The Top 6 Tech Questions to Ask an AI Surveillance Provider

Introduction

All AI Surveillance technology is not the same. Some of it isn’t even good. But how can you tell the difference unless you’re an engineer? Here’s a cheat sheet of things you’ll want to ask.

The Questions

1. 01Where is the video stored and processed - onsite, on the edge, or in the Cloud?
2. 02Is the Machine Learning Supervised or Unsupervised?
3. 03How is the network architecture secured?
4. 04How good is the AI?
5. 05Who is going to respond to all the AI alerts?
6. 06Who owns/controls your surveillance data if you subscribe - you, or your provider?

Where is the video stored and processed - onsite, on the edge, or in the Cloud?

Video surveillance is Big Data; it takes a lot of storage. To compound matters, AI is a huge computational challenge, and also takes a lot of processing power. Providers that require your onsite storage and processing make you responsible for managing RAID arrays, OS and application updates, large-scale data backups, sufficient HVAC for cooling, and more.

Processing video on the edge has some advantages, such as reduced latency and lower bandwidth usage, but it also comes with notable challenges: limited processing power for AI plus higher costs, scalability issues, and finite storage.

Look for:

An AI Surveillance provider that stores and processes in the Cloud. Cloud brings business continuity, disaster recovery, etc. If your building floods or burns down, the data is all safely offsite, still ready for you to access.

Is the Machine Learning Supervised or Unsupervised?

Supervised Learning (otherwise known as Human-in-the-Loop) is essential for a high responsibility environment like security. Ask every potential provider whether they do:

• In-house supervised learning and curation of content.
• Offer training for customers to continue to curate their surveillance content, particularly when it comes to facial recognition.

Look for:

Providers that have a human in the loop model to monitor and accelerate their AI model. Ask if they have a customer interface for your own human-in-the-loop training as well, so your system can continue to improve with your usage.

How is the network architecture secured?

There have been some pretty catastrophic hacks into enterprise video surveillance over the years. Ask about all cybersecurity measures. Look for a system that:

• Doesn’t need a fixed IP address
• Introduces no holes in your firewall
• Nor any port forwarding in your router.

Look for:

A closed system as above. Because an ounce of prevention is worth a pound of cure.

How good is the AI?

Everyone has a cool AI demo nowadays. However, demos and real world performance are two very different things.

Where does the data for their training models come from?

You may have heard the expression “Garbage In = Garbage Out.” Computer vision AI is developed from training models where engineers teach the computer to recognize and detect objects, people, and cars by giving them millions of images to sort.

But the images need to reflect real world use. If the company’s training models work off of what is essentially “clip art” it’s not going to work well in the wild. Want to detect a gun/backpack/bike/dog/etc? You first need to train your model on many images of guns/backpacks/bikes/dogs/etc from every angle, including partial views. Most commercially available models don’t train off of real video surveillance data. Here’s the difference:

Who owns the computer vision technology - the provider or a third party?

Every Big Tech company is selling some kind of computer vision technology now - meaning you can buy rights to their gun detection, or facial or license plate recognition. Not only are solutions from Google, MSFT, AWS, etc. built for this general purpose, but they are crazy expensive to run at scale. Any AI Surveillance provider that doesn’t have their own AI technology has to take expensive tech from Big Tech and then mark up the price and resell it to you.

Also, bear in mind that those who give can also take it away, particularly when it comes to facial recognition software. Big Tech providers have been turning critical features on and off to meet the whims of the market.

Who is going to respond to all the AI alerts?

If you have multiple buildings, each with dozens of cameras, yes, the AI can do a great job of filtering out the vast majority of the noise and only present alerts that are likely relevant but … that’s still a lot of alerts. Who is going to monitor all these alerts – you, the provider, a third party? Utilizing AI Surveillance the way it’s meant to be used - to actually stop crime in real time - will require Remote Guarding software as well as Remote Guarding services.

Look for:

A platform that offers fully integrated Remote Guarding software and services to keep your AI Surveillance economical, as well as functioning as a real time, crime fighting tool.

Who owns/controls your surveillance data if you subscribe - you, or your provider?

Always read the fine print. While it’s your business and your cameras, a number of surveillance providers have in their contract that they control the data, even when you want to directly access your footage and analytics. They will insist they do the investigation for you and charge you an additional fee for the privilege of learning the results. They may even use your surveillance data for marketing purposes without your permission.

Look for:

A provider that ensures that you exclusively own and control your data.