SmaCCS: Smart Camera Cloud Services
Towards an Intelligent Cloud-based Surveillance System
Sven Tomforde
1
, Uwe J
¨
anen
1
, J
¨
org H
¨
ahner
1
and Martin Hoffmann
2
1
Organic Computing Group, University of Augsburg, Eichleitnerstr. 30, 86159 Augsburg, Germany
2
Volavis GmbH, Schuckenteichweg. 31, 33818 Leopoldsh
¨
ohe, Germany
Keywords:
Intelligent Surveillance, Cloud Computing, Automated Machine Learning, Smart Camera, Organic Comput-
ing.
Abstract:
Today, high performance and feature rich surveillance systems are very costly as they require an expensive
set of infrastructure components. As a consequence, such systems including, e.g., complex automatic video
content analysis, are restricted to large scale applications, such as airports or train stations. In smaller settings,
e.g. in shop surveillance, mostly low-cost display or record-only systems are in use. In this position paper we
propose to combine two well-known approaches in order to make Intelligent Video Surveillance applicable
and affordable in small to medium-scale scenarios. The proposal includes to combine the concept of Smart
Cameras, i.e. cameras equipped with local processing resources, with the ideas of Cloud Computing, i.e. the
on-demand provisioning of computing and storage services for complex calculations, and the management of
large amounts of data, i.e. video storage. The former allows for the cost effective pre-processing of video
data close to the sensor, while using the latter concept does not require large initial investments into expen-
sive infrastructure components such as powerful compute servers. The paper presents research issues of the
necessary system design, including precise system goal and system model aspects. Based on this, we discuss
several research issues required to be addressed for solving the overall goals.
1 INTRODUCTION
Within the last decade, a strongly increasing usage
of video-based surveillance has been observed driven
from both, the industry- and academia-side. The
application spectrum reaches from analyses of cus-
tomers’ flows in retail business to semi-automated
surveillance of safety-critical areas like airports and
railway stations. Most of the currently used systems
are proprietary and isolated applications serving just
one specific purpose. Typically, they consist of a pre-
defined set of cameras and a central point of oper-
ation, where all video streams are combined, stored,
and observed by human operators. This system model
results in non-scalable, hardware-intensive, and con-
sequently cost-inefficient solutions, see e.g. (Javed
and Shah, 2008).
One promising solution to alleviate these unde-
sired system properties is to combine Cloud Comput-
ing (Vaquero et al., 2008) concepts with more auton-
omy for the cameras: Smart Camera Cloud Services
(SmaCCS). The term Smart Camera (SC) refers to
a standard camera that is equipped with an on-board
computation unit which is able to fulfil video prepro-
cessing tasks (e.g. feature extraction and annotation of
video files with meta-data). In addition, a variable set
of SCs is able to self-organise a surveillance network.
Thereby, the Cloud-solution serves as interface to the
user and as basis for computation-intensive tasks like
video analysis (e.g. detection of persons and move-
ments, see (J
¨
anen et al., 2012)), pattern recognition,
and learning of conspicuous and abnormal behaviour.
A possibly large set of SCs will most probably result
in a huge amount of data that needs specific analy-
ses methods following the Big Data principle (Bryant
et al., 2008), which is most promisingly tackled using
a Cloud-based approach.
This paper outlines the general SmaCCS system
and names the upcoming research challenges to be ad-
dressed. The paper is structured as follows. Section 2
describes the state of the art regarding Smart Cam-
era and Intelligent Surveillance Systems, followed by
concepts for moving parts of the functionality into the
Cloud. Afterwards, Section 3 introduces the system
design of SmaCCS and highlights the most impor-
tant research issues. Finally, Section ?? summarises
the paper and gives an outlook to current and future
work.
288
Tomforde S., Jänen U., Hähner J. and Hoffmann M..
SmaCCS: Smart Camera Cloud Services - Towards an Intelligent Cloud-based Surveillance System.
DOI: 10.5220/0004590902880293
In Proceedings of the 10th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2013), pages 288-293
ISBN: 978-989-8565-70-9
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
2 RELATED WORK
Intelligent Surveillance Systems and Smart Cam-
eras. According to (Velastin and Remagnino, 2006),
intelligent surveillance systems can be classified by
their degree of autonomy and capability to sat-
isfy self-X properties like self-organisation and self-
configuration. The first class consists of analogues
CCTV (Closed Circuit Television) techniques for im-
age distribution and storage in a single control room.
