Gun Detection Analytics Are Here, Says StartupBy: Brian Rhodes, Published on Jul 03, 2014
A timely innovation, or cashing in on tragedies?
A new video analytics company is touting its 'gun detection' analytics that can trigger as soon as a weapon is drawn, potentially alerting authorities in seconds and saving lives.
Are they crazy, or will this be a powerful defense against active shooters and lower the risk of gun violence? We spoke with the company and take a look at its features, pricing and positioning inside.
The promo clip below demonstrates SafeZone's Firearm Detection analytic. The red boxes below calculate the distance of the gun from the camera, the relative speed of the the shooter, and the confidence interval that a firearm is recognized:
SafeZone's manufacturer, ReardenIT [link no longer available], claims the analytic alarms regardless of gun type; handguns, rifles, or shotguns alike. This video below is the same demo on 'assault rifles' compared to the handguns in the previous clip:
The company has specific requirements on the cameras combined with SafeZone. At a minimum, cameras must have:
- 720p or better resolution (no PPF stated)
- 10fps or greater
- Network or IP based stream
- Fixed FoV Standard: The company disclaims the use of PTZs as "available, but must be analyzed on individual camera basis"
- Low Light Performance advised (but no firm specification given)
- Cameras are optimally mounted at a height of 10 feet above ground (although 'any placement works, but affects range'.)
SafeZone resides on a customized server (called Sentinel) containing 'parallelized GPUs' that can handle up to 100 fps total throughput. At minimum recommended framerate, one Sentinel can support 10 cameras at 720p resolution.
SafeZone is expensive compared to even high end analytics. The company emphasizes that exact pricing depends on customer needs and site adjustments, but states MSRP is ~$4000 per camera for tear one, and subsequently $400 per year for software maintenance.
Initial purchase includes all required SafeZone hardware, deployment services, qualification and performance validation.
Given the bold claims, we pressed for detail on just how SafeZone picks firearms out of a scene, but avoid alarming on benign items 'like a broomstick' for example. SafeZone responded with:
"[We use] a highly developed classification-based system that recognizes shape primarily, but color is also factored in. The highly refined classifier can instantly "see" the difference between a broomstick and an AK."
The company also reports the analytic uses "separate classifiers for long guns and handguns" and SafeZone "has less sensitivity to silver colored firearms than black".
ReardenIT is the exclusive dealer of SafeZone. All sales, install, configuration, and support come only from the company, with no plans to expand to outside vendors. The company claims this is in part because of the customization and configuration required for each scene, suggesting:
"SafeZone calibration and installation testing will quantify detection capability for each camera. Factors such as physical scene geometry, lighting, and optics are examples of scene-specific characteristics that SafeZone is designed to adjust to."
They will also source suitable cameras or repositioning labor if needed, but will work with a customer's existing surveillance vendor if needed. SafeZone asks that customers provide "Floor maps containing each area secured with SafeZone, allows for the calibration of each camera, and a list of e-mail addresses and text message numbers to be notified when SafeZone detects a threat."
How Much Gained?
Presuming it works as advertised, what does it gain you? While the matter of notification time could be lessened if the system correctly identifies the firearm of an active shooter, it does not guarantee response will be any faster or more effective than without SafeZone.
Indeed, if the shooter already has a firearm drawn, firing on targets will most likely occur within seconds of visual verification. At that point, the noise of discharge and the clamor of the event will likely be a stronger indicator of trouble than video analytics.
False Alarm Ratio
In most facilities, even schools where the fear of shootings is high, guns are rarely pulled. The frequency that a real gun is pulled in front of a SafeZone enabled camera is likely very low - once a year, perhaps longer.
By contrast, even if the system is extremely accurate and only false alarms once a week, that would be 50 false alarms for every valid one - a ratio that would cause many operators to lose faith in the analytics.
This is a fundamental issue for any analytic that attempts to identify extremely rare events.
Since guns come in many size, shapes and colors and camera positioning can warp what objects look like (shadows, glare, bad angles, etc.), it is also a concern if the analytics fail to identify a real drawn gun.
Going To Draw Interest
No doubt, though, gun detection is one of the great hopes for security managers, so we expect that the product will draw interest and perhaps heartache.