our interest include data size, frequency, data update interval, the network
distances between agents, and some relevant factors. Finally, simulation and implementation
results are discussed in Section 5. More realistic data will be given in future work. Our scheme
shows the efficiency of profile data deploying resulted in better context inference.
ACKNOWLEDGEMENTS
This work is supported by the GRRC program of Gyeonggi province. [GGA0801-45700, Center
for U-city Security and Surveillance Technology]
REFERENCES
[1] A. Schlicht and H. Stuckenschmidt, (2008) “Towards Distributed Ontology Reasoning for the
Web,” WI-IAT08 pp. 536-539.

21
[2] J. Gomez-Romero, M. A. Patricio, J. Garcia, and J. M. Molina, (2011) “Communication in
distributed tracking system: an ontology-based approach to improve cooperation,” Expert
Systems (The Journal of Knowledge Engineering), Vol. 28, No. 4, pp. 288-305
[3] L.Yin, G. Cao, (2006) “Supporting Cooperative Caching in Ad Hoc Networks,” IEEE Tr. On
mobile computing, Vol. 5, No. 1.
[4] Sem Borst, Varun Gupta, Anwar Walid. (2009) “Self-organizing algorithms for cache
cooperation in content distribution networks,” Bell Labs Technical Journal,
[5] Kam Fung Yeung, and Yanyan Yang. (2009) “Mobile information retrieval in a hybrid peer-to-
peer environment, ” Mobility '09 Proceedings of the 6th International Conference on Mobile
Technology, Application & Systems
[6] K. Chitra and Dr. G. Padmavathi, (2011) “FAVQCHOKE: To Allocate Fair Buffer to a
Dynamically Varying Traffic in an IP Network,” International Journal of Distributed and Parallel
Systems (IJDPS) Vol.2, No.1, pp.73-81
[7] U. H. Rao, (2011) “Challenges of Implementing Network Management Solution,” International
Journal of Distributed and Parallel Systems (IJDPS) Vol.2, No.5, pp.67-76
[8] ONVIF (Open Network Video Interface Forum) http://www.onvif.org
[9] PSIA (Physical Security Interoperability Alliance Specification Package Q12009,
http://www.psiaalliance.org
[10] R. Martinez-Tomas, M. Rincon, M. Bachiller and J. Mira, On the correspondence between
objects and events for the diagnosis of situations in visual surveillance tasks, Pattern Recognition
Letters, vol. 29, Issue 8, pp. 1117-1135, (2008).
[11] Roberto Vezzani and Rita Cucchiara, (2010) “Video Surveillance On-line Repository (VISOR):
an integrated framework,” Multimedia Tools and Applications, Vol.50, No.2 pp.359-380
[12] T R Gopalakrishnan Nair, P Jayarekha, A Rank Based Replacement Policy for Multimedia
Server Cache Using Zipf-Like Law, Journal of Computing, Volume 2, Issue 3, (2010)
[13] A. Paknikar, M. Kankanhalli and K. Ramakrishnan, (2000) “Caching and Streaming Framework
for Multimedia,” Proc. of the 8th
ACM International Conference on Multimedia
[14] S. Yang, (2005) “An Efficient Access Control Model for Highly Distributed Computing
Environment,” Lecture Notes in Computer Science, Vol. 3741, pp.392-397
Authors
Soomi Yang received the B.S., M.S. and Ph.D. degrees in computer engineering
from Seoul National University of Seoul, Korea, in 1985, 1987 and 1997
respectively. From 1988 to 2000, she was a researcher at Korea Telecom
Research Center where she worked on telecommunication network, internet and
information security. From 2000 to 2001, she was a visiting scholar at UCLA,
USA. From 2002 to 2004, she was a faculty of the Suwon Science College.
Since 2004, she has been on the Faculty of the University of Suwon, Korea,
where she is a professor of computer science. Her research interests in
information security include access control, network security, and secure system
software.

PROFILE DATA DEPLOYING STRATEGY FOR EFFICIENT COOPERATIVE REASONING ON ONTOLOGY

More Related Content

Similar to PROFILE DATA DEPLOYING STRATEGY FOR EFFICIENT COOPERATIVE REASONING ON ONTOLOGY (20)

More from samueljackson3773 (20)

Recently uploaded (20)

PROFILE DATA DEPLOYING STRATEGY FOR EFFICIENT COOPERATIVE REASONING ON ONTOLOGY