In contrast, systems of the second class already com-
bine computer vision and CCTV which results in
semi-automatic visual surveillance. Finally, fully au-
tomated wide-area surveillance systems represent the
third class. These systems are characterised by the
distribution of intelligence among a possibly large set
of collaborative cameras.
The targeted SmaCCS observation system con-
sists of several pan-tilt-zoom (PTZ) capable Smart
Cameras (SC), see Fig. 1. A SC is an automated sys-
tem and combines an optical sensor with a computa-
tion unit for preprocessing and on-board analysis of
video data, see (Schneiderman, 1975). The output of
the optical sensor is preprocessed by the local com-
putation unit; hence, only limited image data has to
be transferred using communication and the network
traffic can be mostly reduced to event- and status-
messages.
Figure 1: Schematic overview of a SC network.
Surveillance systems have captured increasing in-
terest of both, the research and the industrial worlds,
in recent years (Javed and Shah, 2008). In the last
decade, surveillance systems of the third class, self-X
properties like self-organisation and the distribution
of hardware components have been the focus of in-
vestigations (Collins et al., 2001; Lipton et al., 2004;
Monari, 2012). The gap between research and real
robust industrial systems is wide. The initial costs
to build up a commercial full-automated surveillance
system are very high. Even large cities like Lon-
don run semi-automated systems, e.g. a static cam-
era is able to detect persons entering forbidden ar-
eas, (Cernium, 2013). Nonetheless, great parts of
the initially demanded research (Chu et al., 2004) has
been successfully performed for pure video surveil-
lance systems. The SmaCCS project aims at closing
the remaining gap by reducing the initial costs with a
pay-per-use concept and an outsourcing of hard- and
software components in the Cloud.
Cloud Computing – Privacy Issues and the Ap-
plication to Surveillance or SC Systems. Cloud
Computing (Vaquero et al., 2008) emerged as a pay-
per-use approach for accessing computing resources
(hardware and software) that are delivered as a service
over a network (i.e. the Internet). In this context, the
term “Cloud” describes the virtualisation and abstrac-
tion of the potentially complex infrastructure and the
remote access by customers. Today, a variety of com-
mercial Cloud-solutions are available focussing on
data centre (e.g. Amazon’s EC2), storage (e.g. Drop-
box), or Software (e.g. Microsoft’s Office 365). The
focus of the SmaCCS approach is to make use of the
data centre functionality as this is an economical (i.e.
reduce initial hardware cost) and highly scalable al-
ternative to centralised system architectures of current
surveillance systems.
Due to the potentially computation-extensive pro-
cessing tasks caused by continuous video data from
a large set of sources, a scalable Cloud-based com-
putation unit is necessary. Similar ideas have already
been presented in the literature. For instance, (Zhang
et al., 2009) describe a quality monitoring process in
industrial automation that is based on visual inspec-
tion (i.e. to detect surface defects). Here, the standard
approach is to make use of extensive image logging
and off-line human interpretation. In contrast, the au-
thors introduce their concept of measuring microm-
eter defects from a distance using a cloud of cheap
vision sensors.
Besides this first industrial-based video analysis
scenario, other researchers worked on a “Cloud-based
algorithmic framework which is scalable and adap-
tive to online smart city video sensing system” (Wen
et al., 2010). Here, spatio-temporal relationships are
derived from large-scale camera networks and simu-
lated using a Cloud-setup in order to build the best
possible topological structure. Hence, the focus is
more on design time decision support than on runtime
usage of the Cloud as access and analysis platform.
Finally, privacy is an important issue if sensible
data is transferred to the Cloud (i.e. in terms of le-
gal compliance and user trust). Therefore, (Pear-
son, 2009) state that this has to be covered at every
phase of designing the system. Corresponding aspects
concerning the privacy challenges will be considered
within the SmaCCS system.
