Page 52 - Security Today, October 2018
P. 52

License plate recognition is another common AI function.
Deep learning is the recent breakthrough within the field of AI that allows for machines to classify objects and make decisions based upon what they have “learned.” Deep learning algorithms that are already being deployed can not only enable AI devices to learn and perceive their environments but can also learn to differentiate everyday occurrences from abnormalities. This pattern recognition may seem simple, but it is a huge victory for analytics software. By giving machines “brains” to match their “eyes,” these offerings al- low for higher levels of accuracy and reliability in both object and
behavior classification.
The Merging of Radar and Deep Learning
While still a relatively new integration, the concept of enhancing pe- rimeter systems, featuring radar and other video technologies, with deep learning is taking hold in the security industry. The idea is sim- ple, combine the best technology for target detection with the best technology for target classification and merge them to create a fused engine that yields the lowest possible number of false alarms. Some AI software work only off of a radar signal, but more advanced so- lutions are extending the AI to the accompanying video stream to analyze data from both sources to get the most accurate results.
In the latter scenario, it all begins with radar detection. Once the radar filters out false alerts, it sends the validated target tracks detects to the VMS, which in turn, cues an integrated PTZ camera to follow the movements of said target. As the PTZ camera follows the intruder, deep learning software analyzes the video stream and tracked move- ment from the radar to classify the object. Once the target has been classified and validated, which takes less than a single second, a veri-
fied alert is generated, and the system will log and record the event. To better illustrate the point, take a large tech company with data centers in various geographic areas. One of the greatest challenges faced by these facilities is the fact that many are in remote areas with wildlife surrounding the perimeter. While the complexes must be pro- tected from vandalism and physical intrusion, these sites are regularly plagued by nuisance alarms. The largest companies own tens of data centers. Even if the number of daily false alarms is below five, the company could receive dozens a day from all the data centers com-
bined. This inundates security systems.
Many times, security guards or law enforcement will end up re-
sponding to erroneous alarms, which take their attention away from true threats. By utilizing a deep learning target classifier that analyzes data from both the radar and video stream, the nuisance alarm rate (NAR) can be significantly reduced. In fact, for successful deploy- ments, the system can replace physical guards, heavily reducing secu- rity operating costs.
There is a clear demand for perimeter security solutions that are effective in all conditions, and that reduce the NAR. While the in- tegration of radar, video and deep learning technologies are still in its infancy, development has been rapid and there is no doubt that this will continue. This new integrated solution
enables security to run at maximum efficiency while keeping costs minimal, simply by marrying the most reliable sensors with the most advanced analytics software available.
Yaron Zussman is the general manager at Magos America.
DEEP LEARNING
Make us your homepage!
Our website uses responsive design to adapt to whatever device you’re using.
32
1.5988 in
securitytoday.com
• Breaking news
• Relevant industry news and trends
• The newest security products
• Online product database and directory
• Trending topic and product videos
• Training through Security Today Academy
Go to sp.hotims.com and enter 7 for product information.


































































































   50   51   52   53   54