Quividi Tweaks Audience Measurement Software To Address A Very 2020 Problem – Face Masks

The French computer vision company Quividi says it has upgraded its patter detection software platform to deal with a very 2020 problem – counting and categorizing audiences for screens when the viewers are wearing face masks.

Quividi says it is about to release an upgraded version of its audience measurement platform that can count Digital OOH audiences even if they are wearing some sort of  face mask. This will ensure, the company says, that all audiences are correctly measured and that Quividi’s data remains high fidelity.

From PR:

In an effort to slow down the spread of Covid-19, an increasing number of countries have recommended or even made it compulsory to wear face masks in public or in crowded indoor spaces. As a result, the number of people covering their faces outside their home has drastically increased, bringing new challenges for DOOH and Digital Signage Anonymous Video Analytics solutions that rely on face detection and face analysis. So far, people wearing masks are inconsistently counted, if not ignored by these solutions.

“As lockdown restrictions are being eased, it is critical for our clients – malls, public transports or stores operators – to monitor in real-time the audience recovery,” says CEO Laetitia Lim. “In the Covid-19 context, this requires to measure the full extent of DOOH audiences, including the ones wearing masks. Being able to access high-fidelity audience data in real-time will enable our clients to drive a faster and more agile business restart.”

I had not thought about this, but it makes sense that biometric software that’s built around scanning and analyzing the geometry of human faces will have a bit of a problem if the lower half of faces are obscured by PPE or other kinds of masks or face coverings.

The AI/machine learning  algorithms used by face pattern detection companies are generally going to be less accurate when there’s less information available in the video stream from the camera/sensor.

I’m going to make the safe assumption that things like emotion (happy? sad?), that computer vision companies have added to base capabilities like age range and gender, is a real problem when there’s a mask obscuring the mouth.

I’m also going to assume the overall challenge masks present is not as great as for those companies that are on a different track with computer vision – doing facial recognition versus face pattern detection (this Wired piece talks about the challenge). The simple, and important, distinction is recognition means the technology is matching what it sees against a database of faces, to ID individuals.

Pattern detection means the technology – what is typically used for Digital OOH and digital signage applications – is just looking for faces, and anonymously counting and categorizing them. It doesn’t reference a database or archive the faces captured in video streams.

It’s a critically important distinction and there are STILL, even though this stuff has been around for something like 15 years, people describing these audience measurement systems as facial recognition – inevitably causing freak-outs over privacy.