SmaCCS:SmartCameraCloudServices-TowardsanIntelligentCloud-basedSurveillanceSystem
289
3 SYSTEM DESIGN
The following section defines the goal and describes
the system design of SmaCCS. Afterwards, the corre-
sponding research challenges are outlined.
3.1 System Goal
The goal of SmaCCS is to investigate a scalable
system allowing for an affordable semi- or full-
automated intelligent video surveillance system. In
this context, affordable means that the user in-
stalls one gateway and as many out-of-the-box SCs
as he needs for the specific tasks, let these SCs
automatically configure themselves based on the
bought license-model, and use a scalable access- and
processing-solution in the Cloud. The user will be
taxed for the Cloud-part on a pay-per-use approach.
The Cloud-based solution reduces the initial invest-
ment cost and increases the scalability of the system
and the possibility to automatise image-analysis and
recording management tasks. The system can consist
of arbitrary SCs, whereas the user is taxed on the par-
ticularly used number. The system aims especially
at small- and medium-sized surveillance tasks, like
the observation of business areas, large private house-
holds, or railway stations and airfields.
3.2 System Model
The system model consists of a user-side and a Cloud-
side. Details are illustrated by Fig. 2. The Cloud’s
server components are an application-, a Nagios-
(Nagios Enterprises, 2013), and a message-queueing
(MQ) server. The image-database as well as the
Cloud-analysis unit are the core components to per-
form exhausted image-analysis on Cloud-side. On
user-side, the SCs and a gateway to connect to the
Cloud have to be installed. The user accesses the ob-
servation system as well as the image-analysis results
via an application-interface (GUI).
3.2.1 User-Side
As mentioned before, a SC consists of an optical sen-
sor (e.g. a pan-tilt-zoom capable camera), a light-
weighted computation unit (SC-analysis), and a com-
munication interface. Each of the SCs is able to per-
form parts of the image analysis like movement de-
tection and detection of persons. In addition, the
SC can observe suspicious events (like persons ap-
proaching forbidden areas). Based on this informa-
tion, the SC decides autonomously about the frame-
rate and the image quality to be stored at server-side
(within the Cloud) and annotates the video data with
the derived meta-information. This reduction of ir-
relevant information is an important step for image-
data-compression which is essential for Cloud-based
storing of data.
3.2.2 Cloud-Side
The pre-processed image-data is streamed to the
image-database for storing and the Cloud-analysis
for post-processing. Based on the annotated data
stored in the Cloud’s database, the user can perform
computational-intensive analysis tasks or search for
specific situations (e.g. besides a standard search like
“all video data between time a and b”, he might look
for persons with yellow clothes). The application
server is the connection-interface to the user. It pro-
cesses the image analysis results and presents it to the
user. Single SCs communicate via XML messages
likewise to the Sensor Model Language (SML, 2013).
Therefore, the MQ-server is the transferring commu-
nication component, as well as the communication
between application server and SC network. As the
whole system consists of Cloud- and distributed com-
ponents, it is necessary to capture system states for
maintenance services. The Nagios-Server is an ade-
quate solution to handle component malfunctions.
3.3 System Components and Research
Challenges
The overall system as outlined in Fig. 2 consists of
a variable set of SCs, a Cloud-based server, and a
web-based user-access system. The following part of
this paper describes the corresponding system compo-
nents to be developed and the resulting research chal-
lenges related to these components.
1. Smart Camera System. SmaCCS builds upon
the results of the CamInSens system (D’Angelo et al.,
2012) and QTrajectories project (J
¨
anen et al., 2012).
Fig. 3 depicts the data stream within a single SC. The
image-data will be transferred from the optical sen-
sor (chip) to the internal-processing unit (PROC) of
the camera – this provides a video stream via its com-
munication interface (COMM). Typically, this opti-
cal sensor is a standard pan-tilt-zoom camera. In
addition, the sensor annotates the current timestamp
and current pan-tilt-zoom configuration to the JPEG-
header. For hard real-time constraints, this informa-
tion should be annotated as close to the data source
as possible. The SC-analysis unit is responsible for
the first image analysis (like person detection). The
enriched and annotated image data is accessible via a
COMM Interface by the Cloud services.
ICINCO2013-10thInternationalConferenceonInformaticsinControl,AutomationandRobotics
290
Figure 2: System Model of the SmaCCS system.
