The public transport infrastructure in Paris, France, which includes the metro, RER, buses and tramways, uses more than 10,000 cameras on one contiguous system with streaming and storage at different data rates.


Although virtually every video surveillance system is equipped with a viewing monitor, the plain fact is that security personnel often do not react to real-time security breaches or events because hours of staring at the system’s monitors have dulled their cognitive senses.

Tests have been performed which show that a typical person can watch and react to changes on a relatively static video display for approximately 15 minutes, after which changes in the viewed scene are not recognized by the viewer.

Although some video surveillance images are relatively static, what about a camera watching a crowd entering a public building or sports arena? How can a human viewer distinguish important events, such as an abandoned briefcase in an airport, or a person going the wrong way on an escalator? Most video surveillance systems include multiple view feeds, so the problems of human recognition and response to security breaches are multiplied. The security industry’s traditional response to these human cognitive problems is to record all video feeds onto DVRs and/or NVRs, providing the ability to review security issues after they have occurred.

Although the reviewing capability is important, many clients would benefit from proactive video analytic systems that can pinpoint security problems as viewed by a surveillance camera and initiate immediate responses from security personnel.

Video analytics is a growing technology in the electronic security industry, minimizing the need to rely on human viewers to distinguish important or actionable events from video feeds.

A number of manufacturers are providing software and hardware sets that can be connected to existing or new video surveillance systems and provide a wide variety of human behavioral and object-event analysis.

The goal of these systems is to increase the efficiency of surveillance systems by providing real-time alerts and alarms when preset activities or conditions appear on a video feed. These immediate notifications can greatly improve the response times when critical events occur while reducing the manpower required to watch the monitors.

Basically, video analytics packages are intelligent video viewers that respond to pre-programmed events, watch security video feeds 24/7 and never take a break.

GENERAL FEATURES

Different vendors provide different types of event analysis; however, the general features provided by video analytics fall into four primary categories:

Live processing – Connected video feeds are processed “live,” providing almost immediate recognition of preset events or types of movement that are deemed as a security threat.

Forensic processing – Pre-recorded video can be screened for specific events, allowing security personnel to quickly review, for example, yesterday’s recorded video to find all the red automobiles that entered the parking lot.

Review indexing – Preset events can be indexed when recorded, so that instead of having to visually review multiple hours of video to find a specific instance of behavior, the system can be programmed to index events by their type. As an example, security personnel can quickly find all videos of people entering a building and going to an elevator without stopping at the entry security desk.

Bandwidth reduction – If using networked cameras, encoders or DVR/NVRs, video analytics can reduce the bandwidth used for video signal transmission over an enterprise’s LAN. These systems can be configured so that only video segments that include preprogrammed event types are stored and/or transmitted to the remote viewing computer or monitor. This reduction in bandwidth also can reduce the amount of drive space needed to record video for later viewing or analysis.

ANALYZE THIS

The various vendors in the video analytics field provide a wide range of programmable behavior and event alarms and alerts that can be applied to specific video feeds.

For example, Cernium Perceptrak offers such options as lurking, erratic, or converging people alerts, while also providing slow-moving, fast-moving, and multiple-vehicle detection.

Another vendor, iOmniscient, can detect objects left or removed in crowded or busy areas. It also counts people in a crowd and provides slip-and-fall and perimeter/intrusion detection.

ObjectVideo’s technology provides flow control, so that individuals’ movements through an area can be monitored and people who walk in the wrong direction for a specific distance will be flagged.

Many of these programmable analytics are similar across the spectrum of manufacturers, but each has its own unique blend of capabilities.

HOW IT’S DONE

Video analytics use computer processing power to analyze the differences between one video image and the next. Pixels that are different between the two images being compared are grouped into objects. This process is called segmentation. The objects and their movement are then compared to preset behavioral and motion parameters, and alarm sequences are initiated if certain criteria are met or exceeded.

It is important to understand that many factors, such as camera movement, lighting changes, compression artifacts, and RFI/EMI interference can cause pixel changes from one video image to the next.

Although video analytic software can filter some of these unwanted effects, a certain amount of inaccurate segmentation with corresponding false positive alarm activations is likely to occur.

Video streams from analog and IP cameras are analyzed in real time with the flexibility to expand easily to hundreds of cameras. The video analytic system illustrated here uses video processors.

ANGLES AND IMAGES

The accuracy of any video analytic system is directly related to the quality of the video images being processed. For the video analytic system to be reliable, the image streams must be of the highest quality.

Cameras must be securely mounted so that vibrations are minimized, and proper cabling and connectors must be used to reduce RFI/EMI noise.

Because compression artifacts can appear to the analytic software as segmentation changes, high rate compression settings should be avoided when using network cameras or encoders.

In terms of frames per second, Nik Gagvani, vice president of Cernium, states, “...the best possible quality should be used, trading off frame rate or size to conserve bandwidth.” This all means that only high-quality cameras and lenses will provide an adequate video stream for accurate video analytic results.

