We are creating some awesome events for you. Kindly bear with us.

Automated surveillance monitoring for Singapore’s malls

The story of Vi Dimensions is the story of a successful business use case.

Finding Hidden Treasure

Three years ago, a group of likeminded individuals already working in the security industry came together. They were already working on video analytics systems. Yet, they noticed a discernible gap between what the customer demanded and what the market offered. Realising they were limited by a rule-based problem under an existing Israeli software they were using. For the software to work, a lot of input is needed to train the machine what to look out for. Eager to discover video analytics’ full strength, the team set out to prototype their own surveillance product.

Backed by a Spring Singapore’s Technology Enterprise Commercialisation Scheme, the team set sail. They were headed southward to the island resort of Singapore. There, the prototype proved successful and even commercially viable.

Now, that team has grown from strength to strength.

Mr Raymond Looi, Co-Founder and CEO of Vi Dimensions, shares with us the technology behind his winning product. ARVAS is the company’s patent pending product. An unsupervised machine learning system, the product is able to analyse huge quantities of real-time streamed videos to identify abnormal behaviours and events.

The gold for the company’s product is in abnormality detection which requires little training and intervention. Ultimately, it improves an organisation’s safety, operation and maintenance.

Mapping Shortcuts

As a recipient of the Call for Innovative Solutions (CFIS) jointly issued by the Ministry of Home Affairs (MHA) and the Infocomm Media Development Authority (IMDA), Vi Dimensions will receive substantial funding from the SGD 2.5 million allocated to the four CFIS projects.

In partnership with CapitaLand Retail Management Pte Ltd and Gold Ridge Pte Ltd, ARVAS will be adapted to meet new operational demands. This is achieved by system architecture enhancements and new deep learning inference engines to extend the technology’s abnormality detection capabilities. In the process, heavy reliance and dependence on security guards will be reduced.

At their booth, Mr Looi explains how the analytics software is able to pick up salient frames from a dizzying myriad of concurrent video streams. One glance and it is easy to understand how security breaches happen. A security officer inundated by the information overload can easily overlook abnormal situaitons. The pilot project they offer can filter pertinent information onto a single screen where only perhaps two to three videos are presented.

Retaining a control room’s existing hardware such as CPUs and cameras, the technology does not incur any additional significant cost. Mr Looi says that the mall doesn’t have to change anything. All they have to do is to provide their IP address and put their existing devices on the server.

This is particularly impressive given that many malls use older generations of cameras with poor resolutions. According to Mr Looi, the resolution the cameras provide are secondary because only recognition of abnormal activities is necessary. The adaptive learning technology behind the pilot does not require extensive programming nor does it need large data sets for learning. Learning is real-time and initial learning may only require a few days.

On his screen, Mr Looi shows how the user is able to identify user-specified scenarios, such as the presence of an unexpected vehicle. The machine quickly picks up the information and highlights abnormalities to the security officer.

Armed with the same capabilities, the pilot project can also reduce bandwidth utilisation in transmitting video data. Moreover, their product offering is cloud-based. Dynamic scaling is possible, and clients can access the analytics anywhere.

Plaza Singapura, Westgate and Nex, are three of the seven malls which will trial the technology. As a product with 90% accuracy, safety seems pretty certain.

Send this to a friend