Page 18 - Security Today, March 2020

we need to take appropriate action. Applying artificial intelligence and ma- chine learning, we can scan video data for specific images based on physical character- istics, movement, behavior, color and other criteria. These sophisticated algorithms can help security staff differentiate between benign events and menacing danger and automatically initiate a suitable response (e.g., auto-lock a gate, alert a guard, launch a drone, trigger an audio warning to tres- passers, etc.) based on previous experience. Applying predictive analysis to the video metadata assists in projecting future events and behaviors – a decided asset for security personnel who strive to be more proactive
in heading off trouble.
Creating Smarter Perimeter Protection
At its roots, video analytics falls into three categories: pixel-based, object-based and application-specific.
Pixel-based analytics sends an alert when it detects changes in the pattern of pixels in the video frame. It’s used in appli- cations like motion detection and camera tampering.
• Object-based analytics is far more so-
phisticated in that it can recognize and differentiate between objects such as cars, people, trees and buildings and track them.
• Application-specific analytics use pixel- and object-based information to exam- ine the video for specific criteria such as license plate identification, facial recog- nition or fire detection.
Innovative developers are continuing to add to that portfolio, designing ever more powerful, targeted analytics to address the evolving challenges to perimeter and sub- perimeter protection. Let’s look at some of the significant advances to date:
Characteristics analysis. We’ve come a long way from basing event alarms solely on detecting pixel change. Now we have analytics able to categorize and under- stand the behavioral characteristics of those moving pixels. How does this benefit perimeter security?
Security can use video analytics to au- tomatically detect unusual behavior – such as a person walking by the perimeter of a restricted area and stopping for a short pe- riod of time. Before alarming, the analyt- ics could direct the camera to zoom out to look for associated concerns like whether the person is wearing a backpack or car-
rying a weapon. The analytics can direct the camera to zoom in close to the face to categorize certain behavior such as eye movement or expressions that might indi- cate suspicious intent.
Minimal pixel detection. In the past, video analytics needed quite a few pixels for reliable detection and classification of a person or object. Today there are ad- vanced analytics that only need one or two pixels to reliably detect a person or object over a mile away. This is a major break- through. Historically, for a camera to de- tect something long-range, it would have needed to be outfitted with a larger, more expensive lens. Unfortunately, the larger lens would narrow the field of view. This latest innovation takes long-range detec- tion to a new level.
Not only does the analytics enable the network camera (visible light or thermal) to detect a distant object using only one or two pixels on target, it does so while maintaining the camera’s normal field of view. This means users no longer have to sacrifice width for distance. Coupling this long-distance, minimal-pixels analytics with advanced camera features like optical zoom provides security staff with greater situational awareness much earlier than previously possible.
Thermal gradient analysis. Thermal cameras have always excelled at detect- ing people, objects and incidents under adverse conditions such as complete darkness, smoke, haze, dust, light fog and even bright sunlight. But today’s thermal analytics have gone well beyond simply detecting and identifying the heat signa-
ture differences between people, cars and objects. Now we have thermal analytics that can monitor temperature variances and the speed of temperature change and trigger an alarm if they detect unaccept- able variances. They translate data into an isothermal color palette to make it easier for operators to see which areas need their attention.
The implications for better safety and security are enormous. Thermal analytics can raise the alarm that self-igniting mate- rial – such as dust or oily rags – are about to combust. The thermal gradient algo- rithms can be used to predicting trans- former and switch gear failures at power substations in time to forestall wide-scale outages. Factories can use thermal analyt- ics to identify overheating machinery and leaky pipes. Refineries can use thermal analytics to monitor stacks for gas flares that might indicate pressure buildup that could damage critical equipment.
Specific application analytics. In recent years, we’ve seen a rise in application- specific analytics: everything from license plate recognition and facial recognition to perimeter defense, direction and speed detection, as well as smoke and fire recog- nition. We’re seeing facial recognition ana- lytics moving beyond the realm of simply recognizing and comparing facial features into the realm of recognizing and identify- ing emotions. We’re also at the forefront of artificial intelligence and machine learn- ing that will take video analytics into the realm of predictive analysis – giving secu- rity even greater situational awareness and earlier warning of potential threats.
16
0320 | SECURITY TODAY
PERIMETER SECURITY
Scharfsinn/Shutterstock.com