Areas of the video image that display constant motion may cause difficulties for the analytic software. In these circumstances, portions of the video image can be masked so that those areas are not included in the analytic process.

Camera placement and angle can make a critical difference in the performance of video analysis products. Gagvani says, “Camera angle is the single biggest factor determining the accuracy of recognition.” Ideally, cameras should be angled so that target objects move across the viewing field without significant change in their scale.

Mega-pixel security cameras, such as those available from IQEye, Covi, Panasonic, and others, would likely be the best possible devices to use for video analytic systems, provided that their compression format is supported by the selected video analytic vendor.

Security integrators considering the implementation of video analytic systems for their clients with existing systems should factor in the strong possibility that some cameras or lenses may need to be upgraded to higher quality devices, and some or many cameras may need to be re-aimed or repositioned to deliver usable image streams for the processing system.

HOOKING IT UP

Depending on the vendor, there are different methods of connecting new or existing camera feeds into a video analytic system. Some manufacturers provide DVR/NVR replacement devices.

Cernium has three such components, which the company terms video processors: the RVP-16 which provides 16 analog camera inputs; the RVP-16-IP which can connect and record 16 IP cameras; and the SVP-4, a four-analog camera input device. Each of these devices has a hard drive for local video storage.

A master console (MCON) or operator console (OPCON) computer is connected to the video processors and provides a user interface, video viewing and management.

Other vendors, such as iOmniscient, promote the capability of interfacing their systems to existing DVRs, which may reduce costs while retaining the value of pre-installed equipment.

Most vendors provide for the connection of both analog and IP camera/encoder video feeds. It is very important that security integrators carefully investigate the compression format capabilities of the video analytic product they are considering for a particular installation. The lack of industry standards for video compression may result in some camera types not functioning with a particular vendor’s equipment or analytic software.

Another important issue is the PC to be used for processing, control and viewing. Not every off-the-shelf computer has the necessary processing capability or video graphics needed to allow the smooth operation and viewing of video analytic programs.

VIDEO ANALYTIC SYSTEM COST

Most vendors provide their systems based on the number of channels of video that will flow through the analytic process. This can be confusing, because some vendors require that multiple channels be used for the same video feed if multiple behavior or event analysis is to be performed on one video stream. Other vendors’ products use one channel per video feed with multiple-event analysis.

As can be imagined, the cost of these systems may appear steep. The Cernium Perceptrak system costs approximately $1,500 per channel for a mid-sized system, with cost-per-channel dropping as the system becomes larger, according to Phil Robertson, vice president of Cernium. However, the analytical power of these systems can greatly reduce personnel costs, which may result in substantial savings for the end user.

Also, integrators should ask their client to consider the value of being immediately notified of behavioral or specific event activity as it is happening. Crimes can be stopped, people’s lives can be saved, and by reacting immediately when trouble occurs, end users can avoid costly litigation and bad publicity.

THE FUTURE

As the number of security video installations grows, the need for computer-based video analysis also will expand. Already vendors such as ObjectVideo can provide digital signal processing (DSP) chipsets that can be integrated into network cameras and encoders at the manufacturing level.

This will place the video analysis at the edge of the security network, further reducing the amount of video using network bandwidth. The camera or encoder will analyze the scene being viewed and transmit only the video that includes segments that match preprogrammed behavioral or event criteria.

ObjectVideo’s software can be put in multimegapixel cameras from Lumenera Corp., Ottawa, Canada. An alliance with Texas Instruments will enable Lumenera to load several types of video analysis software into an intelligent camera platform.

“We are able to throttle back the amount of frames or data being transported back to the server until an event occurs,” explains Greg Bell, Lumenera’s vice president of business development. Then a high-resolution image is sent to the surveillance operator.

In the foreseeable future, high-end video security will fully integrate video analytics, converting the always-on monitor into an event-driven display, only showing human responders what they need to see right now.

Sidebar: How to Select a Video Analytic System

Ask these Questions
  • How will this system connect to existing analog cameras, network cameras and DVRs?
  • What are the expansion capabilities of the system? How many additional analog and/or network video feeds can be connected, and for what cost?
  • What compression formats will work with the analytic devices and software?
  • Will the system work with mega-pixel security cameras?
  • What is the minimum resolution or video lines needed for proper analysis?
  • Are separate software modules needed to provide specific types of analysis at an additional cost?
  • Is the system based on Linux or Microsoft Windows?
  • Does the vendor provide on-site system setup and initial programming of event/behavior parameters?


Sidebar: More Information about Video Analytics

The following companies are some of those that offer video analytic systems and software. The latest information and contact information are featured on these Web sites.

ActiveEye
www.activeye.com

Cernium
www.cernium.com

iOmniscient
www.iomniscient.com

Lumenera Corp.(intelligent cameras)
www.lumenera.com

Nice Systems Inc.
www.nice.com/nicevision

ObjectVideo
www.objectvideo.com

Pyramid Vision
www.pyramidvision.com

VisioWave (owned by GE Security)
www.visiowave.com

Vistascape
www.vistascape.com

Links