Video analytics represents something of a holy grail to those in the security industry. Computers have long been able to scan text and even audio for keywords or phrases, but analyzing video — especially in real time — is considerably more challenging. In recent years, however, major improvements to artificial intelligence (AI), machine learning and deep learning capabilities have given rise to impressive new tools capable of analyzing video with minimal input from security personnel.
As companies look to invest in these new technologies, it’s important to establish a baseline understanding of what terms like artificial intelligence, machine learning and deep learning actually mean — and what these technologies are capable of. Education will be increasingly critical as we move away from relying on human-based security and lean more on technology to identify and alert us to anomalous or troubling behavior.
Drawing Important Distinctions
Companies hoping to portray themselves as cutting edge have thrown around the terms AI, machine learning and deep learning almost interchangeably over the past several years. Unfortunately, some companies claim to use AI or deep learning under the pretense that it sounds innovative, without clearly defining the benefits or differentiators. But the truth is that each term refers to something very different. A device with true deep learning capabilities will function much differently than a device with AI capabilities. And while it’s true that some users may not know the difference, more discerning users must be able to trust that a company advertising deep learning functionality actually means what it says.
The distinction is simpler than it might seem at first glance. AI is an overarching term that encompasses both machine learning and deep learning and refers to any demonstrated machine intelligence. Although the idea of superintelligent AI has been popularized by pop culture and science fiction, the real-world definition is much less grandiose and might refer to something as simple as an algorithm designed to win a match of tic-tac-toe. AI-powered machines tend to be able to perform a specific task well and develop better skills over time, but they may be limited in their application.
Machine learning is a more complex subset of AI, and the term is used when signs of basic cognition (the ability to learn) become apparent. Machine learning technologies use algorithms to solve problems like basic AI but can more effectively adjust to match different input data. This means that machine learning can be trained to complete a certain task or serve a certain function — a major step forward from the hand-coded software routines of the past.
With deep learning, an even more advanced form of machine learning, things become even more complex. Inspired by the way the human brain processes information, deep learning-capable machines can use large amounts of data to identify patterns, classify information and make decisions by labeling and categorizing what they see. For those in the security industry, the advent of deep learning has proven to be the key to more effective video surveillance and analytics.
Smarter Cameras Mean Stronger Security
It’s hard to overstate how important video analytics is for the security industry. Human beings are only capable of devoting so much attention to what’s happening on a screen — let alone the thousands of cameras in large city surveillance or corporate security systems — and even the most observant humans can’t be expected to see everything with perfect accuracy and recall. The value in traditional video surveillance often comes in the form of recording video to review later if an event is brought to the security team’s attention, so it isn’t an exaggeration to say that the vast majority of video captured by surveillance cameras without analytics is discarded without ever being looked at.
The implementation of deep learning and AI has helped to ensure that surveillance footage no longer goes to waste. Deep learning is used to train video analytics to better recognize and identify things like activity in an off-limits area, with new applications for the technology in development every day. When training AI, the more data you feed it, the more effective it will be. Cameras and sensors provide a massive amount of data, giving the associated algorithms plenty of opportunities to refine their capabilities, improve their recognition and reduce false alarms.
By using deep learning to engage in video analytics, today’s cameras are augmenting human security personnel, helping them improve their ability to both identify and react to potentially alarming activity. To highlight one application, the accuracy of surveillance cameras equipped with automatic incident detection in areas with heavy shadows, cover or glare is often questioned. Research conducted by my company indicates that deep learning increases detection accuracy from 85% to around 95%, reducing false alarms by two-thirds — a huge help for staff.
One final point: Understanding what today’s AI technology is capable of is important not just from a practical standpoint, but from an ethical one as well. Even companies that are on the forefront of AI research continue to grapple with how the technology should be ethically deployed. The importance of AI to the future of security is clear, but fully examining the societal implications of the technology is a must.
Know What You Want — And Know What You’re Getting
The security market — and the technology market in general — is filled with companies promising AI, machine learning or deep learning capabilities. Improving your knowledge of those terms will provide a more thorough understanding of what today’s security technologies are actually capable of, which can mean the difference between investing in a product that actually meets your security needs and investing in one that doesn’t. This can help you not only identify when a company is overpromising on its capabilities, but better understand the technology’s ethical implications as well — enabling you to make better, more informed decisions. AI and its subsets have helped make today’s cameras and sensors more advanced than ever, and understanding these new capabilities will make it easier than ever to match your security needs with the right products and technologies.