Machine Vision – Video Surveillance
The Digital Watchman
Video monitoring systems that automatically detect unauthorized persons and follow their every step are edging out of labs and moving into the mainstream. On the way are systems that differentiate between humans, animals and vehicles, identify unusual behavior, and track suspects even in crowded environments.
Today’s most advanced video surveillance centers (above) provide an overview of activity in major installations, such as the Munich airport (below), which has approximately 1,600 cameras
1:34 a.m.—a suspicious person stealthily approaches an airport area under cover of darkness, climbs over a high fence, sneaks across an open area and takes cover. Nothing happens. There’s no sound from an alarm system. But anyone who might assume that security isn’t assigned the highest priority here would be very much mistaken. When the intruder approached the fence, an alarm was triggered in the facility’s central monitoring office. But the intruder, noticing nothing out of the ordinary, believes he has gone undetected and leaves his hiding place. What he cannot know is that a video camera has long since detected his presence and is tracking his every move. When the unauthorized person enters a hangar, specialists from the security service are waiting for him.
At today’s most advanced airports, hundreds of cameras may be deployed to monitor perimeters and detect intruders. As soon as an intruder is detected, a monitoring system reports this development and automatically follows the person’s movements—regardless of his location, in any kind of weather and around the clock.
The technology that makes these automatic motion detection, analysis, and object tracking functions possible is one of the world’s most innovative security systems: the Sistore CX EDS (Enhanced Detection Solution) monitoring system from Siemens Building Technologies (SBT). "The heart of the system is intelligent image processing," explains Klaus Baumgartner, Product Manager for digital video systems and a specialist in digital video monitoring—otherwise known as closed circuit television systems (CCTV)—at SBT in Karlsruhe, Germany. "With Sistore CX, we have succeeded for the first time in designing a complete, permanently-installed monitoring platform with such flexibility that it can be equipped with new functions including automatic people tracking by means of a simple update."
A lot has changed since video monitoring was first used in the late 1950s. In those days, the objective was to use proven TV technology for the security sector. The first systems consisted of a few cameras that were connected to monitors, directly or via relay technology. The displays were intended to provide, in a sense, an "expanded field of vision” for watchmen, who suddenly were able to observe several areas simultaneously from a single control room. But as the number of areas to be monitored grew, so did the number of cameras. As a result, by the late 1960s, space limitations had made it impossible to connect each camera to an individual monitor.
This triggered the birth of the electronic video crossbar on the interface between camera and monitor, thus allowing numerous video cameras to be connected to this control device, and their signals to be transmitted to any monitor. However, although the camera signal on the monitors can change at five-second intervals, this quickly strains the limits of human perceptual abilities. "A study in the United States showed that with just two monitors with automatic image switching, an observer overlooks up to 45 % of all activities depicted in the images after 12 minutes," says Baumgartner. "After 22 minutes, that figure reaches 95 %." Furthermore, the video crossbar has a significant problem: It’s permanently wired to the cameras, so expansions are very expensive.
Beginning in the mid-1990s, thanks to digital technology, new algorithms for image analysis, and Internet technology, a solution to this problem began to emerge. At Siemens SBT, developers envisioned systems that could simply be added to a company’s intranet structure. Thanks to growing computing power, it was becoming possible to evaluate image data in real time. Furthermore, image compression technology based on the MPEG standard was finally up to the task.
Signals and Substance. "Thanks to the MPEG process, the enormous quantity of data from a high-resolution video signal can be reduced from approximately 160 Mbit/s to 1 Mbit/s," explains Baumgartner. In fact, his development department is working with a team at Siemens Corporate Technology that is developing an even more efficient compression processes as part of the international MPEG committee. The process exploits knowledge of the physiology of the human eye—that is, it removes hardly any perceptible information from the image. "This is similar to DVD technology, where the signal is reduced to five Mbit/s."
As a first step toward digital video monitoring, in 1999 Siemens installed a video signal converter between an analog camera and a video crossbar. This "video codec" with processor digitized the signal, evaluated it for motion, and transmitted it to the monitor. In the processor were algorithms that made automatic motion detection possible. "For example, in South Africa this digital video sensor was very popular with customers, especially in gold and diamond mines," adds Baumgartner.
