The approach to security and surveillance changed dramatically after the events of September 11, 2001.  Virtually overnight, the number and types of facilities and locations requiring or considering video surveillance grew exponentially.  In many cases the demand for warning systems that provide pre-incident actionable intelligence identifying suspicious incidents or behaviors was driven by legislation.   In other cases, it was driven by a new awareness of vulnerability.  Use of human guards proved costly, inefficient and in many cases unreliable.  Video surveillance became an obvious and potentially powerful alternative.   

Unfortunately, installing cameras alone does not result in increased security.  In the past a camera might deter a vandal or a thief by its mere presence.  In today’s world, presence of cameras would hardly deter a terrorist, as was evidenced recently in London, where the London Underground system has more than 7,000 cameras.  Cameras are only sensors, with no intelligence.  They simply collect vast amounts of video data, some relevant and most irrelevant.  It is the real-time interpretation of that data that delivers security.

Fortunately, behavior recognition technology or Video Content Analysis (VCA) offers a cost-effective solution for many of today’s needs.  VCA refers to artificial intelligence software that is able to detect, classify and track very specific behaviors and events involving people, vehicles, and other objects.  When rules are violated, Behavior Recognition systems will automatically alert a human operator in real time, providing a visual assessment of the situation.

Along with improvements in behavior recognition systems video-coding technology has also seen progress, enabling suppliers to provide remote video surveillance systems to their users.  We can now send video data via digital networks, instead of analog videos or closed networks.  System efficiency has also improved as multi-location surveillance videos can now be centrally controlled.

DATA: FROM USELESS TO USEFUL

The definition of ‘Critical infrastructure’ has expanded from a few high profile, high risk government properties to include vast numbers of public and private locations from transportation centers to pharmaceutical processing plants to all manner of utilities and mass production facilities.  Sites whose needs were limited to low-level security to protect against theft and vandalism four years ago now have been added to the critical infrastructure protection list.  These sites require proactive advance warning systems that can deter people and vehicles from threatening or destroying lives and assets critical to a corporation, industry or entire nation.

Today, corporate buildings, sports stadiums, bridges, airports, ports and harbors, and other critical infrastructure sites such as utilities, transportation terminals, telecommunications networks, petrochemical plants, government buildings, military installations and even national monuments have all been identified as potential high risk targets.  This new and expanding need requires the combination of technology solutions, in many cases combined with major changes to operations, policies and procedures.

The deployment of complex enterprise wide systems requires integration of Command, Control, Communications and Intelligence (C3I) systems to efficiently extract intelligence from the streams of data arriving from disparate sensors.  However, data itself is useless.  Data must be evaluated and filtered and intelligence must be extracted from it before it is useful.  In the case of video, it requires automated detection, classification, tracking and alerting of specific events and behaviors that may be precursors to a major incident.  Humans must be alerted to “real” threats while not being overwhelmed by “normal” activity.

BEHAVIOR RECOGNITION TECHNOLOGY

The timely convergence of a host of complementary and necessary technologies has enabled video analytics, often referred to as artificial intelligence, or simply as, “smart video” to progress from a laboratory research tool to a highly intelligent, rapidly deployable, easy to use, reliable tool for enhancing video surveillance.  Several years ago, our industry began to deploy video motion detection on a widespread basis.  Unfortunately, many scenes have motion of some kind, much of the time.  This is not useful.  What the industry needed was the introduction of computer vision algorithms to analyze each motion and determine if it is relevant and important to address specific security risks.  Video analytics based on computer vision algorithms began to be commercially available for exactly this type of behavior recognition in late 2002.

Concurrently, the rapid growth of complementary technologies has enabled the video surveillance field to make great strides in features and performance.  No longer is the user burdened with VCR tapes, poor video quality, the false alarms and irrelevant video images generated by motion detection, or the daunting task of locating incidents on hundreds of hours of video tapes.

Cameras are becoming smaller, higher performance and less expensive enabling users to triple and quadruple the numbers of cameras used for surveillance.  In addition, IP network cameras are rapidly gaining market share especially in new installations, permitting enterprise level installations to record and view cameras remotely from a desktop anywhere in the world.  New hardware video compression technology, embedded into professional DVRs has made great strides in increasing the quality of recorded video and at the same time reducing the storage volume of video data.

The rate of recorded video data is growing exponentially.  Fortunately, this is balanced by rapidly decreasing costs per gigabyte of massive RAID data storage devices, permitting vast amounts of video data to be stored increasingly cost effectively.  The great improvements in broadband communication technology such as gigabyte Ethernet, high-speed broadband wireless Local and Wide Area Networks (LAN/WAN) now permit any camera to be viewed from any desktop in the world.  Such trends lay the foundation to enable widespread availability of intelligent video-based behavior recognition systems.

That being said, all behavior recognition solutions are not created equal.  As with any technology, the potential buyer needs to arm himself with a solid working knowledge of the benefits, applications and limitations of this technology.

ActivEye, a spinout from Philips Research, has been developing VCA or “intelligent video” for nearly 10 years.  Active Alert®, the company’s flagship product can process many video channels using a Pentium 4 PC or a server, and in some cases can even be embedded in a DVR or a camera.  An example of an important core technology employed by the company is the use of a Digital Signal Processor (DSP) platform to drive its VCA software.  At ASIS 2005 in Orlando, Florida, U.S., the company has announced its new multi-core DSP platform which will permit up to 16 channels of the full Active Alert software suite on a single chip.  This may have a major impact on the surveillance industry by making ‘smart video’ easier to deploy, requiring far less hardware and offer much more cost-effective solutions. 

