More Accurate and Secure Visual Recognition

Advantech FaceView® uses Intel® technologies to make AI visual recognition more accessible and valuable to businesses.

At a Glance:

  • AI visual recognition technologies are emerging as an economical way for businesses and governments to enable smart infrastructure, help control access to secure areas, generate more personalized retail experiences, and encourage health-conscious behaviors such as wearing masks.

  • Advantech FaceView, built with the CyberLink FaceMe® SDK, uses key Intel® technologies to deliver both the visual recognition accuracy and edge AI performance needed to drive emerging characteristic recognition use cases.

author-image

By

Advantech FaceView, Built with the CyberLink FaceMe® SDK, Uses Key Intel® Technologies to Make Visual Recognition More Accessible and Valuable to Businesses Everywhere

Advances in AI vision technology are making facial characteristic recognition more accessible and economical for organizations to deploy. New use cases are springing up that enable smart infrastructure, help control access to secure areas, generate more personalized retail experiences, and encourage health-conscious behaviors such as wearing masks during global health events. Quality-of-life improvements are emerging with growing enthusiasm and innovative uses for object recognition technology. Business end users and consumers who want to enjoy the benefits that AI vision technology brings to their work or life environments are volunteering or preregistering to participate. Now the focus is on developing quick, economical means to deploy these solutions while adhering to safety standards and expectations.

Use Cases for Facial Object Recognition Technology

Smart building

  • Contactless elevator
  • Tenant recognition
  • Special facilities access such as gym or garage

Retail

  • Self-service kiosk
  • Personalized omnichannel experiences
  • Mask detection, temperature sensor integration

Factory

  • Heavy equipment authorization and access
  • Employee clock-in and metrics
  • Secured area control

Transportation

  • Anonymized passenger counting
  • Contactless fare and ticketing
  • Social distancing analysis

Challenge: The Need for Accuracy and Performance

Traditional visual recognition deployments have required enormous processing power. Faces, in particular, are highly differentiated subject matter compared to inanimate objects such as vehicles or consumer goods. AI models and parsing need to be extremely accurate and nuanced in order to be useful. And with greater accuracy come greater performance requirements. In traditional configurations, cameras would generate dense visual data that would then flow upstream to cloud networks for processing. However, this required not only intensive computing performance in the cloud, but also a robust network infrastructure to handle the massive backhauling of data. These challenges have made visual recognition deployments aspirational at best—and beyond reach at worst—for most organizations, until recently.

“In addition to its high ranking on the National Institute of Standards and Technology (NIST) list for facial characteristic recognition technology, FaceView has achieved a 99.7 percent accuracy rate.1 Customers know the name and are aware of our potential value.”

—Alan Kao, product manager at Advantech

Solution: Edge AI Visual Compute with 99.7 Percent Accuracy1

The Advantech FaceView solution, enabled by the CyberLink FaceMe® software development kit (SDK) with Intel Atom® processors, Intel® Movidius™ Myriad™ VPUs, and optimized with the Intel® Distribution of OpenVINO™ toolkit, delivers both the visual recognition accuracy and edge AI performance needed to drive emerging characteristic recognition use cases. By offloading visual processing and AI object recognition to edge-level devices, the FaceView solution eliminates the need to backhaul data to the network for processing. This saves on infrastructure costs while enabling fast results and responsive experiences for end customers at the location of the edge devices. And organizations can still send data upstream to centralized clouds and server systems for additional processing or analytics.

Performance and capabilities aside, the combined Advantech and CyberLink solution delivers exceptional accuracy, which is essential for success. Alan Kao, product manager at Advantech, comments, “In addition to its high ranking on the National Institute of Standards and Technology (NIST) list for facial characteristic recognition technology, FaceView has achieved a 99.7 percent accuracy rate.1 Customers know the name and are aware of our potential value.”

