AI Is the Latest Step in Video Analytics

Surveillance video implements many cool technologies to enable us to prevent or solve crimes. If you think about the whole system end to end, we have cameras that take in photons, turn them into electrons, which are carried across a variety of networks and come out the other end as photons from a monitor. Along the way, these electrons might also get encoded into magnetic fields to be stored on hard drives or, in many cases, stored as a captive electrical charge in a solid state drive.

We have descriptive terms for the various processes that make up a modern networked surveillance system. At a high level, video codecs handle encoding of what is by default a very high-density signal from a data perspective into something that trades small data losses for easier transmission, commonly referred to as video compression. Examples of video codecs used over the years include MJPEG, H.264 and H.265. Now AV1 is emerging as a codec standard. No matter what mechanism or compression algorithm is used, we call these collectively codecs (sometimes capitalized as CODECs, for coder/decoder).

Once video is encoded, it needs to be moved across a network. This is where video transport technology comes into play. You may have heard of protocols like RTSP, RTMP, DASH or WebRTC. Much like video codecs, they all have pros and cons, and some are broadly considered better (or worse) than others, or more or less modern, secure, etc. But they are all transport protocols.

When it comes to digitally storing video, once again we have a variety of solutions to the problem. From the media used (hard drive, SSD, tape archival) to how the data is structured and written to the media. And, as always, pros and cons, common approaches, etc.

Surveillance video ultimately exists so that we can see what is happening in real time, or so we can go back and review what happened. We now produce far more video than can be watched or viewed by a person. Fortunately, video analytics can automatically find the most relevant or interesting bits of information in the video stream.

Like everything else discussed so far, video analytics dates back to the earliest days of digital video (and, in some cases, even to analog video). And also like the other technologies mentioned, video analytics has continued to evolve, advance and incorporate new methods of delivering results for the part of the stack it is meant to address. If you’ve been following this segment of the industry for a while, you may have heard of things like “background subtraction,” “pixel motion detection” or “scene calibration.” And you have almost surely been hit over the head with use of the term “artificial intelligence” (AI) as a mechanism for performing video analytics.

Much like AV1 represents the latest technological progress in video codecs, or WebRTC the latest mechanism for streaming video, AI is the latest approach to video analytics. In short, AI is video analytics. Trying to classify them as two completely separate and distinct concepts is as foolish as trying to say that AV1 is not a codec, or that WebRTC is not a streaming protocol.

From a broader perspective, AI, or more specifically deep neural networks (DNNs), are being used to solve a wide variety of problems (while, in some cases, creating new categories of problems). Video surveillance applications are but a tiny part of where AI/DNNs are now being used.

When video surveillance, or even access control companies, say “We are using AI,” this is the equivalent of saying “We are using transistors.” Of course you are. It is the obvious solution to advanced data processing in the current technological landscape.

The views and opinions expressed in guest posts and/or profiles are those of the authors or sources and do not necessarily reflect the official policy or position of the Security Industry Association.

This article originally appeared in All Things AI, a newsletter presented by the SIA AI Advisory Board.