Newer facial recognition systems are scalable and provide more accurate results, and the end-user selects the parameters of system performance to suit their own application needs.
“We have developed algorithms that can be optimised in various compute environments, whether in a small chip inside an edge device or on a large server device,” says Ido Amidi, Oosto’s Vice President of Products and Business Development. “It all depends on client needs. We can do it all without giving up performance. There is no significant loss in performance across various platforms.” The system’s “containerised environment” operates efficiently in a customer’s IT architecture.
Facial recognition implementation
Oosto, formerly AnyVision, was founded in 2016 based on deep learning research into facial recognition. The focus has been on how to use face recognition to identify people in real-world scenarios. Products began coming to market in 2017, targeting verticals that involve a need to identify “bad actors” or very important people (VIPs).
The implementation of the technology involves the software operating in an on-premise appliance or server
The most common implementation of the technology currently involves the software operating in an on-premise appliance or server, with inputs coming from cameras at the edge and outputs going to a network video recorder or access control system. As concepts of the Internet of Things (IoT) trickle down into the security marketplace, edge-based deployments will become more common, supplying real-time actionable information, and avoiding a flood of unstructured video data.
Oosto helps users deploy, set up, and calibrate the system; then, it is managed by the customer based on their needs. Privacy features are built into the system, part of the company’s commitment to “ethical” facial recognition.
The technology, in general, has changed a lot since early implementations a decade or so ago failed to perform as promised. Expanded capabilities are fueled by developments in deep learning.
Facial recognition can help end-users identify people of interest but is specifically designed not to violate anyone’s privacy.
There are no databases or watch lists involved; the user typically compiles their selection of “bad actors” against which facial recognition algorithms can be compared. While Oosto provides a valuable tool, the end-user customer decides how to deploy that tool in their business, says Amidi.
Oosto's “liveness detection” deploys algorithms to analyse images from video and/or 3-D cameras
Recently, Oosto has adapted the approach to provide access control; too, including “liveness detection” that deploys algorithms to analyse images from video and/or 3-D cameras to ensure a presented face is not a printed image or mask.
Their latest software release is a single platform aimed at recognising and providing insights into how people behave in physical spaces. New functionality includes the ability to detect unknown individuals entering restricted areas.
Neural net algorithm
Improved facial recognition functions well even for those wearing masks. A new neural net algorithm improves accuracy when identifying those wearing masks and the system can alert if a person is not wearing a mask (to aid compliance).
Expanded video forensics features, designed to expedite investigations, include an ability to do in-depth searches on captured video related to body attributes, colour of clothing, etc.
How well can Oosto recognise a face among 20 different people on a casino floor? It all depends on variables, such as the number of people, lighting, and angle of view, image quality, and even networking issues. Recognition algorithms are “trained” using data collected from real-world use cases; training takes place in the lab environment.
The system can adapt to various lighting conditions, such as in a glass-enclosed lobby in the morning versus in the afternoon. As people enter an area, they may be looking down at their smartphones, for instance, which would mean some of their faces would not be in view. Such variables impact accuracy.
Test requirements of a system apply to each specific customer and can be adjusted and tweaked as needed
Each customer must decide their parameters regarding how to best use the system, with the fewest acceptable number of false alerts, in real-time, and using existing infrastructure.
Test requirements of a system apply to each specific customer and the system’s sensitivity and other operational factors can be adjusted and tweaked as needed, including at any time after a system has been installed.
Meeting specified application
“No customer is alike, so the needs of each customer’s system are unique,” says Amidi. For example, some situations might have a higher tolerance for false alarms, such as when there is an operator available who can make the final decision on whether a face matches. In other situations, such as when there is a missing child, false alerts are a bigger problem, and rapid response is especially of the essence.
“What we give customers out of the box applies to any scenario, and customers can tweak the system with the click of a button to meet a specific application,” says Amidi. “We empower our customers to understand the pros and cons so they can react in real-time.”
Oosto can help to enhance the customer experience by identifying important customers
Oosto sells through systems integrators and has a partnership program with more than 150 integrators worldwide certified to manage and install systems. Oosto targets Fortune 500-size companies in financial services, gaming, and retail. The system helps to create a safe environment that is devoid of any “bad actors.”
Alternatively, Oosto can help to enhance the customer experience by identifying important customers that warrant special treatment. The system also can provide an alert if any unauthorised person enters a restricted zone; that is, anyone who is not an employee, a registered visitor, or a contractor.
In many parts of the world, facial recognition is widely accepted and used for applications such as payment and access control. Anxiety about technology, especially in the United States, is based on a lack of understanding.
The public needs to “be better educated” on the subject, and Amidi expects the technology to become more socially accepted over time. “It’s a tool, and it needs to be supervised,” says Amidi. “From a technology perspective, we trust the accuracy. It just needs to be better explained.”