Security_Camera

For someone who, if he were 12, would have Edward Snowden posters on his bedroom wall, it pains me to utter the following phrase – the world has never needed surveillance more than ever. Shoot-outs and bombings at even the seemingly safest of places has impelled our society down this route.

Now, before you come at me with your blood-stained pitchfork and stones, let me qualify my statement: I am not championing for an Orwellian society where phones, emails and chats are all tracked and archived by the big G-man; I am talking about an on-the-ground surveillance that can help put a muzzle on criminal activities before they even happen.

Yes, we are still a long way from using three clairvoyants submerged in liquid à la Minority Report to point out potential offenders, but the advent of advanced machine learning algorithms and Big Data have accelerated the development of A.I-enabled video surveillance systems.

Computers are accelerating to the point where they can detect and track behavioural anomalies faster and more accurately than human beings.

Now, Singapore-based video survelliance company Vi Dimensions has laid claim to such an accomplishment.

Using software to sniff out the bad guys

First off, using software to enhance video survelliance capabilities is not new. Video analytics software helps reduce the reliance of humans for detecting criminal activity. This solves a critical pain point because there are more CCTVs than there are security staff at most command centres – making it easier for deviant behaviour to slip past.

But there are constraints in the existing framework.

“The current approach in these existing video analytics software is that the user has to first specify rules for detecting various already-known behaviours or incidents, for example, perimeter intrusions or loitering. In other words, existing video analytics software needs to be told what to look out for,” says Raymond Looi, CEO and Co-Founder of Vi Dimensions, in an interview with e27.

Such a system, while more efficient than manual detection, is still not the optimal solution – human beings can be unpredictable as not every evil-doer would exhibit the same pattern of nefarious behaviour. And what if the camera is surveying a large group of people? How would it be able to detect the lone wolf in the crowd when there is so much activity going on?

Also Read: 5 tips for hiring your company’s first data scientist

“There are almost limitless security-related scenarios and it is not possible to know beforehand, thereby ineffective, to try and define a precise rule to detect a specific event, such as a terror attack, for example” says Looi.

Throw out the rules

So how does Vi Dimensions tackle this pain point? Constantly having to implement new rules is time-consuming and a strain on resources – also, they imply a reactive approach because it requires the help criminologists to dissect the anatomy of a crime that has already happened.

The solution – discard the rules, remove all conventional human notions of criminal behaviour, and leverage on Big Video Data Analytics to construct Unsupervised Machine Learning system.

That, is in a nutshell what Vi Dimensions’s ARVAS is about.

“The system automatically surfaces or discovers unusual or abnormal behaviours without having to first apply any rules, thereby removing the restrictions on prior human knowledge. It enables the machine to learn and tell us what is deviant from a scene instead,” says Looi.

This enables the system to surface up almost limitless variations of suspicious behaviour. And inside, the brain that run everything is the Abnormality Detection Algorithm.

“Our Abnormality Detection Algorithm is based on a unique and novel approach. We have adapted it for surveillance videos where multiple motion patterns are occurring simultaneously.” he says.

“It can effectively infer their various patterns and starting times. Since the system is autonomous, it provides the means to automatically analyse hours of video easily,” he adds.

In addition to detecting criminal patterns, the ARVAS claims to be able to “discover events which have safety implications”, “alert rare and inconsistent occurrences that could signify an unauthorised activity taking place,” and, check this – “surface inconsistencies and defects occurring in the video streams.”

Looks like Keanu Reeves in Speed would have lucked out if ARVAS was installed in the bus.

Also Read: Behavioural analytics security firm Fortscale pulls US$16M in pre-B funding

What if there is an error?

Obviously, no system is without its flaws and hiccups from time to time. I posed this question to Looi about the implications of raising a false alarm, and whether ARVAS can ensure that its detection is accurate.

“Before we ask whether we are detecting ‘suspicious criminal activity accurately’, the question is whether we are even surfacing enough suspicious activities from the thousands of CCTV video streams for scrutiny. Vi Dimensions’ technology fills this gap by trying to discover as many unusual and suspicious activities as possible without having the CCTV operators manually go through each camera. The CCTV operator then reviews and checks if this ‘suspicious activity’ surfaced is indeed a potential security threat they have to act upon,” Looi explains.

So, to set your minds at ease — in case memories of that scene in Minority Report where Tom Cruise undergoes an eye transplant to escape the authorities has you feeling all paranoid — ARVAS does not make the call to escalate the situation to the law enforcement agencies by itself. At the end of the day, it still boils down to a human decision.

The idea is really to help “law enforcement agencies adopt a more pro-active approach rather than a reactive one where the incident had already happened and they are only using CCTV footages for forensics and post-event investigations,” says Looi.

Also Read: What’s the big deal with Big Data in Israel and China?

Moving forward

With technology developed, Vi Dimensions is currently being evaluated and undergoing trials with the Ministry of Home Affairs and other government agencies. It will also explore partnerships with other security companies.

“We definitely need partnerships as we do not provide the entire surveillance solution for example, the infrastructure and security cameras,” says Looi.

It will also scale up the system for enterprise deployments to systems and command centres streaming thousands of CCTV cameras in the months to follow.

“We are also selecting key customers for trials not only in Singapore but also in other countries where there are acute security concerns,” he says.

In terms of funding, Vi Dimensions recently closed a US$1.5 million round from ICT Fund I, which is part of the Early Stage Venture Fund (ESVF) by the National Research Foundation (NRF).

Looi says it will be used to accelerate VI Dimensions’s machine learning technology for discovering abnormal behaviour and events.