By Mike Sherwood
Over the next year, the market will see products such as cameras, NVRs and DVRs, encoders, and video servers with embedded analytics that support an ever growing library of applications. Historically, video analytics have required the full flexibility and processing power of mainstream, rack-mounted servers or highly customized, dedicated analytics processing platforms. The new trend in the market is that powerful devices, packaged in a variety of edge-deployable form factors, will allow security consultants and designers greater flexibility in CCTV systems design, deliver more accurate analytics, and effectively manage bandwidth consumption of video data.
Embedded analytics devices are now targeting applications requiring the deployment of intelligent video capability in remote or harsh environments, in applications where limited wired communications bandwidth is available, or when it’s impractical to provide wired network access at all. Prior to this distributed architecture, customers were forced to deploy either analogue or digital cameras and provide high bandwidth connections to a central data center where analytics processing could be supported. Putting the analytics functionality at the edge of the network (near the camera) offers many advantages over a centralized approach. These advantages include:
Enables the Transition to IP Video Devices
One of the barriers of adoption to IP cameras is the massive network traffic they create. This is exacerbated by the increasing popularity of mega-pixel devices. Standard resolution cameras can easily generate 2 Mbytes/sec of data which saturates a 100 Mbytes/sec (fast Ethernet) network with only 30 cameras (assuming 60% network efficiency). Today, the most common method for managing bandwidth consumption is through the use of motion detection methods. The method used is based on sensing pixel changes from frame to frame. This approach has limited effectiveness since most scenes have constant or frequent movement of some kind, even if it is ‘non-interesting motion’ such as that from trees, puddles, flags, or normal human traffic. With video analytics installed at or near the camera, the sophisticated analytics algorithms can be used to identify true suspicious or noteworthy behaviors. Only then, the transmission of high frame rates and high resolution streams can be comfortably accommodated. Analytics insure that the desired activities are recorded, while not consuming bandwidth and disk space on non-interesting movements.
More Accurate Analytics
Edge-based analytics devices have the benefit of receiving and analyzing pre-compressed, high quality images increasing the probability and accuracy of detection. Video analytics accuracy, particularly for perimeter applications, is largely based on the number of pixels that make up the object that are available for processing. Many systems can detect movement of objects with as few as 8-15 pixels of data (e.g., 2x4 or 3x5). Locating analytics near the camera allows the analytic processor to operate on pre-compressed or higher resolution video, increasing the effective range of detection. Analytics operating on 4CIF (Common Intermediate Format) images will have twice the range of detection as 1CIF, increasing the effective range and requiring fewer cameras.
Video analytics edge devices are often also legitimate network devices. One such product, the Vidient Intelligent Video Router 2400, includes a 5 port network switch, Power-over-Ethernet for cameras and other digital devices, support for wireless protocols, and a firewall, all of which would be required separately without such a device. This greatly simplifies information and power distribution networks, decreasing complexity and cost.
Physical Security Integration
Along with remote cameras, many sites utilize a variety of physical security devices that are deployed at the site boundary or other remote locations. Many embedded analytics devices can also include contract relay input/output ports for integration to devices such as gate or door control systems, taut wire devices and other physical security mainstays. A single intelligent device at the edge can replace and/or support a number of smaller devices and simplify the network for control and communications between physical security devices.
Conserving Data Center Resources
Tightly configured embedded devices located at the edge of the network can also reduce precious data center resources, such as space, power and cooling support. With the rapid expansion of CCTV surveillance systems, large sites need alternatives to constructing massive data centers to house the computers to run the analytics and video encoding.
Although a strong case has been made for embedded encoding and video analytics, it has also become clear that ‘a single architecture does not fit all’. Significant reasons remain to continue with solutions based on industry standard rack-based servers. Each solution has its advantages. Server-based solutions may be more appropriate for the following situations.
Most current sites are designed based on a centralized video architecture, where all the video signals are already routed to a central facility. In these cases, changing to a distributed architecture, just to support video analytics, may not be cost effective. Also, many new sites are installing gigabit or multi-gigabit wired, fiber, or wireless networks to transport the video data to a central facility minimizing or completely eliminating the need for edge-base bandwidth control.
Limited Computing Capacity
Many of today’s edge devices and embedded processors are not yet powerful enough to support full functionality, full frame rate encoding and video analytics. For example, many of the digital signal processors in today’s cameras expend much of their computing capacity, performing the video encoding and compression, with little power remaining for analytics. As a result, the quality of the analytics is compromised, undermining the original purpose. For example, many intelligent cameras on the market today, can track only 1 or 2 objects within the field of view, not 10 or 20 that can be supported with stand-alone embedded encoders or with server-based solutions.
Proprietary Technology, Fewer Choices
Today, camera-based analytics are limited to a specific family of most manufacturers’ cameras. This often locks a customer or application into proprietary technology, providing few options, both for the camera and the analytics. Additionally, in today’s market, one can not expect or assume that embedded devices will lower cost than server-based solutions for a given application.
Until volumes increase, embedded products won’t benefit from the same economies of scale as industry standard servers and workstations. Over time, as volumes increase, prices will naturally decrease.
A MUST-HAVE TECHNOLOGY
Both approaches of embedded analytics devices and server-based soulutions have advantages. Most sites will have a mix of needs some which require embedded devices located at the network edge and others where an industry-standard computing model makes more sense. Rarely do site owners want to deploy multiple video analytics systems due to the cost of integration, training and deployment. Therefore, it often makes more sense to select a vendor that offers both solutions, so that the entire site can operate compatibly. Finally, regardless of deployment approach, the market continues to demand that any intelligent video solution provide high detection probability and high accuracy, with minimal false alarm rate.
As the industry moves toward the recognition of video analytics as a ‘must-have’ technology, volume and mass market applications will arise. These ‘killer applications’ could be:
Smart Motion Detection
Effective motion detection is a valuable function for both IP cameras and encoders, as it supports the conservation of both communications and storage bandwidth. Nearly every commercial camera or encoder has it today, but it has limited effectiveness due to the generality of the response; any type of motion. Video analytics has the potential to make motion detection truly useful, by limiting communication and storage to object and behaviors of interest. Human detection, often at the core of many ‘interesting’ behaviors is an important basis of comparison of analytics algorithms. Progress may come in steps, with less effective approaches requiring lower compute resources, coming to market first.
Another trend tied to the use of video analytics is the continuous extraction of events and motion from the video stream into a relational database. Such a database can be searched using simple queries or expressions and stored in a fraction of the space of the video itself.
Other applications include controlling lights based on room occupancy, automotive safety applications and integration with other technologies such as radar.
Embedded analytics devices have enabled a new range of possibilities for video analytics. For the first time, high powered analytics can be deployed in remote or harsh locations without wired network support. New processors are becoming available that deliver orders-of-magnitude (>10X) more performance than previous generations. As more power is packed into smaller devices, new applications will emerge, further driving growth. Ultimately, as in many high technology markets, rapid, robust growth will be dependent on finding the ‘killer app’ -- the ubiquitous application of intelligent video.
Mike Sherwood is Director of Channels for Vidient (www.vidient.com).
For more information, please send your e-mails to email@example.com.
ⓒ2007 www.SecurityWorldMag.com. All rights reserved.