|
|
| Photo by Bosch Security Systems |
Smart surveillance is not new. Essentially it involves digitally analyzing the signal from a CCTV camera using smart Video Content Analysis (VCA) algorithms capable of identifying and marking specific features in the video content that may require action. But the algorithms nowadays are getting much smarter and have come a long way from the relatively simple Video Motion Detection (VMD) algorithms of a few years ago. VCA compares real-time video with ¡®known rules¡¯ for detecting alarm situations or suspicious activity. Today¡¯s VCA algorithms deployed in the more advanced video security solutions are capable not only of motion detection, but also of detecting specific objects like abandoned baggage, direction of movement and even suspicious behavior such as loitering -- all in real time. These systems also operate 24/7, are never distracted or bored and never have an off-day -- in contrast to their human counterparts. Moreover, they are clever enough to spot patterns of behavior that may pass unnoticed by even the most vigilant human operator. VCA-enabled systems can also easily be configured to trigger a recording only when events that violate the ¡®Known rules¡¯ are detected, dramatically reducing the amount of recording disk capacity required. What¡¯s more, today¡¯s smart surveillance systems can significantly cut costs by reducing the number of security personnel needed to monitor the video, and allowing them to concentrate exclusively on incidents that require human intervention or decision making.
By Ruud Toonders
THE ALL-SEEING EYES
The latest VCA algorithms utilize the increased processing power and intelligence inherent in today¡¯s digital video security products.
Changes in environmental conditions are one of the major causes of false alarms in VCA systems. To eliminate this, the IVMD functionality includes an advanced background-learning algorithm that suppresses unwanted notification from, for example, moving tree branches, leaves, clouds, shadows and falling rain and snow. The algorithm is also capable of dynamically adapting to changes in background so that alarms are triggered only in real alarm situations.
These advanced VCA algorithms also allow the user to create independent detector fields with individual alarm trigger parameters. Even completely overlapping or identical shapes with different trigger parameters can be set. In the case of Bosch¡¯s IVMD 1.0, alarm parameters such as size, speed and direction discrimination can be set up independently for each of the detector fields, allowing, for example, a single camera to monitor an outside parking lot and detect whether cars are speeding or driving the wrong way. It is also possible to calibrate the cameras to allow objects to be distinguished by their size. This calibration includes the addition of 3-D camera perspective correction necessary for reliable size and speed discrimination in all directions. With this function, alarms can be triggered only on objects below or above a specified size, for example just for trucks or cyclists, or just for objects travelling below a specified speed, such as a pedestrian walking along a highway.
MONITORING SUSPICIOUS BEHAVIOR
It is difficult to precisely define what constitutes suspicious behavior. Often it¡¯s a matter of experience and even intuition, as any long-serving police or security officer will attest to. But there are certain patterns of behavior that can be interpreted as potentially suspicious such as people moving in the opposite direction to the normal flow of traffic when approaching an airport security checkpoint, or continually returning to the same point for no apparent reason. Advanced VCA algorithms can be set to trigger on such behavior.
Today¡¯s advanced VCA systems are even capable of automatically distinguishing between people and vehicles or other objects, which could be important when monitoring for suspicious behavior on a busy street full of traffic. This can be done on the basis of relative size but the effects of perspective can confuse this.
With the heightened fear of terrorism particularly at airports and other transportation hubs, unattended luggage is an immediate cause for suspicion. Aspect ratio can be used to trigger alerts, but more powerful functions that can trigger an alert on a change in the idle state of an object being monitoring are also available for this. This idle object detection function can, for example, detect whether a car is stopping on a highway carrying fast-moving traffic. The difficulty here lies in distinguishing between suspicious and innocent events, and there is currently some skepticism about the effectiveness of these functions in venues with lots of human traffic where they could generate too many false alarms. For the present, more likely applications for these alert functions are in detecting vehicles stopping in the wrong place (for example, stopping too long at an airport drop-off zone or in red zones where they are not allowed to stop at all), in museums and art galleries and in high-quality shops where they can be useful for monitoring shop displays, for example, a rack of highly-priced coats.
