WPG Group launches new facial recognition solution

WPG Holdings, a leading semiconductor component distributor dedicated to the Asia-Pacific market, announced that its subsidiary Youshang has launched a facial recognition solution based on SOPHON BM1880 and ON Semiconductor AR0130 & AR0230.

With the development and progress of society and the gradual popularization of artificial intelligence, people have gradually become accustomed to the convenience that AI brings to daily life. For example, work attendance, community access control, office building access control, high-speed rail station ticket gates, etc. do not need to carry keys or identification cards, and can be quickly passed by scanning the face.

Dalian Dayoushang launched a binocular face recognition solution based on SOPHON's computing chip BM1880 and ON Semiconductor's AR0130 & AR0230. Its BM1880 supports up to 1 TOPS computing power and local storage of 30,000 face profiles; AR0230 Sensor supports internal HDR synthesis output, which can effectively solve the overexposure and darkening caused by backlight and strong light; AR0130 has good performance in near-infrared. The application scenarios supported by this solution include: community access control, face recognition attendance system, and gate gate; the AI ​​functions supported by this solution include: human attribute and posture analysis, human detection, recognition and expression analysis, object measurement and recognition, license plate recognition, and voiceprint recognition.

Figure 1- A display board of the facial recognition solution launched by Dalian Dayoushang based on SOPHON and ON Semiconductor

Core technology advantages

【SOPHON  BM1880 】

BM1880 is a deep learning inference artificial intelligence chip focused on edge applications. BM1880 TPU can provide 1 TOPS computing power under 8-bit integer operations; under Winograd convolution acceleration operations, it can provide up to 2 TOPs computing power.

The specially designed TPU scheduling engine can effectively provide extremely high-bandwidth data flow to all tensor processor cores. The chip contains 2MB of memory, which can provide optimal programming flexibility for performance optimization and data reuse.

Provides users with a powerful deep learning model compiler and software SDK development kit. Mainstream deep learning frameworks such as Caffe, Pytorch and Tensorflow can be easily ported to the BM1880 platform, supporting common CNN/RNN/DNN

Neural network model analysis, compilation and tuning compatible with Linaro 96boards

Supports ResNet50, Yolo V2, Google NetV1, Mobile Netv1 / v2, SSD300, Alexnet, VGG16 and other models

SOPHON BM1880 is a multi-functional acceleration platform designed for artificial intelligence applications. It can be used as a coprocessor for deep learning reasoning acceleration, or as a main processor to receive video streams, pictures or other data from the Ethernet interface or USB interface, and then perform reasoning and other computer vision tasks; other hosts can also send video streams or pictures to BM1880, which will do reasoning and return the results to the host.

【ON AR0230】

Excellent low-light performance

ON Semiconductor's latest 3.0um pixel, DR-Pix™ technology with dual conversion gain

Linear or high dynamic capture

Pixel or line scan interleaved T1/T2 output

Support on-chip HDR internal synthesis output

Data interface: Four-channel serial high-speed pixel interface ( HiSPi ) differential signal (SLVS and HiVCM) or parallel output

【ON AR0130】

Excellent low-light performance in both VGA and HD modes

Excellent near infrared performance

Figure 2-Block diagram of the facial recognition solution launched by Dalian Dayoushang based on SOPHON and ON Semiconductor

Solution Specifications

【Application Processor Subsystem】

Dual-core ARMA53@1.5GHz

Neon acceleration, integrated floating point unit (FPU)

RISC V processor subsystem

Single core RISC V @ 1.0GHz

Tensor Processor Subsystem

512 MACs, 1 TOPS@INT8

Winograd convolution operation, up to 2 TOPS@INT8

【Video subsystem】

H.264 Video Decoder

MJPEG Encoder/Decoder

Video Post-Processor

1080P@60FPS or dual-channel 1080P@30FPS

【Physical specifications】

Typical power consumption is 2.5 watts

0.9V core voltage, 1.8V / 3.3V input and output

14mm x 14mm FCBA package

Audio subsystem

Dual I2S input and output

Using software to implement multi-protocol voice encoding and decoding

【Peripheral interface】

One USB 3.0 / USB2.0 host device

Two integrated GMACs supporting RGMII and RMII

5 x I2C / 16x PWM / 4x UART / Multi GPIO

【Memory interface】

DDR3 / LPDDR3 / DDR4 LPDDR4

Supports DDR capacities from 1 Gb to 4 GB

SPI NAND / Emmc / SDIO interface

eMMC capacity can support up to 32 GB

【ON AR0230】

Excellent low-light performance

ON Semiconductor's latest 3.0um pixel, DR-Pix™ technology with dual conversion gain

Full HD supports up to 1080P 60 frames per second superior video performance

Linear or high dynamic capture

Optional Adaptive Local Tone Mapping (ALTM)

Pixel or line scan interleaved T1/T2 output

Support external mechanical shutter

On-chip phase-locked loop oscillator

Integrated position-based color and lens shading correction

Slave mode for precise frame rate control

Stereo / 3D camera support

Data Engine

Data interface: Four-channel serial high-speed pixel interface (HiSPi) differential signal (SLVS and HiVCM) or parallel output

Automatic black level calibration

High-speed configurable A and B mode switching

Temperature Sensor

【ON AR0130】

Excellent low-light performance, both in VGA mode and HD mode

Excellent near infrared performance

HD video (720p60)

In-disc AE and statistics engine

Automatic black level calibration

A, B mode switch

line-by-line scan

Support 2:1 zoom

Internal master clock generated by an on-chip phase-locked loop (PLL) oscillator.

Parallel Output