HOW HAS MACHINE LEARNING PROGRESSED IN THE LAST 5 YEARS?
One word: processing. It’s the ability and speed at which complex algorithms can be performed in order to be effective for real-world applications. Traditionally, software has relied on CPU processing which has served us well for decades. Over the past 5 years, we’ve seen how GPU processing has been leveraged to perform these functions due to its far superior processing power. You’re witnessing this progression through self-driving cars, artificial intelligence and medical simulations to help cure diseases to name a few.

We’re also witnessing its benefits within the security industry. I’ve seen many video analytics companies over the past decade either fail or under deliver. For the most part, it wasn’t because the coding was terrible, it was because the technology available at that time couldn’t keep up. Thanks to this thing called innovation, the processing that is available today is much, much faster and its footprint has shrunk. And because of this, we have the ability to perform processing out at the edge (camera) versus having to send back all of that information to a server with limited processing power. Couple this with machine learning, and we’re starting to see analytics really take off.

WHAT IS CONSIDERED STANDARD VS. INNOVATIVE FOR VIDEO ANALYTICS IN 2018?
What has long been considered standard has been basic level analytics, such as object left behind, motion detection, tripwire, object tracking and object recognition: human vs. vehicle vs. animal etc.

Today, you’re starting to see a big push for greater intelligence. Clients want far more than facial detection. They want to be able to determine gender, clothing, gestures, accessories, etc., with the ability to detect in real-time, perform forensic searches, and cataloguing. This requires heavy duty processing made practical through GPU’s.

Another example is addressing firearms. The tried and true methods require sensors strategically placed to determine that a firearm has been discharged and then triangulate its whereabouts. Today, there are video analytics that can begin to identify firearms within view and hopefully prevent an active shooter situation.

Pattern and behavioral analytics have become far more accurate in recent years. For example, determining patterns within a scene such as traffic flow and generating an alarm if a vehicle or human veers off course. Determining patterns can also be used for developing heat maps which has its own slew of use cases.

DO YOU THINK THE CURRENT HYPE WILL BE SHORT LIVED OR IS THIS TECHNOLOGY MAKING A LONG-TERM IMPACT ON THE INDUSTRY?
You can tell by my previous responses that I’m a firm believer in the technology. We’re seeing the progression and benefits of machine learning and AI outside of our industry, and that won’t be any different within our industry either. Unless of course there is an overwhelming number of bad products out there. We’ve seen that before and it has made buyers more jaded. But, I still believe advancements in technology should limit a full repeat of the past.

WILL AI MAKE TRADITIONAL VIDEO MONITORING (OPERATORS) OBSOLETE?
I’d like to say never, but you know the saying. I believe the two will co-exist and they will rely on each other. With that said, there will be efficiencies where you may not need the same number of operators to perform certain tasks. Much of it will be automated, including the decision making. However, humans may still need to be involved in verification.

WHAT DO YOU CONSIDER TO BE BASIC OR MANDATORY ANALYTICS THAT ANYONE WHO DOES MONITORING SHOULD HAVE IN PLACE?
This is somewhat difficult to answer as it depends on the threat vectors, the purpose of monitoring, and the scene within view. Most systems have a good handful that are readily available and should be applied on a case by case basis. For anything advanced that would be considered an add-on, I would consult with technology experts who should have a full understanding of what you’re hoping to accomplish and set expectations. Its important to understand any system’s capabilities and limitations.

WHERE WOULD YOU EXPECT TO SEE MACHINE LEARNING FOR PHYSICAL SECURITY GO IN THE NEXT 5 YEARS?
Machine learning spans far beyond video. I can also see it applying to the cyber aspect of our products: being able to detect anomalous access or a breach on the backend (i.e. servers and databases), or developing behavioral characteristic of its cardholders based on badge activity, also, with understanding how its own operators are using the system.

Beyond that, I see video playing a critical role in access control. The holy grail has always been frictionless access. There are ways to mitigate some of that pain through biometric and Bluetooth technology, but imagine being able to freely walk through a campus without having to perform any special gestures and only being limited by who you are. Its been tried before and its success is still hit or miss, but I still believe that’s where we’re headed.