Figure 3: Architectural view of a single SC.
Typical observation jobs (like counting people
in an entrance area of a building) are usually fixed
assigned to dedicated SCs located at specific posi-
tions (e.g. above a door). More sophisticated surveil-
lance jobs like “look at position X”-commands initi-
ated by a user to get an image of a special location,
need a more sophisticated job scheduling method.
Therefore, each observation job is represented by a
software-agent. This agent is responsible for the ful-
filment of the user command. Such an observation
job can be initiated by the user or by the SC sys-
tem itself as a reaction to the observation result of
another surveillance job. The agent itself searches
self-organised for an adequate SC resource to fulfil
its job. Therefore, the agents trade camera resources
within the system. Another possibility is that more
important agents simply suppress other agents from a
camera resource.
In addition, the control of each SC can be im-
proved. Thereby, the Multi-level Observer/Controller
(MLOC) framework (Tomforde, 2012) will be ap-
plied to each SC in order to develop a self-organising
camera alignment solution. Furthermore, the MLOC
framework serves as classification system for the cur-
rently detected situation of the SC and therefore de-
cides about the data-quality to be submitted to the
Cloud, similar to the classification of network-traffic
situations in (Tomforde et al., 2011).
2. Cloud System. Besides the “normal” techni-
cal Cloud-side tasks as outlined before, the main re-
search challenges refer to a) an efficient data-storage
solution for video and image data and b) to intel-
ligent data-analysis method (e.g. based on machine
learning techniques). Storing video data efficiently
and accessing the contents based on a database con-
cept is an interesting task and has attracted researches
for years, see e.g. (Collins et al., 2001) for a good
overview. Here, the challenge is to provide an effi-
cient technique for searches and queries that scales
well with possibly huge data – i.e. as provided by re-
lational databases. Video and image data are not suit-
able for relational databases – hence, a data partition-
ing concept is needed. The second part the Cloud-
analysis has to find best-possible answers to user
queries. Thereby, the query should be more powerful
than standard solutions.
In addition, machine learning and data mining
techniques are needed to detect re-occurring complex
situations that might induce a security-threat to the
surveillance system. Therefore, video-data from sev-
eral SCs needs to be aggregated and suspicious move-
ments (in terms of trajectories, cf. the QTrajectories
project that serves as input (J
¨
anen et al., 2012)) have
to be detected. Additionally, the movements and sus-
picious events are categorised by a pre-trained alarm-
classification system. This alarm system informs the
user in case of events that have been classified as sus-
picious and reduces the false-alarm-rate significantly.
SmaCCS:SmartCameraCloudServices-TowardsanIntelligentCloud-basedSurveillanceSystem
291
3. User Interface. The system is designed to
be used by both professional security operators and
private users. We therefore aim at developing a user
interface that is easy to use but also comes with all
features needed to be used in safety critical environ-
ments. The notification of users in case of alarms is
an important task. We will extend our existing work
as described in (Hoffmann et al., 2010) by carrying
our work on the user interface design further. In (Yee
et al., 2012), practical solutions are given to adapt a
GUI to most effectively suite the needs of different
user groups under psychological aspects. Among oth-
ers, this approach may help to improve the useability
and acceptance of the system.
A first draft of the GUI is depicted in Fig. 4. A
web browser is used on the clients in order to display
the GUI whereas the business logic is running on the
server side. By offering a web application users may
access the system from any location with any device
available (notebook, smartphone etc.). Some func-
tionalities of the system are described in the following
on the basis of Fig. 4: A screen displaying video data
is taking most of the space to the right side. On the left
side, the menu is displayed that contains links in order
to browse a camera network and select cameras. In or-
der to browse historical video data, a calendar will be
included (in the middle of screen). Further menus will
give access to further screens. E.g. reports generated
by an algorithm for people counting may be accessed.
In order to administrate the system, various param-
eters can be adjusted. As mentioned before, a shop
system will be integrated in order to sell services to
the user. This services can be algorithm for analysing
video data or the enhancement of the system (more
cameras, storage upgrade etc.).
Figure 4: Possible GUI design for SmaCCS’ start page.
