printer friendly version

From the editor's desk: When the rules change

Andrew Seldon, Editor

Welcome to the SMART Surveillance & AI Handbook 2026. We were a bit nervous about including AI in the title, since it either has a good or bad reputation depending on the individual – very few people seem neutral about it. However, when considering what end user companies demand from surveillance these days, it is definitely a very real part of the industry.

Naturally, there are many environments where AI and even traditional analytics are not used, but newer installations are increasingly considering what AI delivers. And this is where the challenge appears. We are at a time when an AI sticker must appear on the box if you want to be taken seriously. Many years ago, I wrote the same thing about video analytics – you couldn’t say your cameras weren’t able to support it.

But whether you need AI or traditional analytics depends on your camera’s job description. And the AI included may not be ‘real AI’. As noted in the publication, some companies take traditional analytics and slap an AI badge on it, while others do the hard work to ensure their AI systems can learn as they go. It’s about a rules‑based approach as opposed to systems driven by machine learning and deep learning.

Traditional video analytics refers to the automated analysis of video streams using predefined rules, pixel change detection, and simple algorithms. These systems operate on fixed instructions such as “Alert when an object crosses this line”, “Notify when motion occurs in this zone”, or “Count the number of objects entering or exiting an area”. Simple rules, unless you are the one who has to program the rules, then it’s not so simple.

AI‑enhanced video analytics, on the other hand, recognises complex patterns, objects, behaviours, and contexts within video streams. Instead of relying on predefined rules, AI learns from large datasets and continuously improves its understanding of video content. It is supposedly about understanding the context of situations – two people are dancing, not fighting, for example.

As noted in our roundtable discussion, while some are all-in on AI, there is still a place for rule-based analytics in many scenarios. Once again, scenarios where context and understanding are not required. In my experience, AI sometimes seems to really understand what you want, but at other times it misses the point completely. Maybe that’s just me.

Another topic we delve into in the handbook is hybrid scenarios, with a greater focus on the edge. Many cameras today offer embedded AI in their NPUs to handle the hard AI stuff in-camera rather than sending video to a VMS or cloud service for analysis. The cost per camera is higher, but there are benefits such as faster analysis and the need to send only real events to a control centre.

Nevertheless, as AI continues to advance and become embedded in nearly every industry – sometimes for mere display purposes – we can expect that in the near future, everyone will be AI-literate. However, whether we will still have jobs that enable us to afford AI is another question.

Feel free to send any comments and criticisms of the handbook to [email protected]. Human-generated emails preferred.

Share this article:

Further reading: