|
E-TOUCH:OVERLAY-BASED VIDEO SURVEILLANCE SYSTEM
FOR URBAN SECURITY
Xuesong Cao, Zhaoping Wang, Ruimin Hu
National Engineering Research Center for Multimedia Software
Wuhan University, Hubei Province,China
ABSTRACT
This paper presents a novel distributed video surveillance
system (E-touch) on Overlay Network. E-touch is
constructed by a set of overlay nodes that can provide
intelligence media service. The system allows service
composition of distributed stream processing applications
dynamically, to satisfy their end-to-end QoS demands.
Compare to traditional centralized-group management
scheme, our approach achieves better scalability and
resource utilization by efficiently allocating media service
resources. Large-scale experimental results demonstrate the
scalability, efficiency and performance of the E-touch
system.
Index Terms-Video Surveillance, Overlay network,
Resource Allocation, QoS
1. INTRODUCTION
In the past decade, the development of video
surveillance systems has captured the interests of both the
research and industrial worlds. The application of
surveillance sensors is becoming pervasive: They are
increasingly exploited in transport monitoring, urban and
building security, tourism , and bank protection [1,2].
A fully digital approach exploiting the use of
distributed processing resources within intelligent networks
has remained in the past few years. Distributed intelligent
Video surveillance systems are now becoming feasible on a
larger scale and also useful for simultaneous applications of
multiple functionalities. Research on the context of large
distributed surveillance systems is mainly from: video
compression techniques [3], network and protocol
techniques [4], distribution architectures and real-time
communications [5].
However, the approaches suffer from the problems of:
(1) scalability problem, where system can be short of
network resources (e.g., audio channels, network
bandwidth) for a large urban-surveillance application (2)
poor reliability due to media service point or IP links of
failure (3) degraded quality-of-service (QoS) when different
users need various level services for the power-limit.
Aiming at these problems, we would like to develop a
novel Video Surveillance System based on overlay network
architecture. In the system, distributed nodes are interconnected and comprised into an application-level
service overlay network for resource aggregation and failure
resilience. We approve that our special service model based
on overlay network outperform traditional architecture
specially under increasing workload conditions.
The layout of the paper is as follows. Section 2
describes the relative works of Overlay Network. Section 3
describes our surveillance system model based on Overlay
Network and proposes a QoS-aware routing scheme for
service composition. Section 4 presents the experimental
results to evaluate the scalability and efficiency of the
scheme. Finally, we present our conclusions in section 5.
2. RELATED WORK OF OVERLAY NETWORK
Currently, many overlay models are designed for
distributed Stream Application, such as multicasting,
content distribution networks and peer-to-peer file sharing.
And Overlay network is developing rapidly within past two
decades. MBone [6] deployed a large overlay multicast
system in internet based on IP tune technique. The main
focus is to manage and allocate overlay links and router
resources to different overlays. Yair Amir et al. [7]
developed a hop-by-hop reliability approach that
considerably reduces the latency and fitter of reliable
connections using an overlay network system. The OverQoS
[8] project proposed an overlay-based QoS architecture for
enhancing Internet QoS. The key building block of
OverQoS is the controlled-loss virtual link (CLUL)
abstraction. Resilient Overlay Network (RON) [9] is also
based on strategically placed nodes in the Internet domains.
It is proposed to quickly detect and recover from path
outages and degraded performance. The overlay networks
effectively use the Internet as a lower-level infrastructure to
provide higher-level services to end-users. The approaches
are opening new ways to Internet usability, mainly by
adding new services (e.g., built-in security) that are not
available or cannot be easily implemented in the current
Internet.
In this paper, we design an overlay service network
model for distributed surveillance applications, which have
urgent requirements of real-time and reliable security. In
following sections, we will describe this system model and
performance analysis results.
3. SYSTEM MODEL
Digital Video Surveillance Systems is a complex,
distributed multimedia application for urban security. In
[10], we have developed an object-based video surveillance
system implemented with ICE middleware [11]. Our system
is designed for large-scale urban surveillance application.
But actually, it is difficult to adapt to increasing QoS
requirements of surveillance business. Many modules must
take heavy load due to lots of task requests, such as Media
Relay Module and Transcoding Module, and not be able to
meet the end-to-end QoS requirements either. So we raise
our system model based on overlay network, and employ
effective service composition and resource allocation
scheme for media service.
3.1 Media Resource Overlay Network
E-Touch project is implemented for DongGuan Urban
Security System in China. The project builds a Video
Surveillance System involved more than 2000 cameras, and
stores about 600 TB video files at the data center. We
extract each sort of media service into one single overlay
node and detach all service nodes to form an overlay service
network. The middle network layer will provide all media
processing modules for session layer. Figure 1 describes a
three-layer framework of E-touch.
The system model is composed of three layers:
Underlying Networking Layer, Overlay Service Network
Layer and Session Layer.
1. Underlying Networking Layer (UNL)
UNL is a collection of all underlying physical nodes,
including routers, hosts and endpoints. UNL points to an
existing physical topology such as backbone network,
wireless channels, or LAN. In our project, the UNL is a
heterogeneous entity contacts wired/wireless access.
