Nowhere is the convergence of IT and physical security more evident than in video analytics, where software becomes a vital part of video surveillance. The most basic form of video analytics, motion detection, dates back more than a decade. In areas where there is normally little motion, such as a remote perimeter fence, motion detection can draw attention to anomalies rather than requiring someone to stare for hours at a static image. It can also conserve bandwidth. A surveillance camera in a normally quiet area can transmit low-resolution video most of the time, says Michael Martin, digital media architect at IBM Canada Ltd, but boost the bandwidth as soon as motion-detection software sees signs of activity. That’s just the beginning. Intelligent video can also watch for other patterns that might indicate security issues. For instance, if someone sets down an object in a public place, walks away and doesn’t return within a certain time, it might alert security personnel. The “bag left behind” scenario is one that video analytics tools often look for today, says Steve Hunt, founder of Chicago security consultancy 4A International, LLC. Loitering is another. If intelligent video systems can spot suspicious activity and draw it to security guards’ attention, you don’t need so many security guards staring at so many screens, on most of which nothing is happening. They can focus on potentially interesting events the software flags for them. And when something does happen, analytics can help them get more information faster. Video analytics software can recognize a face and then watch for that face to appear again, or search video archives for it. It can also search video for other objects, such as a red car. It can read licence numbers on passing vehicles. There are some stumbling blocks, though. One is video compression. To conserve bandwidth on IP networks, video signals are often compressed or the frame rate reduced, which makes video analytics more difficult. However, as storage gets cheaper, it has become increasingly practical to use only lossless compression that doesn’t hurt the quality of the image. The solution is to do most of the analytics at the edge of the network – where the camera itself sits – so high-resolution images don’t have to be transmitted to a central site. “You may not want to send all the video images all the way down the wire to a big computer somewhere,” Hunt says. But analytics at the edge of the network has its own problems. “When you have a camera and you start adding more and more at the edge there becomes a point where the return on investment is lost,” says Martin. Hunt says analytics works well if you have fairly specific requirements like spotting bags left behind, but those hoping to do many things will be disappointed because it requires too much processing power. He sees some promise in “machine learning” video analytics systems. While most analytics uses algorithms to identify specific patterns of interest, the machine learning approach learns to recognize normal activity and respond to anomalies, which requires less processing power.