Video Analytics: Separating Fact from Fiction

Intelligent video analytics, or computer vision, hold a great deal of promise, but the results so far have been disappointing. Alan Lipton, Chief Technology Officer for ObjectVideo, discussed reasonable expectations for video analytics at the Global Digital Surveillance Forum, part of SecuTech Expo 2007, in Taipei, Taiwan.

Video analytics seems to have become the must-have video surveillance technology of the 21st century. The promise of video analytics has generated much hype over the last few years, but what actually is it? Why would anyone want it? And how can would-be buyers separate the useful wheat from the over-hyped chaff? Video analytics has reached the point where it is no longer a sell, but a buy. The days of vendors scrambling to prove the value of the technology to prospects are over. Rather, customers are increasingly turning to analytics to help them solve real business problems. After seeing the technology early successes, it is important to have appropriate expectations when dealing with video analytics, and more importantly, ask the right questions.

What is video analytics?

Put simply, intelligent video analytics is software that extracts useful information from video imagery: It turns video into data. The underlying technology, called computer vision, dates back to artificial intelligence research from the 1960s. But it has only bore fruit after decades of research and funding by such Forwardlooking organizations as DARPA (U.S. Defense Advanced Research Projects Agency), which was also responsible for funding the development of the Internet. Computer vision techniques include sophisticated approaches to modeling the world as seen through the lens of a video camera. Environmental phenomena such as wind, rain, sunshine and shadows are detected, analyzed and rejected as immaterial. Legitimate objects such as people and vehicles are automatically identified and tracked through the scene.

Video analytics software uses computer vision techniques to detect and track objects, and then analyzes them to identify specific behaviors. For example, video analytics can detect when a person is breaching a perimeter, loitering around an ATM, or when a car is parked in an illegal space or for an unusual length of time. The ability to automate the process of monitoring video feeds may sound like science fiction, and considering the exaggerated claims made by some vendors, it is possible that unrealistic expectations about the current capabilities of the technology have surfaced. When confronted with an unlikely claim about video analytics, a good rule of thumb is to ask, "Would a person watching the video perform the same task?" If the answer is no, then it is unlikely that video analytics can do it either. Even if the answer is yes, it is still possible that the abilities of video analytics will be stretchedremember humans have been interpreting visual images for about five million years. Computers have only been at it for about 40!

Why buy it?

 The value proposition of video analytics is easyto catch bad guys. Surprisingly, it is more complicated than that. Early adopters have tried new technologies for their own sakes, but more conservative buyers require a measurable return on investment before they are willing to upgrade their systems. Sophisticated buyers are creating video analytics strategies that define the role analytics will play within the organization. A strategy starts with understanding how analytics will affect your company's missionwhat can it do for you? Obviously, efficient data gathering for security or business intelligence is a reason for adopting truly intelligent analytics. But there are other fundamental drivers. Video analytics can have a significant impact on the efficiency of personnel and infrastructure. If a company's staff are not glued to monitoring stations, they can be more mobile. This means they are more visible to act as a deterrent and can provide faster response should an incident occur. Furthermore, video-based investigations can be automated with intelligent video, so they take minutes rather than hours. Understanding which analytics features and functions are required is a key step to selecting appropriate analytics vendors and products.

A second part of a strategy involves integrating video analytics with CCTV infrastructure to increase the overall system value. Over recent years, businesses have begun to switch their CCTV infrastructure from legacy analog video to new IP-based digital video systems. This means that video data, which was once completely ignored on coaxial cable networks, can now be completely ignored over IP networks. It can also consume massive amounts of network bandwidth and digital storage in the process.

There is much press recently about video analytics at the edge, as there truly is great value in putting analytics in cameras, IP-video encoders and other front edge devices. A vast majority of video imagery is completely useless for security or business intelligence, but still consumes large amounts of network bandwidth and storage. Putting analytics at the edge provides a mechanism to filter the video at the source. When nothing important is happening, video can be transmitted and stored at a lower frame-rate, resolution and quality, or be deleted. Only video deemed important need be transmitted and preserved at high quality. This allows for intelligent savings of bandwidth and storage.

It is also valuable to put intelligence inside the network itselfsuch as within a network switch or router. One of the holy grails of IT is the data-aware network. If the network infrastructure understands the content of the data flowing through it, it can make sensible decisions about what information is important, how to route it and how to optimize quality of service.

Analytics embedded in a router gives the network the ability to peer inside a major data streamvideoand intelligently determine the importance of the content. Analytics also has a place at the back-end of a video storage device or management system. The same argument about bandwidth throttling applies to storage capacity and putting analytics at the back-end allows a video management system to save only important video information. In addition, many customers do not want the same analytics applications running on the same set of cameras all the time. Having centralized analytics allows customers to configure different application sets on different cameras at different times. And finally, putting analytics in the storage solution allows a device to tag stored video with analytics derived metadata that can make video investigations significantly more efficient.

Finally, look to the future. End users can select analytics vendors and products that fit their mission today, but they should be aware that as they use the technology, new uses will present themselves. They should make sure to select a partner that can grow with them as analytics becomes a more significant part of their operational paradigm. And for the same reason, they should select an analytics provider who can offer full set of truly intelligent capabilities, not just an a few algorithms sold a la carte.

What to look for and look out for?

There is much hype surrounding intelligent video and there are some things to look out for if those considering the technologyeither equipment manufacturers looking to incorporate analytics into product linesor as end customers looking to integrate analytics into their infrastructures and processes.

Look for strength in three separate areas: technology, product and company. End customers should look for technology that is proven in the real world. Make sure it is capable of providing the applications you require in environments relevant to your mission. It is easy to create compelling sales presentations, but a flashy PowerPoint slideshow does not always guarantee strong operational performance. End customers do not want to be someone's pilot project. Look for truly intelligent analytics products, not those based on dinosaur technology video motion detection (VMD).

The software must have well-tested features and functions that support a company's mission, and seamlessly integrate with its preferred CCTV infrastructure. Do not buy hardware products from analytics vendorsor home-grown analytics produced by a hardware vendor. Choose best-of-breed analytics functionality from an analytics market leader and a hardware platform from an equipment market leaderand make sure they are integrated together. And look for an analytics company that has a real track record and is capable of providing the necessary support.

Everyone loves video. It is cheap and intuitive, and therefore, fairly ubiquitous. Unfortunately, up until recently, video systems have had thousands of eyeballs, with absolutely no brains behind them. Video analytics gives the power of intelligence to CCTV systems by extracting useful data from video imagery. But it is up to the end user to drive how those brains are implemented. The time of the pushy vendor is overnow is the time for the wise buyer.

Share to:
Comments ( 0 )