A platform for a flexible infrastructure for digital security systems finally emerged with the arrival of the Internet and intranets. Experts at Siemens quickly recognized its potential and went to work on a solution that could connect any number of cameras with video codecs to monitors via the intranet. The result was a combination of hardware and software that makes it possible to access the cameras from a PC, which in turn made the video crossbar superfluous. "This way, a network is designed redundantly and can be expanded flexibly," explains Baumgartner. At the same time, the video codec can also be supplied with algorithms, which no longer have to be permanently programmed; instead, they can be flexibly updated.
With these developments, the foundation for the Sistore CX video monitoring system was in place. The existing intranet infrastructure transmits the signal from any number of cameras via Ethernet, while the software—in addition to providing automatic motion detection—offers additional functions, such as Enhanced Detection Solution (EDS) software upgrades. For example, the video sensor can learn the normal state of the monitor image by memorizing the most frequently-occurring state during a particular period, and then focusing more closely on any deviation from that state. "It’s about much more than just the difference between two pictures," says Baumgartner, explaining the changes that have taken place over the years. "Today’s system actually learns what normal means."
A surveillance system detects an intruder in a camera’s video image (bottom left) from his behavior pattern (bottom right) and tracks him on a monitor with multiple images (middle)
Man or Beast? EDS software also works with algorithms to achieve a "feature extraction" function. Using pre-defined parameters, such as size and speed, it is possible for it to distinguish a person from an animal or vehicle, for example. And with the help of "foot points" that mark the site of the object and compare it with its knowledge of the image, the system can determine the object’s precise location. Still another advantage of knowing what a normal background image looks like is automatic sabotage detection. If someone turns a camera to point in another direction, changing the background image, the camera no longer recognizes its usual environment and sounds an alarm.
Ensuring that these functions would also be possible with analog cameras was a must for developers of the system, who felt strongly that when old systems are replaced with Sistore CX, most of a facility’s existing technology should still be usable—as at Munich Airport. There, the old video system was still largely intact. "Only the video crossbar from 1992 no longer functioned," explains CCTV Product Manager Oliver Wiesner, who is responsible for video monitoring at the airport. "There were no spare parts available for that year of manufacture." Wiesner had the video crossbar replaced with a video codec and connected the system to the airport’s Ethernet. The decision to have Siemens handle the modernization was made quickly. "In addition to Siemens’ great quality and service, the compatibility of Sistore CX with the existing 1,600 cameras ensured the best price-performance ratio for us," said Wiesner.
Baumgartner and his colleagues are working on additional solutions, including continuous monitoring with digital cameras—a feature that will mean even more substantial improvements in image quality. Working along these lines, Siemens has installed a completely digital Sistore system for the 2006 Asian Games in Doha, Qatar. More than 1,300 digital cameras will automatically detect, evaluate and, if necessary, track suspicious movements in the city’s sports compound.
Siemens’ fully-digital monitoring system will also soon be deployed in Germany. The Federal Border Police in Berlin, for instance, are currently building a new headquarters that will be outfitted with a high-quality surveillance system. "The headquarters’ fence, which is 1.2 km long, will be protected by a system that will safely cover the entire area with only 40 digital cameras," reports on-site Project Manager Markus Sasse.
Looking ahead, Sistore CX developers will focus primarily on new algorithms. "Soon our system will be able to locate abandoned objects in airports," predicts Baumgartner. "The video sensors will memorize not only the background, but also every permanent change that takes place in the foreground."
And considering the steady increase in computing power, Baumgartner foresees the automatic identification of persons. "Even in crowded pedestrian zones and sports events, cameras will be able to detect troublemakers and people wanted by the police—simply by recognizing their faces—and report them to the authorities," he says. But don’t hold your breath. It will still be a years before security systems will be able to identify people regardless of weather and lighting conditions, not to mention their constantly-changing appearances.
Sebastian Webel
An EU-funded project should make airliner highjackings more unlikely than ever before. Using video cameras and microphones, the SAFEE (Security of Aircraft in the Future European Environment) computer system records what happens on a flight and compares the events with stored images and records of critical situations. If the system detects suspicious movements or conversations, it sounds an alarm and automatically sends an encrypted emergency call. And if terrorists actually manage to penetrate the cockpit and attempt to divert the aircraft, SAFEE compares the flight position with off-limits zones that are stored in its memory, blocks access to the steering controls, and automatically returns the flight to its authorized path. Siemens is one of 30 participants in the 36 mill. € project.