Facial recognition has seen huge breakthroughs since the U.S. National Institute of Standards and Technology (NIST) first began testing in 2010. Accuracy has seen massive gains, especially from 2013-2018.
In the 2018 test, the most accurate algorithm was 20 times more accurate than the 2013 equivalent. Essentially, 95 percent of the matches that failed in 2013 now yield correct results. Compare that to 2010-2013, when the most accurate algorithm reduced its error rate by 30 percent. This reduction in error rates since 2013 is due to wholesale replacement of the old algorithms with new ones based on deep convolutional neural networks — completely revolutionizing the technology.
Optimal recognition results
One entrant in the newly energized market is RealNetworks, whose SAFR for Security is an AI-based facial recognition solution for live video that integrates video management system (VMS) solutions. With 24/7 monitoring, SAFR detects and matches millions of faces accurately in real time, enabling teams to manage a watchlist across any number of video feeds.
SAFR says it delivers optimal recognition results with 99.86 percent accuracy in under 100 milliseconds, even in real-world conditions where faces are in motion, at different angles, under poor lighting, or partially obscured. SAFR builds on RealNetworks’ 23-year history in video technologies. Launched in July 2018, SAFR — secure, accurate facial recognition — is enabling new applications for security, convenience, and analytics.
Create security responses
SAFR targets facial recognition for live video, identifying camera-unaware faces moving in real-world conditions. In the April 2019 NIST results, SAFR tested as the fastest and most compact solution among algorithms with less than 0.022 False Non-Match Rate — 62 percent faster than the average speed, according to the company. SAFR now provides capabilities such as live video overlays alerting security professionals to events in real time, automatic bookmarks with rich metadata for investigative work, and alerts that can be customized to create security responses.
Facial recognition algorithms
Five years ago, facial recognition algorithms would struggle to match forward-facing people from still images, let alone camera-unaware moving faces from live video with variations in rotation and tilt. SAFR says they have achieved a balance of accuracy and performance for live video. A contributor to this accuracy is consistency across a range of skin tones. The algorithm was trained on a highly diverse global set of over 10 million non-simulated real-world faces.
SAFR was optimized for speed and can sample a face multiple times during the same period of time as other algorithms, subsequently increasing its accuracy. SAFR achieves the performance through edge processing. Distributed architecture enables efficient bandwidth consumption, reducing the roundtrip latency of facial recognition speed to under 100 milliseconds. The savings lower total cost of ownership (TCO): SAFR uses one-sixth the compute power of competing facial recognition solutions, equating to $500,000 or so in savings on a 250-camera deployment.
Integrated experience
SAFR also uses off-the-shelf hardware and is optimized to leverage inexpensive GPUs. SAFR can be deployed on premises or in the cloud, and supports Windows, Linux, macOS, iOS, and Android. When SAFR is paired with a VMS, such as Milestone XProtect, the integrated experience includes 24/7 monitoring to detect and match faces in real-time.
SAFR for Security provides live video overlays, notifications, and automatic bookmarks directly through the Milestone XProtect interface.
SAFR is designed with privacy in mind. All facial images and signatures are AES-256 encrypted in transit or at rest.
“We’re excited to partner with Milestone Systems to help security professionals achieve focused and prioritized security 24-hours-a-day,” said Dan Grimm, Vice President and General Manager of Computer Vision at RealNetworks. “We look forward to working with Milestone to help their customers achieve greater value with SAFR.”
Find more info about the SAFR integration with XProtect on the Milestone Markeplace.
Read the full article by Larry Anderson on SecurityInformed.com.