By carefully designing the user interface we aim
at reaching a broad range of users and usage scenar-
ios that require different levels of security, privacy an
the consideration of legal aspects as discussed in the
following section.
4. Security and Legal Issues. The consideration
of privacy is of utmost importance for social accep-
tance of surveillance technology. We aim at develop-
ing a trustworthy video surveillance systems by prop-
erly combining different aspects that current systems
do not manage. In particular, we propose the com-
bination of the following issues into the SMACCS
framework.
First of all, we will develop a properly defined in-
terface to detection algorithms mainly based on com-
puter vision techniques. Thereby we make sure, the
user knows exactly the data that the system gener-
ates and stores. For example, an algorithm for peo-
ple counting only delivers (and stores) the number of
persons moving in or out a certain area. In contrast
to this, an algorithm for people detection and recog-
nition might store data containing biometric informa-
tion. By offering a well designed interface, the op-
erators can customise the system to fit their specific
needs (and laws of their respective countries (Zick,
2007)). The content protection relies not only on us-
ing convenient cryptographic techniques, but also law
enforcement and user cooperation in order guarantee
the legal compliance of the system. For example, we
propose user pairing for accessing stored video data.
In order to access data that has been saved on the
system, two user have to agree on this request. Af-
ter both of them logged into the system, a specified
part (with time of beginning and end) of the stream
is made available for viewing. Some of the ideas
are based on a model for large-scale smartphone-
based sensor networks that has to cope with simi-
lar challenges (Kapadia et al., 2010). In addition,
the SmaCCS approach can re-use insights from the
predecessor project CamInSens (Hornung and Desoi,
2011).
4 CONCLUSIONS
This paper presented a concept for a scalable and
cost-efficient Intelligent Video Surveillance system
for small- and medium-sized surveillance tasks, like
the observation of business areas, large private house-
holds, or railway stations and airfields. The ap-
proach combines the advantages concept of Smart
Cameras, i.e. cameras equipped with local process-
ing resources, with the those of Cloud Computing, i.e.
the on-demand provisioning of computing and stor-
age services for complex calculations. Thereby, ex-
haustive image analysis and the management of large
amounts of data, i.e. for video storage, becomes pos-
sible. The Smart Camera approach allows for the cost
effective pre-processing of video data close to the sen-
sor, while using the Cloud-concept does not require
ICINCO2013-10thInternationalConferenceonInformaticsinControl,AutomationandRobotics
292
large initial investments into expensive infrastructure
components such as powerful compute servers.
Current and future work will try to find answers
on the research challenges outlined in this paper. Es-
pecially, the research issues of the necessary sys-
tem design, including precise system goal and sys-
tem model aspects, the machine learning and efficient
video/image data storage and querying are focused.
REFERENCES
Bryant, R. E., Katz, R. H., and Lazowska, E. D. (2008).
Big-data computing: Creating revolutionary break-
throughs in commerce, science, and society. Com-
puting Research Initiatives for the 21st Century. Com-
puting Research Association.
Cernium (2013). Perceptrac System webpage. Online,
http://www.cernium.com.
Chu, M., Reich, J., and Zhao, F. (2004). Distributed atten-
tion in large scale video sensor networks. IEE Seminar
Digests, 2004:61–65.
Collins, R. T., Lipton, A. J., Fujiyoshi, H., and Kanade,
T. (2001). Algorithms for cooperative multisensor
surveillance. In Surveillance, Proceedings of the
IEEE.
D’Angelo, D., Grenz, C., Kuntzsch, C., and Bogen, M.
(2012). CamInSens An Intelligent in-situ Security
System for Public Spaces. In Proceedings of the 2012
International Conference on Security & Management
(SAM), 2012, 16.-19. Jul, Las Vegas, USA. CSREA
Press.
Hoffmann, M., J
¨
anen, U., Fares, A., and H
¨
ahner, J. (2010).
Amidivin: basic algorithms for alarm management in
distributed vision networks. In Proceedings of the
Fourth ACM/IEEE International Conference on Dis-
tributed Smart Cameras, ICDSC ’10, pages 150–157,
New York, NY, USA. ACM.
Hornung, G. and Desoi, M. (2011). Smart Cameras und
automatische Verhaltensanalyse. Kommunikation und
Recht, pages 153–158.
