The Architecture of Analytics
Jun 1, 2008 12:00 PM, By Chris Johnston
At the point of its introduction in the last decade and even as recently as three years ago, video analytics existed mainly in government-related installations — nuclear power plants, select border points and military bases. At a price point of up to $15,000 or more per video channel, only the United States and other nations' governments possessed the funds necessary to support a video surveillance installation with this advanced technology.
Recently, as the cost of video surveillance technology in general has decreased, video analytics likewise has become more affordable and a realistic option for security directors in a broadening range of industries. Improvements in the accuracy of the technology — for identifying activities such as loitering, theft of an object or an object left behind — have also made it a more sought-after feature of video surveillance systems by security directors for casinos, airports, retail stores and other private sector businesses. And, while video analytics is not the silver bullet that can solve all security-related challenges, it may be an appropriate technology to assist the personnel monitoring a facility.
Once one has determined that video analytics is a valuable solution to implement in an organization — whether as a system upgrade or as part of a new video system infrastructure — the next step is to determine the architecture that best suits the requirements. As with video recording and storage, video analytics can be implemented using either a centralized approach - where video is streamed from all cameras to centrally-located servers for processing - or in a decentralized or embedded approach - where the analysis is performed by the IP cameras or encoders at the system's edge.
In many cases, the architecture decision is driven by a customer's network or budget limitations. However, there are a number of other factors, detailed below, that should also be considered during the evaluation process.
Available Processing Power — Video analytics algorithms have a large footprint, which could require up to 40 percent of the available calculation power of an encoder or camera. Some edge devices may not have the memory or processing power to run all of the needed analytics. In most cases, adjusting encoder settings should resolve the issue.
Risk Level — There is some level of risk involved with a centralized system, and there needs to be a back-up plan or some level of redundancy in case of server failure. If a server fails, the ability to analyze video from all cameras streaming to that server is lost. If this level of risk or the cost of redundancy is unacceptable, users should consider intelligence at the edge, as each camera or encoder processes video independently, resulting in no single point of failure.
Total Cost of Ownership — Before installing a system with centralized video analytics, users should involve their IT colleagues and understand its true cost. Total cost of ownership (TCO) incorporates factors — such as ongoing maintenance and operating requirements — that impact the cost of the system over time. Some estimate that for every dollar spent on computer hardware, nearly double the amount of time is spent on maintenance, such as administering operating system patches and anti-virus updates. That time translates into money spent, especially in the case of a managed services contract with an IT outsourcing company. IT professionals likely already understand that the total cost of ownership of server hardware and software can vary anywhere from 1x to 5x of the initial purchase cost per year. This equation means that over three years, for example, the TCO is between 3x and 15x the purchase price. Choosing video analytics at the edge eliminates the added expenses of server technology, resulting in a cost per channel that can drop dramatically to as little as $400, making it a more affordable option.
Bandwidth — With centralized video analytics, video from all cameras being analyzed must traverse the network to the servers running the analytics software. This design could cause network bottlenecks, if there is not enough available bandwidth. Integrators should work closely with IT colleagues to determine what the existing network can support. Using video analytics at the edge helps to reduce the overall amount of video traffic sent across the network for viewing and storage. One can choose to only transmit alarm video — video already deemed important based on established detection rules — thus diminishing the amount of bandwidth and storage resources required, but enabling the camera channels to be monitored effectively. This capability is important, especially in the case of monitoring outlying or remote areas of a site where network bandwidth is limited, such as in perimeter protection.
System Expansion — With centralized video analytics, the system can easily be programmed to send video from additional cameras to be analyzed, as long as there is sufficient processing power available on the existing server(s). However, additional servers may be required depending on the number of cameras being added. For video analytics at the edge, some manufacturers offer analytics as an embedded feature, where the IP camera or encoder already has the capability to perform the video analysis internally without the need for additional server hardware. The embedded analytics just need to be activated with a license. In this case, there will be no need to replace existing IP cameras and encoders in order to add analytics to areas where video was not previously processed. Manufacturers also offer programming tools to configure multiple cameras simultaneously, if they are analyzing scenes for the same set of rules or criteria. This feature makes adding analytics to IP cameras easier, since there is no need to manually replicate the programming.
Clearly, there are pros and cons to both centralized and decentralized architectures. Only by understanding the capability for each option to deliver the results, and understanding how each design will impact the acquisition and ongoing maintenance budgets, as well as network bandwidth, can users make the best buying decision for an organization.
About the Author
Chris Johnston is a product marketing manager for video surveillance equipment at Bosch Security Systems Inc., responsible for the company's Intelligent Video Analysis offerings. He can be reached at firstname.lastname@example.org or at 717-735-6372.
Want to use this article? Click here for options!
© 2014 Penton Media Inc.
Today's New Product
In support of the Privaris family of personal identity verification tokens for secure physical and IT access, an updated version of its plusID Manager Version 2.0 software extends the capabilities and convenience to administer and enroll biometric tokens. The software offers multi-client support, import and export functionality, more extensive reporting features and a key server for a more convenient method of securing tokens to the issuing organization.