SAFR exec on global opportunities and responsibilities of facial biometrics
SAFR has come a long way from its genesis as a project to build a facial recognition feature for a RealNetworks project roughly four years ago to watching over the Independence Day speech by India’s Prime Minister in Delhi this past August. The company’s CTO Reza Rassool had challenged a development team to build the capability in-house, but when the technology they built beat internal benchmarks far ahead of schedule, the company realized its future was not as a feature for something else, but as a product in its own right, with the potential to address a growing need for schools to ensure safety and security by screening individuals while easing access for those who entitled to it, explains RealNetworks Vice President and General Manager of Computer Vision Dan Grimm in an interview with Biometric Update.
Grimm was brought on a year later, after leaving Amazon, to help figure out how to build SAFR to commercial scale. As part of that process, he observed that heads of security at facilities ranging from airports and train stations, to public squares and casinos all have the same need to know right away if a person of interest is picked up by one of their security cameras.
“So many of these security operations centers they’re looking at a bank of screens, which represent a subset of the total number of cameras live on around a particular infrastructure,” Grimm points out. “There’s no way they can actively and effectively notice if a person of interest appears on one of those screens. SAFR solves that problem, and as far as we can tell, better than anyone else in the world.”
The results SAFR has achieved in the University of Massachusetts’ Labelled Faces in the Wild and NIST’s Face Recognition Vendor Test place its algorithm among the field’s elite, particularly for “wild” subjects, Grimm says.
“Our effective accuracy for live video is unsurpassed, based on the performance benchmarks we see, with respect to our accuracy on Wild Images on NIST, and our speed. Because our algorithms are so fast, it means we can actually compete multiple recognition calls before the competition calls once.”
Grimm says SAFR is enabling customers to do things with computer vision in places where they would previously not have been possible. Along the way, SAFR has also added people-counting, and analysis by characteristics, demographics, and sentiment, to allow customers to unlock further value from their existing camera deployments.
“What we’re finding is that they’re not only interested in doing facial recognition, but we’re delivering value for the head of operations and the head of marketing.”
Real-time facial recognition is somewhat controversial even in the most mundane deployments, however, and Grimm acknowledges a certain role for SAFR in the developing social dialogue around the technology’s use.
“As a pioneer in this space, we have a set of responsibilities in terms of how we introduce world-class facial recognition to the world,” he states.
Those responsibilities include a standard of excellence in building and training the system, according to Grimm, who says SAFR’s NIST scores show a very low variance in accuracy rates between white males and black females. While no algorithm is perfect, SAFR’s is among the top 5 in the world with the least “bias,” according to Grimm, and the company continues to work on this area. Another responsibility is to educate customers about how to effectively and appropriately deploy facial recognition. This includes the key variables such as camera position, the nature of the camera sensor, how the camera it is tuned, and how the system and software are tuned. The company also works only through trusted system integrators it has fully trained, he says.
Asked about the dependence of a technology vendor like RealNetworks on the responsibility of system integrators, Grimm acknowledges the need for trust, but suggests that trust is not a simple matter of faith.
“You have to place a certain level of trust in the partners you’re doing business with in any geography, and we’re doing business in lots of geographies in the U.S. and outside it,” Grimm says. “And we have an obligation to train them, and insist that they do things the right way, and pay attention to the training that we’re giving them, and also impress upon them their responsibility to work with the end customer and the end customer’s legal team, and their legal teams, to make sure the technology is being deployed in a thoughtful way.”
SAFR also tries to have visibility right through to end-users, as well, for several reasons, including understanding what works well and less well in deployments, and maintaining “a consistent and helpful feedback loop.”
The third responsibility companies pioneering the use of facial recognition have is to advocate on behalf of the industry, consumers, and even competitors with government to pass legislation that intelligently regulates the space.
Like a growing chorus within the industry, Grimm believes national regulation is badly needed. The first priority, he says, is to keep from losing the technology’s benefits to avoid risks through enacting outright bans, and so far SAFR has worked to be available as a resource to lawmakers to answer their questions and raise the level of understanding about the nature of deployments and how they work. While the company has engaged with policy-makers at the national level, it has focussed largely on the state level, including working with partners to support the 2019 Washington Privacy Act.
“We think the passage of smart laws such as that one at the state level can help set the groundwork for passage of thoughtful regulation at the national level,” Grimm explains. “That’s one strategy that we have taken and will continue to take.”
The dialogue is playing out not just the state and local level, but internationally, and Grimm says having access to RealNetworks’ resources and sales team has been one advantage beyond technology that has helped SAFR scale with customers around the world.
“You can think about SAFR as a hyper-growth startup inside of RealNetworks corporate infrastructure. It’s really the most important growth initiative for the company.”
SAFR recently launched its person detection, and the capability to associate a face with a body for continuous identification, and plans to launch improvements to that feature set, along with object detection capabilities in early 2020. Ultimately, Grimm expresses optimism about the prospects of balanced legislation and about SAFR’s growth prospects.
“If you have a critical need to run facial recognition on live video,” he says, “then SAFR is the clear choice.”