Base on the UNL, we select a set of special physical
nodes to construct a virtual overlay network. We endow
these special nodes with new defines as Overlay Service
Network (OSN).
2. Overlay Service Network Layer (OSNL)
OSNL provides common media service functionalities
such as Media Codec, Media Relay, Media Transcoding,
etc. It also provides multi-control strategies for system
balancing, such as topology discovery, overlay link
performance estimation, overlay routing, resource
allocation, etc. OSNL nodes are classified into three
categories: 1) Resource Access node (RAN), responsible for
interaction with up-layer and allocation of service resources.
2) Media Service node (MSN), responsible for dealing with
the media business. 3) Data node (DN), responsible for
storage of other nodes info and overlay topology. Each
service node periodically updates own status info stored in
the data nodes helping other nodes to construct overlay
topology.
When a new request arrives at Resource Access nodes,
using a QoS-aware Routing Protocol, the OSNL first forms
a logical application specific overlay topology as a subgraph
of the OSNL, connecting the appropriate Media
Service nodes. Then, it allocates the required network
resources for the session according to the application's
requirement.
3. Session Layer (SL)
All overlay applications aggregates in SL. In our paper,
overlay applications are mainly surveillance-oriented realtime
tasks, including video monitoring, Video Replay,
Video Retrieval, Alarm handling and so on. The
applications can be dynamically customized through OSNL.
So, not only surveillance-oriented application but also other
large-scale real time stream application, like Video
Conference and IPTV, could acquire customizing services
from OSNL. In other words, we have separate services
customizing from the basic topology construction. This
three-layer framework is much flexible for different
applications extending.
3.2 Service Path Composition
OSNL defines open interfaces for users to specify the
basic services involved in a new composite service. It is
desirable that the user does not need to know specific
service interfaces and media formats when specifying a
composite service. When tasks from Session Layer arrive at
RAN, OSNL will perform validity check on service
specifications and provides feedbacks to users. The DN
stores the whole topology of OSNL and profiles of basic
services. Based on DN, RAN is able to check if OSNL can
provide a composite service path for the task. Therfore,
OSNL will select appropriate MSNs to perform service
request by a de-Centralized optimal routing algorithm. The
algorithm can find QoS-satisfied paths according to network
bandwidth, CPU load, handoff delay and application's
requirement, which is described as following:
W is weight of selected path, , F(l) is Fair Index [12],
L(P)is average load of selected path, D(P)is total delay
of selected path, and p
d w is weight of delay for
corresponding task. Our fair resource allocation algorithm
distributes the load across multiple nodes, driven by
decisions of individual nodes, and based on global
information in DN. The approach allows dynamically
recomposition of distributed stream processing applications,
to satisfy their end-to-end QoS demands with high
probability.
Figure 2 gives some examples of Service Path
Composition.
Figure 2 (a) shows a service composition graph for a
system that offers video view services. S1->M1->A path
shows a relay model. Destination node A requires MPEG4,
128Kbps, CIF video stream from a Source Node, which is a
Digital Video Server (DVS). But the network bandwidth of
DVS is limited. So, the video stream is relayed to A by
Media Relay Service Nodes (M1). These relay nodes
multicast video stream to users in order to save bandwidth.
Similarly, S1->M1->T1->M2->B path depicts a transcoding
model. Destination node B also requires video stream from
S1. But B is more interested in a different video format as
AVS, 64Kbps. So, the video stream will be transformed to
adapt to B through a Media Transcoding Node (T1).
Because T1 is a shared resource in network, the output
stream is also relayed to B through M2.
Figure 2 (b) shows a service composition graph for
storage video demand. Destination node C requests to watch
video from Storage device. Storage Node retrieve
corresponding stream file in RAID.T2 transcodes the stream
into a special video format accepted by C.
3.3 Service Maintenance
During runtime, the surveillance service can experience
significant QoS violations or service outages due to the
failures of IP-layer network links or Service nodes. To
achieve robust surveillance application, E-touch provides
runtime failure detection and recovery mechanisms to
maintain the availability and QoS of all active conferencing
session. RAN of a stream session is responsible for
monitoring and maintaining the aliveness and QoS of media
service path for each surveillance application. Etouch relies
on the overlay data routing to recover the service outage or
performance failures of IP-layer network links.
3.4 System Application
Figure 3 and 4 describe the implementation of E-touch
project in DongGuan Urban. We deploy a large surveillance
platform in Municipal Center and 2000 surveillance points
across Town Center. From screen of our terminal, we can
observe scenes of different areas in the city. The system has
been officially opened in March, 2007.
4. PERFORMANCE ANALYSIS
4.1 Experimental Setup
We evaluate performance of E-Touch using a real
experimental environment. The project is applied to
DongGuan City Video Surveillance Network System in
China. We employ 2000 DVSs distributed across DongGuan
City for building test bed. All DVS are connected to the
central computer room through fiber. And an overlay
network, including 12 media relay nodes, 2 access nodes
and 1 data node, are designed to support the system
platform. The overlay network can provide load-balance
Media Relay Service for QoS guarantee. Figure 5 illustrates
the framework of test bed.