An important area where these new VCA systems are increasingly being deployed is in what is known as critical perimeter surveillance. This includes, for example, surveillance at the perimeter of airports, which are vulnerable to attack by terrorists using shoulder- or vehicle-launched missiles. They are also being deployed in other vulnerable areas such as oil refineries, large industrial and chemical complexes, nuclear and other power generation facilities and water-pumping stations. These complexes are often surrounded by perimeter fencing which may be more than 50 km long and, until recently, their only security has been barbed wire. Typically these facilities lack video surveillance, and only a small number use some type of microwave or fiber-based intrusion-detection systems for perimeter monitoring. But these systems tend to be very error prone, generating so many false alarms that operators can become desensitized and have been known to go to the extreme of turning the system off because it becomes such a nuisance. Here, in particular, smart surveillance based on advanced VCA systems is expected to have a major impact on security. It will allow huge tracks of perimeter fencing to be reliably monitored by automatically identifying suspicious behavior like vehicles pulling up and stopping for long periods, and distinguishing between animals and possible intruders approaching the fence line.
1 2 3 4
CLASSIFICATION TECHNIQUES
VCA algorithms currently deployed invariably rely on identifying objects simply by defining a frame around the object. For the most part, this approach is effective but it can lead to a degree of uncertainty in identification -- is it a suitcase or a child in a stroller? In the future, however, new advanced classification features will make it possible to identify objects far more accurately by associating them with a set of classification parameters -- precise shape of object, size, color, original movement, and other such criteria. An abandoned suitcase will then clearly be recognized as a suitcase. These new algorithms also allow moving objects to be tracked much more accurately, even over several CCTV cameras. So if, for example, the police are looking for a blue getaway car involved in a robbery, it will be possible to set the search parameters within the algorithm to alert security personnel if a fast-moving blue car appears on screen.
NO HIDING PLACE
Besides creating alarms in real time, the digital data generated by VCA systems is also recorded for use after the fact as a forensic tool. VCA already dramatically reduces the amount of video that needs to be searched through but nevertheless the search could still be laborious if it has to be done manually. To facilitate searching recorded video, the more advanced VCA algorithms also serve a video-management function. They automatically generate metadata which is sent together with the video footage to the recording medium. The metadata is a text string usually comprising a few key words describing specific scenarios with information such as patterns of behavior, aspect ratio, color of objects, date and time of event. The metadata files are therefore much smaller and easier to search through than the digital video files. Typically, a manual search of video footage that may take days or even weeks can be completed within seconds by searching the metadata with smart search facilities similar to those provided by an Internet search engine. And as with a search engine, the search can be refined to limit the number of hits. The final search results are displayed with the associated video, which can then be checked manually to find the exact event being sought.
INTELLIGENCE AT THE EDGE
Traditionally, VCA systems have been implemented as stand-alone software systems running on an industrial PC platform. These PCs are installed in a central location (e.g., in the control room), and continuous streams of monitored video are transmitted over the network for analysis and archiving. This architecture has two significant drawbacks. First, the continuous transmission of video places considerable strain on network bandwidth. Second, the use of PC-based platforms for video analysis adds unnecessary hardware costs. These limitations can be overcome with the new generation of IP-based CCTV systems. By taking advantage of the advanced processing power already available in the IP-based video encoders connected to the cameras to run the VCA algorithms, the VCA functionality can be moved out to the camera site. Better yet, with an IP camera this functionality can be integrated directly into the camera itself. In this scenario, video content is analyzed, compared against the ¡®Known rules¡¯ and events are generated at the camera site and only video footage of interest (e.g., abnormal events) is transmitted to the control center. This approach dramatically reduces traffic on the network, and eliminates the overhead of a separate PC for running the VCA software.
This ¡®intelligence at the edge¡¯ as it is known has become commonplace in the IT world. Besides the reductions in network traffic it provides, it is commonly accepted that the nearer to the edge the intelligence is located, the more reliable a network becomes since if one intelligent device fails you don¡¯t lose the entire network. In the case of smart surveillance, this approach is now being taken on board by many of the leading security-system manufacturers and is expected to become the de facto standard architecture for smart surveillance in the future.
Ruud Toonders is Product Marketing Manager EMEA - IP Network Video, Bosch Security Systems (www.boschsecurity.com).
For more information, please send your e-mails to swm@infothe.com.
¨Ï2007 www.SecurityWorldMag.com. All rights reserved.
|
