Intel FPGAs enable NEC’s face recognition technology
NEC applied Intel (which acquired the line with its purchase of Altera) Arria 10 FPGAs operating on Intel Xeon processor–based servers to increase the performance of its NEC NeoFace facial recognition engine to a level where an individual can be identified smoothly from a high-resolution image with dozens of faces.
“Facial recognition in a moving crowd requires highly advanced techniques when compared to still images because these cameras are affected by many factors: camera location, image quality and lighting, along with the subject’s size, walking speed and face direction,” said Tadashige Kadoi, general manager of IoT Platform Development Division, NEC Corporation. “Intel FPGAs and their parallel processing capability help NEC to enable fast and accurate collection and processing of images from even 4K high-resolution remote cameras.”
Intel FPGA acceleration technology played a role in a recent achievement for NEC. In March, NEC NeoFace was ranked No. 1 in almost all tests by the U.S. National Institute of Standards and Technology (NIST) [note 1] specifically for face-in-video evaluation.
The NIST tests evaluated the accuracy of the technology in two real-life test scenarios [note 2] including a test for entry-exit management at an airport passenger gate. It determined whether and how well the engine could recognize people as they walked through an area one at a time without stopping or looking at the camera. NEC’s face recognition technology won first place with a matching accuracy of 99.2%. The error rate of 0.8% is less than one-quarter of the second place error rate.
In the second test, the technology was asked to detect suspicious individuals at an indoor stadium. This test was conducted with an individual situated far from the camera, with their face direction changing frequently. NEC’s face recognition technology won first place with an error rate half that of the second place error rate.
To create the NeoFace Accelerator, NEC’s NeoFace facial recognition engine software IP is integrated into an Intel Arria 10 FPGA, keeping the same accuracy level while achieving higher performance in facial recognition than the previous solution.
Intel teams also worked with NEC to enhance the performance of NeoFace data centre server technology. NEC NeoFace Accelerator includes not only the Intel Arria 10 FPGA, but also an Intel MAX 10 low-cost FPGA and Intel Enpirion power devices.
Intel technology configuration used by NEC comprised; NEC-branded card with FPGA, and Intel Enpirion power; Intel Xeon processor Xeon E5-2697v4; Arria 10 FPGA (for the face detection engine); MAX 10 FPGA (for configuration, field upgrade, power monitor/control); Enpirion EM2130 (to power Arria 10 core); Enpirion EN2342 (to power the IO and transceivers).
Note 1; full NIST test results can be found here.
Note 2; Intel cautions; “Results shown from the Face in Video Evaluation (FIVE), the Multiple Biometric Grand Challenge (MBGC), the Multiple Biometric Evaluation (MBE) and the Face Recognition Vendor Test (FRVT) do not constitute endorsement of any particular product by the U.S. government.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
