Making security feeds smarter via machine learning

• Surveillance industry stuck with a rule-based system
• Vi Dimensions looking to scale its technology and its operations

 
IN today’s age of surveillance, with cameras even in our homes, security professionals need to monitor an ever-increasing number of screens. In 2014 alone, 245 million video surveillance cameras were installed globally, according to an IHS report.
 
It is increasingly clear that the ‘watchmen’ are having trouble watching. This is where Singapore-based Vi Dimensions hopes to help, by identifying anomalies to alert human operators that a second look is needed.
 
While the security industry is already familiar with rule-based analytics, there is still a gap, according to its cofounder and chief executive officer Raymond Looi.
 
“Video analytics has been in the market for the last 10 years, but the way they have been doing it is not very efficient, using rule-based analytics.
 
“This requires you to know what you are looking for and then specify a rule, when most do not know what to look out for,” he says, in a conversation with Digital News Asia (DNA) in Singapore recently.
 
“We felt that there was a gap in the industry in how they applied automation to cameras – and with the proliferation of surveillance cameras and vast amounts of data being generated, we felt that this was an area in which we could differentiate ourselves,” he adds.
 
Building the machine
 

Vi Dimensions is banking on machine learning technology to identify anomalies, but teaching machines how to see has proven challenging, according to Looi.
 
“We were initially working with researchers to write the algorithms, but this only worked in the lab, so we had to rewrite everything from scratch,” he says.
 
This was not helped by the fact that there is a shortage of IT talent who specialise in the niche area of computer vision.
 
“Finding people to work in IT is difficult; finding someone to work in a startup is even more challenging – and finding people with computer vision and machine learning abilities is an even bigger challenge,” Looi says.
 
“We had to source globally and go on all sorts of websites to get the right candidates,” he adds.
 
This idea of machine learning video analytics is also new, and challenging people’s preconceived notions was yet another challenge.
 
“Not many people have an idea of what we are trying to do – they have a preconceived notion from the old rule-based analytics,” says Looi.
 
“To try and change their mindsets to the new way of doing it is challenging, and it takes a lot of time to create awareness and let customers know how our product works,” he adds.
 
But all this paid off, with Vi Dimensions’ new algorithms producing good results, which was a breakthrough for the startup, according to Looi.
 
There are still limitations to what kind of cameras the video analytics startup can use, according to cofounder and chief technology officer Christopher Tay.
 
“There are some old cameras that have too low quality or frame rate which is below our minimum requirements,” he admits.
 
While the current system only works with a fixed camera, moving the camera does not throw off the machine, which can learn and adapt to the new picture, he adds.
 
Adoption and plans
 
Security is now on most governments’ priority lists, and Vi Dimensions has seen a lot of interest from the Singapore Government, Looi claims.
 
“Anyone who has a lot of CCTV (closed circuit television) camera, especially the LTA (Land Transport Authority) and MHA (Ministry of Home Affairs), has shown a lot of interest,” he says, adding that Vi Dimensions is currently on trials with a few customer sites.
 
The startup is also currently undergoing the accreditation process with the Infocomm Development Authority of Singapore (IDA), or the Accreditation@IDA programme, according to Looi, and is looking to scale in more ways than one.
 
“The next thing we are trying to do is scale it up to thousands of cameras – this is an engineering problem we are trying to solve,” he says.
 
“The other thing of course is to look beyond Singapore – we build a base here with a few good case studies, and then scale regionally,” he adds.
 
As for funding, Vi Dimensions raised S$1.5 million (US$1.1 million) in a pre-Series A round in January from ICT Fund I, a software-focused venture capital fund that qualifies for the National Research Foundation’s Early Stage Venture Fund II (ESVF II) 1:1 matching scheme. [Clarified]
 
“We will probably look for another round of funding at the end of 2017,” says Looi.
 
Related Stories:
 
ViSenze's mission is to make sense of the visual web
 
Embedding innovation in the tech ecosystem
 
Tech Vision: Innovating to zero crime
 
Singapore is using spyware, and its citizens can’t complain
 
 
For more technology news and the latest updates, follow us on Twitter, LinkedIn or Like us on Facebook.