Is AI really necessary for video surveillance?
AI is often used as a term to attract organisations with promises of cost savings and, often,
unrealistic, levels of accuracy. But is the new technology always essential? Chris Price reports.
When it comes to video surveillance, Artificial intelligence (AI) is often presented as a panacea
– a technology providing all the answers to all the client’s security problems. Not only can it save money, it can also offer unlimited functionality and, some claim, close to 100% accuracy.
However, the truth is somewhat different. While AI does have a growing role within surveillance, it’s not without its faults. Nor is it always necessary for video surveillance with many systems on the market able to provide solutions using existing data analytics without the need for investment in complex deep learning (DL) or neural networks that imitate the human brain.
“The way AI works is actually pretty crude,” Dean Drako, Founder and CEO of cloud-based video surveillance provider Eagle Eye Networks, admits. “You basically pick thousands of images that are what you want and thousands of others that aren’t what you want. Then you train the system by teaching it.” Where cloud-based surveillance companies like Eagle Eye Networks come in is that they can do a lot of training of AI systems, saving companies the time and expense of doing it themselves.
However, according to Drako, standard video analytics is more than adequate for many security applications, such as counting people in a particular area, detecting those who are loitering, camera tampering and spotting cars that are travelling the wrong way. “Although some people are starting to use AI for these types of applications, it’s not necessary. You can generally get pretty good accuracy with the latest video analytics.”
Empowering surveillance operators
Inevitably, though, some video analytics systems are better than others. “Traditional systems use BLOB (binary large object) analytics which are not particularly accurate,” remarks Jamie Barnfield, Senior Sales Director at IDIS Europe. “They often cause false alarms which means either customers don’t use them at all or they respond to an event that’s the result of harmless environmental factors such as a moving branch from a tree or a bag blowing across a car park, instead of an actual human intruder.” According to Barnfield the advantage of solutions that use the latest analytics systems is that it helps to reduce the number of false alarms, thereby ‘empowering surveillance operators to better detect crime or suspicious behaviour.
Nevertheless, accuracy is never going to be 100% even with the most advanced analytics or AI systems. “I read about one of the big supermarkets trailing an AI system that used a dataset of 300,000 faces to prevent staff from having to ask how old customers were to buy alcohol and cigarettes,” comments IDIS’ Jamie Barnfield. “There’s no way with a dataset that small that it’s going to be accurate enough.” Eagle Eye Networks Dean Drako agrees: “If you can’t teach a human to reliably tell someone’s age then you are going to have a very hard time teaching a computer to reliably do it.”
According to Barnfield, not only do some clients have unrealistic expectations about the accuracy of AI-based video surveillance systems, they can also be a distraction from the organisation’s real problems. “Some companies are losing tens of thousands of pounds a day in internal and external shrinkage while they wait for the perfect AI solution when they could be installing a regular HD plug and play system that could pay for itself in the first year,” adds Barnfield.
Improving video search
Yet while it’s true that not all companies need AI-based surveillance systems right now, demand is increasing as the technology advances and prices fall. For Eagle Eye Networks, which recently bought Bangalore-based AI company Uncanny Vision, there are two main applications where AI in video surveillance offers real benefits over traditional analytics. Firstly, when it comes to video search and, secondly, for real-time alerts.
For example, using an AI-based video surveillance solution it’s possible to capture metadata, such as the colour of the clothing someone is wearing or the backpack they are carrying – something that wouldn’t be possible using general data analytics. This means that if an eyewitness says they saw someone wearing a black coat and carrying a green backpack acting suspiciously it’s possible to input those details into the video search facility and it will find the relevant footage, potentially saving security personnel from watching hundreds of hours of video footage.
Eagle Eye Networks showed IFSEC Global an extremely advanced AI based video surveillance solution that
captures metadata of everyone entering its US offices as well as an ANPR (automatic number plate recognition) system that matches car number plates to the precise make, model and colour of each vehicle parked in its car park.
Nor is video search the only advantage that AI video surveillance offers over conventional analytics. It also makes it possible to create much more accurate real-time alerts. For example, Ipsotek, which has recently been acquired by French digital transformation company Atos, has developed a smart petrol station solution for the UK’s biggest grocery retailer. Instead of having staff manning its forecourts 24/7, it uses advanced CCTV cameras, coupled with AI-powered BullSequana edge servers, to detect certain events such as individuals trying to use the petrol pumps without a vehicle to fill up, or vehicles spending far too long at the petrol station which could indicate a significant site risk.
These events are automatically sent to a remote monitoring centre which is then able to intervene, often using speakers located on the forecourt to warn potential offenders that they are being recorded. Comments Chris Bishop, Sales Director APAC and Marketing Director at Ipsotek: “The smart petrol station solution, which has been approved by the fire authorities and by UKPIA (UK petroleum industry association), has allowed the retailer to move staff from its petrol stations to the superstores where they can help to stock shelves and refill back of house.” The solution has now been deployed at over 300 retail sites across the UK.
Whereas once video surveillance required large numbers of security staff sitting in a room examining video footage, the latest solutions – many of them AI-based – allow organisations to automate some of the processes, thereby allowing them to make considerable operational savings. For example, Ava Security, now part of Motorola, works closely with SMEs and educational establishments providing cloud-based video surveillance solutions that can be retrofitted to work with existing systems or fitted from new.
And although analytics is still widely used for basic object detection, such as counting how many students are in a particular area, increasingly AI is being used for real-time anomaly detection such as spotting if someone is in a place late at night where they shouldn’t be or a vehicle is travelling the wrong way down a road. However, AI isn’t a silver bullet, warns Sam Lancia, Ava Security’s Co-Founder and Head of Video Engineering: “Like analytics, AI isn’t perfect. There’s no Skynet here that’s magically going to detect that this person walking in is a bad guy and shoot them and lock down the school.”
Inevitably, while AI technology can help with video surveillance, it doesn’t provide all the answers. Nor does it always offer 100% accuracy, even with the most comprehensive of data sets and deep learning embedded over many years. For many organisations more important is getting the basics right first – installing cameras in the right places and making sure they are all working correctly – before they consider in investing in the latest artificial intelligence has to offer.
Source: IFSEC Global