VIDEO SURVEILLANCE TECHNOLOGY

The core technology of remote video surveillance systems is the video-coding technology.  First, a low bit rate video communication system was established by modifying the standard MPEG1 and MPEG2 methods.  Then in 1999, MPEG4 was standardized, enabling video to be sent through general IP networks, such as ADSL and fiber networks.  Coding technology has been evolving since then, and in 2003, H.264 with better coding efficiency became the standard.

Oki Electric, a Japanese company providing network solutions, contributed to the standardization of MPEG4 and H.264 from an early stage.  In 2001, Oki launched its VisualCast, an MPEG4-based video surveillance system.  It is a remote video surveillance system with the video encoder, VBOX-S, as the core component.  It can record multiple channels simultaneously and effectively use IP networks by integrating with various sensors and business applications.  The system also includes a two-way voice communication function.  It distributes high-quality MPEG-4 video with low latency and is scaleable, covering video bit rates from 16kbit/s to 4Mbit/s.  It includes functions to change the data amount and image quality (resolution), according to the network bandwidth. 

Behavior Recognition Systems

The development of practical, easily deployable video content analysis software has faced a large number of challenges over the past decade.  Among these are: the ability to differentiate between humans, vehicles, animals, birds, trees moving in the wind, rain, snow, reflections from reflective surfaces such as marble, glass and mirrored walls, and changing lighting conditions.

‘Smart Video’ also imposes other challenges on users that need to be recognized and accommodated for optimal performance.  Proper camera placement is critical to operational success.

The camera’s physical placement and lens selection should be appropriate to observe the behaviors desired in a given scene.  In general, the rule for camera placement in intelligent video applications is, ”higher is better”.  This is important to avoid occlusion where multiple objects might ‘hide’ behind each other if the camera were positioned too low.  In special cases such as counting people, the camera is always positioned directly overhead and pointed straight down perpendicular to the floor.

Correct lens selection is also important to ensure that the images of objects of interest will be sized appropriately for detection.  For example, Active Alert is capable of reliably detecting objects as small as 18 pixels.  If detection is desired along a very long perimeter two or more cameras may be required each having progressively longer focal lengths to ensure all objects of interest will be detected and to minimize the blind zone beneath the cameras.  The system offers 35 events or behaviors as standard.  Many of these may be combined in a single camera view to provide users with multiple opportunities to detect progressively threatening situations.

Video Surveillance Systems

To prepare the best-fit video surveillance system, users need to purchase the surveillance camera, set the video bit rate, and decide how to store the video data.

First the camera: users need to determine the conditions in which the surveillance camera will operate.  Will they use it at night?  How large is the area that needs surveillance?  Will it be installed indoors or outdoors?  Does it need remote control from the surveillance center?  Grasping these conditions will help users select cameras that fit their needs.

 Next, users need to set the video bit rate.  This is determined based on the user’s network bandwidth and how much media they need to store.

Finally, users need to decide whether they want to store the video in a distributed model or a centralized model.  In the distributed model, data will be stored on the video-encoder-embedded-storage media that is installed in the surveillance locations; in the centralized model, data will be stored on the storage server in the surveillance center.  There are also two ways to store the data: the cyclic-type, in which data is stored on the media chronologically, or the alarm-linked type, in which data is stored when an alarm occurs in the video encoder.

BUILDING INTELLIGENCE INTO VIDEO SURVEILLANCE

In July 2005 Frost and Sullivan published an industry report on the status and growth of the video surveillance software market.  In it they predict that the market for surveillance enhancement software will experience a 25% compound annual growth to US$670 million by 2011.  Behavior recognition technology is having a major impact on the security industry from several different perspectives.

For end users, behavior recognition software provides vastly increased probability of detection of critical incidents.  Specific threats to operations, security or the safety of personnel can now be specified as ‘rule breaking’.  If an end user’s selected ‘rules’ are broken, the system will catch this automatically and provide actionable intelligence in the form of early warning.   In addition, this automated detection can reduce human monitoring costs by 50% or more.

From the perspective of the security system integrator and engineering design firm, the introduction of behavior recognition technology offers multiple opportunities for new revenue.  These include up-selling ‘intelligent video’ to virtually every existing CCTV customer, as well as including it in new bid specifications to provide specific solutions.

Manufacturers of cameras and DVRs will be able to differentiate themselves from competitors by building very sophisticated, multi-channel behavior recognition technology into their products.  A trend we predict for the future will be for OEMs and system designers to build as much intelligence into the ‘front-end’ of systems and thus greatly reduce the real-time bandwidth required to backhaul video to a command center.  Instead, minimal real-time bandwidth will deliver the alarm data and the channels of interest can be streamed back as required.

By combining Active Alert with VisualCast, behavior recognition can be conducted through a remote network.  This means that a complete remote video surveillance system can be established.  Moreover, such a system can improve operation efficiency because the Behavior Recognition System can be installed at the surveillance center itself, when there is limited space at the surveillance locations or when installation is outdoors.

Co-authored by:

Yoichi Yamada is with Oki Electric Industry Co., Ltd.                

Carolyn Ramsey is CEO of ActivEye, Inc.      

Richard Smolenski is VP of Business Development, ActivEye, Inc.

For more information, please send your e-mails to swm@infothe.com.

ⓒ2007 www.SecurityWorldMag.com. All rights reserved.