J
¨
anen, U., Feuerhake, U., Klinger, T., Muhle, D., H
¨
ahner,
J., Sester, M., and Heipke, C. (2012). QTrajectories:
Improving the Quality of Object Tracking Using Self-
Organizing Camera Networks. In ISPRS Annals of
the Photogrammetry, Remote Sensing and Spatial In-
formation Sciences, Volume I-4, pages 269 – 274.
Javed, O. and Shah, M. (2008). Automated Multi-Camera
Surveillance: Algorithms and Practice. Springer Pub-
lishing Company, Incorporated, 1 edition.
Kapadia, A., Myers, S., Wang, X., and Fox, G. C. (2010).
Secure cloud computing with brokered trusted sensor
networks. In Collaborative Technologies and Systems
(CTS), 2010 International Symposium on, pages 581–
592.
Lipton, A. J., Clark, J. I., Brewer, P., Venetianer, P. L., and
Chosak, A. J. (2004). Objectvideo forensics: activity-
based video indexing and retrieval for physical secu-
rity applications. In Intelligent Distributed Surveil-
liance Systems, IEEE, pages 56–60.
Monari, E. (2012). Dynamische Sensorselektion zur
auftragsbasierten Objektverfolgung in Kameranetzw-
erken. KIT Scientific Publishing. ISBN: ISBN 978-
3866447295.
Nagios Enterprises (2013). Nagios webpage. Online,
http://www.nagios.org/.
Pearson, S. (2009). Taking account of privacy when design-
ing cloud computing services. In Software Engineer-
ing Challenges of Cloud Computing, 2009. CLOUD
’09. ICSE Workshop on, pages 44–52.
Schneiderman, R. (1975). Smart cameras clicking with
electronic functions. Electronics, pages 74 – 81.
SML (2013). Sensor Model Language webpage. Online,
http://www.opengeospatial.org.
Tomforde, S. (2012). Runtime adaptation of tech-
nical systems: An architectural framework for
self-configuration and self-improvement at runtime.
S
¨
udwestdeutscher Verlag f
¨
ur Hochschulschriften.
ISBN: 978-3838131337.
Tomforde, S., Hurling, B., and H
¨
ahner, J. (2011). Dis-
tributed Network Protocol Parameter Adaptation in
Mobile Ad-Hoc Networks. In Informatics in Control,
Automation and Robotics, volume 89 of LNEE, pages
91 – 104. Springer.
Vaquero, L. M., Rodero-Merino, L., Caceres, J., and Lind-
ner, M. (2008). A break in the clouds: towards a
cloud definition. SIGCOMM Comput. Commun. Rev.,
39(1):50–55.
Velastin, S. and Remagnino, P. (2006). Intelligent Dis-
tributed Video Surveillance Systems. Professional Ap-
plications of Computing. Institution of Engineering
and Technology. ISBN: 978-0863415043.
Wen, Y., Yang, X., and Xu, Y. (2010). Cloud-computing-
based framework for multi-camera topology inference
in smart city sensing system. In Proceedings of the
2010 ACM multimedia workshop on Mobile cloud me-
dia computing, MCMC ’10, pages 65–70, New York,
NY, USA. ACM.
Yee, C. K., Ling, C. S., Yee, W. S., and Zainon, W.
(2012). Gui design based on cognitive psychology:
Theoretical, empirical and practical approaches. In
Computing Technology and Information Management
(ICCM), 2012 8th International Conference on, vol-
ume 2, pages 836–841.
Zhang, L., Malki, S., and Spaanenburg, L. (2009). Intel-
ligent camera cloud computing. In Circuits and Sys-
tems, 2009. ISCAS 2009. IEEE International Sympo-
sium on, pages 1209–1212.
Zick, T. (2007). Clouds, Cameras, and Computers:
The First Amendment and Networked Public Places.
Florida Law Review, 69(St. John’s Legal Studies Re-
search Paper No. 06-0062):1 – 66. Available at SSRN:
http://ssrn.com/abstract=956160.
SmaCCS:SmartCameraCloudServices-TowardsanIntelligentCloud-basedSurveillanceSystem
293