Arm® Cortex®-A processor enables embedded AI facial recognition

In recent years, due to the high accuracy of facial recognition solutions, facial recognition solutions have been used as an alternative to security products that were unlocked using traditional cards and passwords, such as attendance management, biometric authentication for ATM/mobile banking, parking lot and locker keys, etc. ” compatible products have been released.

Recently, we have been receiving requests from customers asking if it is possible to create a facial recognition solution using a processor equipped with the low power consumption and high performance Arm® Cortex®-A, which is often used in embedded products. We handle facial recognition solutions realized with Arm® Cortex®-A processors and can provide a comprehensive introduction.

Therefore, in this article, we will introduce the ``i.MX 8M Plus equipped with an Arm® Cortex®-A53 processor and NPU that operates at up to 2.3TOPS'' provided by NXP Semiconductors and the ``i.MX 8M Plus with a recognition accuracy of 99.83%'' provided by CyberLink. We would like to introduce the built-in AI facial recognition solution realized by FaceMe®, an AI facial recognition solution.

FaceMe® - Face Recognition Designed for AI/IoT Devices | CyberLink

i.MX 8M Plus – Arm® Cortex®-A53, Machine Learning, Vision, Multimedia and Industrial IoT

Embedded AI performance in NXP i.MX 8M Plus

Equipped with up to four Arm®Cortex®-A53s operating at 1.8GHz, it is an application processor that can meet various needs that require high performance processing with low power consumption.

It is equipped with an ISP that can process MIPI CSI input camera images, and can be connected to an image sensor without an external ISP.

Equipped with a Neural Processing Unit (NPU) running at up to 2.3 TOPS, it is an ideal processor for edge IoT, allowing you to implement AI applications even in environments without network connectivity.

We provide BSPs that are compatible with various OS such as Android, embedded Linux, Windows 10 IoT Enterprise, etc.

i.MX Application Processor - NXP Semiconductors - (macnica.co.jp)

CyberLink FaceMe® built-in AI facial recognition performance

CyberLink's FaceMe® is AI facial recognition software for embedded devices such as edge devices.

It is possible to add highly accurate facial recognition functionality to mobiles, embedded devices, IoT, servers, etc.

It is a software development kit (SDK) that can realize "face detection", "face recognition", "facial attribute analysis", "spoofing prevention", and "mask support" with high precision, and can be used on various platforms such as Windows, Android, Linux, and iOS. It can be implemented on the OS.

99.83% facial recognition accuracy ・Face recognition ・Singularity extraction ・Face detection ・Comparison and identification	Analysis of age, gender, and emotions ·sex ·age ・Emotional reasoning ・Direction of head
Anti-spoofing function ・Biometric authentication using 2D camera ・Biometric authentication using 3D camera	Facial recognition when wearing a mask ・Mask detection ・Recognition rate when wearing a mask is 98% or more

Introducing a demo using i.MX8M Plus EVK and FaceMe®

In the demo environment, the i.MX8M Plus EVK implements Android OS.

NXP provides pre-built images, so you can start testing the operation right away. We explain how to start it here.

[NXP MPU Beginner's Guide] I tried booting Android OS with i.MX8 development kit

By connecting a display to HDMI, you can operate it with a keyboard and mouse like a normal Android device. A USB camera is used.

The FaceMe demo provided by CyberLink is available as an Android app, and implementation is complete by simply installing the app.

The table below summarizes the inference speed (FPS) results for two patterns: "When AI inference is offloaded to NPU" and "When inference is performed only by CPU".

By turning on the NPU, the inference speed can be increased to about 10FPS, making it a practical solution for the walk-through facial recognition shown in the application example.

*Walkthrough facial recognition: Applications that require real-time authentication even when you move your face without stopping in front of the device.

Purchase an evaluation board at Macnica Mouser

Related information

Inquiry / Quotation

If you have any questions regarding this site, purchasing an evaluation board, or would like a quote, please contact us using the form below.