For comparison, we implemented traditional
Centralized-Group management scheme (CG), de-
Centralized random management scheme (Etouch-R). The
approaches are compared with de-Centralized optimal
algorithm of E-touch under increasing load. In traditional
Centralized-Grouped model, each media relay node is set up
to visit fixed number monitors. But in de-Centralized
random model, service nodes of path are randomly selected
instead of optimal routing algorithm.
We define the metric Load Fluctuation, Bandwidth
Fluctuation, and Average Delay for evaluating throughput
and performance of system under different number of
sessions. Load Fluctuation L' denotes CPU utilization
fluctuation of each working node,
defines  , Ln is CPU load of node i , and
L is average load of nodes. Bandwidth Fluctuation denotes consumed bandwidth fluctuation of each working node,defines  , i B is consumed bandwidth of
node i , and B is average load of nodes. Average Delay denotes delay of establishing session.
4.2 Results and Analysis
We have got a group of compare experiments in the DongGuan project arming at load, bandwidth, average delay and recovery rate.
Figure 6 and 7 illustrates throughput variation of all
media service nodes. We observe that E-Touch achieve
much higher throughput than the other schemes for efficient
resource allocation. Figure 8 shows average task delay time
comparison. The E-touch will cost more time to implement
a task than other schemes due to complexity of routing
algorithm. E-touch is better than other schemes under high
workload. The results demonstrate that the E-touch system
presents much better scaling property than the other two
approaches.
Figure 9 shows the recovery ratio, defined as the ratio
of the number of recoverable connections to the number of
failing connections. We varied the number of failing IP links
and the size of the overlay network. This graph expresses
the number of connections that can be recovered by routing
at the overlay layer when their direct physical connection
fails.
5. CONCLUSION
In this paper, we have presented a novel QoS-aware
Video Surveillance System called E-touch using a combined
application-level service overlay network. E-touch achieves
better QoS provisioning and resource utilization than a
traditional surveillance system that uses a centralizedgrouped
management scheme. E-touch can be easily
deployed, which does not require any IP-layer or application-layer multicast support. Actual experimental results demonstrate the efficiency of the Etouch system.
This work was supported by the National Natural
Science Foundation of China (60272097).
6. REFERENCES
[1] I. Pavlidis, V. Morellas, P. Tsiamyrtzis, and S. Harp, "Urban
surveillance systems: From the laboratory to the commercial
world," Proc. IEEE (Special Issue on Video Communications,
Processing, and Understanding.), vol. 89, pp. 1478-1497, Oct.
2001.
[2] V. Kettnaker and R. Zabih, "Bayesian multi-camera
surveillance," in Computer Vision and Pattern Recognition, Fort
Collins, CO, June 23-25, 1999, pp. 253-259.
[3] Liu, L.-C., Chien, J.-C., Chuang, H. Y-H., and Li, C.C.: 'A
frame-level FSBM motion estimation architecture with large
search range'. IEEE Conf. on Advanced Video and Signal based
Surveillance, Florida, 2003, pp. 327-334
[4] Saad, A., and Smith, D.: 'An IEEE 1394-firewire-based
embedded video system for surveillance applications'. IEEE Conf.
on Advanced Video and Signal based Surveillance, Florida, 2003.
[5] Huihui Huang, Weirong Chen, Qingquan Qian. Implementation
of Wide Area Communication in Distributed Remote Video
Monitoring System for Substations, Proceedings of the 2nd
International Workshop on Autonomous Decentralized System,
2002, pp.294-298.
[6] H. Eriksson, "MBone: The multicast backbone," Commun.
ACM, vol. 37, no. 8, pp. 54-60, Aug. 1994.
[7] Y Amir, "Reliable communication in overlay networks," in
2003 International Conference on Dependable Systems and
Networks (DSN'03), (San Francisco, California), June 2003
[8] A. Subramanian, I. Stoica, H. Balakrishnan "OverQoS: An
overlay based architecture for enhancing Internet QoS," in Proc.
1st Symposium on Networked Systems Design and
Implementation (NSDI), (San Francisco), ACM, March 2004.
[9] D. G. Anderson, H. Balakrishnan, M. F. Kaashoek, and R.
Morris, "The case for resilient overlay networks," in 18th
Symposium on Operating Systems Principles, (Lake Louise,
Alberta, Canada), ACM, October 2001.
[10] Xuesong Cao, Ling Jiang, Ruimin Hu. Design and
Implementation of Distributed Multimedia Surveillance System
based on object-oriented Middleware, Proceedings of SPIE - The
International Society for Optical Engineering, v 6418,
Geoinformatics 2006, 2006, p 641818.
[11] Michi Henning, Mark Spruiell. 2006. Distributed
Programming with Ice Reading. http://www.zeroc.com/
[12] R. K. Jain, D.-M. W. Chiu, and W. R. Have. A quantitive
measure of fairness and discrimination for resource allocation in
shared computer systems. Technical Report DEC-TR-301, Digital
Equipment Corporation, 1984. |
