AI is becoming increasingly important in the face of more sophisticated security threats and increasing workloads. But what is it and how does it help automate video surveillance?
We’re now seeing a revolution in AI (artificial intelligence). AI gives computers human-like performance in a specific area. The benefit is that a computer can process huge amounts of data. Also, a computer doesn’t get bored or tired, and can work around the clock. For example, AI might find patterns in what people buy from a supermarket. In another example, it might detect credit card fraud.
“Deep learning” is one common approach. Deep learning is “trained” on hundreds of thousands of images and can recognise images with human accuracy. For example, deep learning could learn to distinguish a cat from a dog or recognise a gun. Deep learning is well known but there are other AI approaches as well. Developers select the best approach for the type of problem. AI is used in new ways every week. For example, Facebook uses AI to label faces in photos. Also, Google uses AI to read street signs. Banks use AI to approve housing loans.
Computer vision is a growing area that is related to AI. Video cameras are the “eyes” of the computer. AI is the “brain” that makes sense of the incoming video data.
AI and computer vision together is ideal for the security industry. Now, AI can automatically monitor security cameras. AI can work with hundreds or thousands of cameras. It can identify unusual events as they occur and alert an operator.
Human operators are easily bored. In fact, studies show that a person can’t easily monitor camera feeds for more than a few minutes before losing focus and missing events. But humans are good at judgement and context. The best approach is for operators to work together with AI.
Video analytics is a growing area. It may include AI as well as non-AI techniques. There are two different security needs:
Real time alerts – alerting an operator as an event occurs, so that a security guard can attend. This allows intervention, which reduces the impact of the event.
Labelling video – this helps speed up review of footage. You can use this to find problem points which you can then addressed.
At a basic level, video analytics lets you set up a rule. For example, you might want to alert an operator every time someone crosses a line. You can define this as a “rule”. You may need to setup hundreds of rules for a larger installation. This is a low level of AI. Rules based systems are not new and are commonly available.
The next level of video analytics uses deep learning for object recognition. Deep learning works well for recognising faces and objects (such as letters, numbers or physical items). Object recognition can find a license plate on an image of a car and read the characters. It could also identify the make and model of the car. It can identify an openly carried gun.
Face recognition is used for access control or border control. This works best when you register each face. For example, Australia now uses face recognition for passport checking. This works best when the person is directly facing the camera.
One issue with deep learning is that it can require quite a lot of computer power. You will need to buy high end servers. Deep learning based video analytics is now becoming widely available.
Behavioural analytics is the most advanced form of AI video analytics available. It can tell the difference between ordinary movement and unusual movement. For example, in a shopping mall, normal movement is people walking along, whilst fighting is unusual. A behavioural analytics system will detect a fight and alert the security team.
These systems are not widely available yet. The challenge is building a system that can recognise all types of unusual activities. Unlike object recognition, which uses still frames, behavioural analytics must use video motion. It is more difficult to recognise actions than objects. Also, a system must be able to generalise without needing hundreds of thousands of scenes for training. It must also be practical and not have excessive server needs.
iCetana is a behavioural analytics solution that is installed all over the world. iCetana is used at all types of sites, including universities, shopping malls and office buildings. It can provide both real time alerts, and label video. iCetana works with your existing camera network.
iCetana is based on a patent developed from a university AI project. One strength is that it can learn what is normal. This allows it to highlight any unusual actions, as the events occur. It can identify abnormal events without needing to see thousands of examples first. Also, it learns and adapts to what is normal for each camera.
Also, iCetana has moderate server needs. One server can support up to 300 cameras. This keeps the total cost per camera low.
AI can help you make the most of your existing video security system. AI video analytics can help you reduce business risks by giving better visibility of events. Reduced risk means better protection of business reputation and fewer injuries and losses.
You can start with a smaller number of key cameras to show value, before moving to a larger number of cameras.
For practical advice how your organisation can start leveraging AI, contact the iCetana team or get in touch with one of our Partners.