Page 25 - Campus Security & Life Safety, March/April 2019

would be removed from campus. In the case of this university, they upgraded their access management security measures, as well as hired more security personnel and promised to add more security cameras. An alternative to more security guards (i.e. more wages) is a Safelist, powered by face-rec, that would only allow residents to enter the dorms after-hours.
After the assault was reported, police searched surveillance footage for the offender. In any real-time search, or post-event investi- gation, identifying the target as quickly as possible is an incredibly high value task. Unfortunately, humans are terrible at video review and/or monitoring. After only 22 min- utes of sustained video review of a single monitor, humans lose up to 95 percent of visual acuity. Add another screen to review or live-monitor, and the results are just as dire. Augmenting human resources with AI is a force-multiplier in time-sensitive situations.
With an on-premises video analytics solu- tion, police can immediately search live cam- eras for, say, a man wearing a red shirt, or a
blue pickup truck, heading away from cam- pus. In addition to facial recognition, solu- tions powered by deep learning can create a unique signature of the whole person or object, using 260 dimensions instead of the normal three that we humans experience life in (has your mind exploded yet?), allowing operators to track the individual from cam- era to camera, even if their face is not visible.
Critical Facilities
At many universities, critical facilities such as nuclear research or robotics labs, medical centers with individual’s health records, toxic chemicals, or even opioids, are prime targets for theft. Instead of relying on costly security guards to patrol the premises, and still more guards to passively monitor mul- tiple screens in a command center, video analytics can be integrated into any VMS to better secure both the interior and exterior of these critical facilities.
Features like, Activating Intrusion Zones are a force-multiplier for critical facility secu- rity. Solutions with robust rules engines enable operators to create alarms based on time, zone, and action. If the medical storage facility that houses Fentanyl, for instance, has normal operating hours of 6 a.m. to 6 p.m., security can set intrusion rules to alert if any- one enters the premises after 6 p.m. Monday through Friday, and 24/7 on the weekends.
Additionally, if cleaning crews clean the facilities during open hours, security can set a rule to be notified if a person enters the opioid storage area wearing a yellow top (the hypothetical cleaning crew’s uniform) and remains in the storage room for longer than it typically requires to empty the single trash receptacle. Deep learning-powered video
analytics will enable security stakeholders to automate much of the mundane, but vital tasks that humans are currently relied on to perform—and repurpose human resources for higher value work.
Bespoke Analytics
Video analytics solutions built on deep learning platforms continuously improve, unlike legacy analytics, and in the very near future will give the power of creating analyt- ics to the end user. Democratizing AI and enabling customers to create bespoke analyt- ics with a few simple clicks of a button is what really gets AI-enthusiasts, like me, excited. Take another use case from a local university on the west coast—scooters.
Rideshare scooters like Bird, Uber and Lime have invaded college and enterprise campuses. Careless scooter drivers abandon scooters in unfortunate locations, such as handicap access ramps, doorways and eleva- tors. My father, who recently had a stroke, would have the right to sue were his only parking spot or accessible entryway blocked by a scooter or, worse, left in front of the elevator causing him to trip on the scooter and fall as he exits.
The answer? A scooter detector. Thanks to the processing power of deep learning solu- tions, it is possible to spin up a new detector in a short amount of time after an operator has flagged the image they wish to be able to detect. In this case, the ability to know when a scooter has been left in unsafe, risk-heavy areas not only increases safety but drives down associated insurance and legal costs. Liberty Mutual Insurance, for example, esti- mates that for every $1 spent on upgrading safety, $3 or more dollars is saved. Bespoke
By Brent Boekestein
Automating Campus
Security with Deep Learning
Security Solutions
Advancements in computing power have made possible ground-breaking physical security solutions
March/April 2019 | campuslifesecurity.com 25
Pasuwan/Shutterstock.com