How It Works

FaceView software is installed on an edge AI inference system enabled by the Intel Atom® processors and Intel® Movidius™ Myriad™ X VPU. Cameras and sensors at the site of data generation will pass data to the edge AI system, where the FaceView software performs AI inference for visual recognition. The result of this inference flows upstream over a 4G/LTE wireless connection to either an edge AI server with more processing power or a private cloud network for analysis and action. At this point, the edge or cloud server can trigger certain behaviors, such as granting access to a building or a piece of equipment or loading a customer’s retail profile for personalized experiences on-site.

Figure 1: The FaceView data workflow from cameras, to edge AI system, to edge server, or private cloud network.

Software Structure and AI Acceleration with CyberLink FaceMe®

FaceView software offers cross-platform OS support for Windows 10 and Linux systems, and users can expand system capabilities with a number of add-in APIs for visitor monitoring, webcam streaming, customer program enrollment, and more. But the real heart of the solution, its AI engine, is driven by the CyberLink FaceMe® SDK. Kao comments, “Advantech tested and validated FaceMe® for the FaceView solution. The FaceMe® SDK is fully integrated with Intel® Movidius™ VPUs and the Intel® Distribution of OpenVINO™ toolkit. It helps deliver strong performance and low CPU loading for characteristic recognition processes.” FaceMe® takes advantage of the Intel® Distribution of OpenVINO™ toolkit to optimize visual recognition workloads, which enables a 5.4x acceleration in AI vision performance2 and makes the total solution much more powerful and cost-effective as an edge AI device deployment.

Antispoofing Measures for More Secure Visual Recognition

The FaceMe® SDK also leverages the capabilities of Intel® RealSense™ technology to generate and process highly accurate points of depth measurements on visual objects. This enables the FaceView solution to not only be more accurate on recognizing familiar objects, but also enables it to detect spoofing attempts or false positives, as in the case of a subject trying to fool a camera or sensor with a flat image or video of a known object. According to measurements by CyberLink, the SDK has a 100 percent detection rate for spoofing attempts.2

Hardware Optimization with a Massive Speed Boost

Advantech chose a combination of the Intel Atom® processor E3950 and the Intel® Movidius™ Myriad™ X VPU for its edge AI system. Within this configuration, the Intel Atom® processor provides a just-right fit in terms of CPU processing power, while the Intel® Movidius™ Myriad™ X VPU is purpose-built for AI vision and object recognition workloads. According to Kao, “We wanted to reduce the deployment difficulty. The Intel Atom® processor E3950 provides just enough CPU power in floating-point arithmetic, while staying cost-effective in combination with the Intel® Movidius™ Myriad™ X VPU.”

With this setup, Advantech was able to demonstrate a 10x acceleration in visual recognition workloads for systems that used both an Intel Atom® processor and Intel® Movidius™ Myriad™ X VPU, compared to systems with just an Intel Atom® processor.3 And, the inclusion of the Intel® Movidius™ Myriad™ X VPU also delivers a 30 percent reduction in total CPU usage, resulting in a more-efficient and less-power-hungry system overall.3 For system integrators and end users, the total cost of an Advantech system also goes down, making it more cost-effective to deploy or scale up with multiple systems.

Multiple Entry Points and Ease of Deployment

Advantech offers multiple entry points for their FaceView solution, with different accompanying edge AI appliances that scale up visual processing power by adding more Intel® Movidius™ Myriad™ X VPUs to an internal proprietary module called the Advantech VEGA. For low-performance use cases, such as an interactive kiosk in a retail setting, system integrators can deploy a basic appliance with just one VPU. For more-advanced use cases, such as intelligent transportation, Advantech also offers a temperature-resistant edge appliance with up to four VPUs. And this is just a snapshot of potential offerings; the full Advantech portfolio is extensive and covers many use cases, with the option to customize solutions based on environmental or AI vision compute needs.

Simple User Experience to Help Engage End Customers and Employees

The key competitive advantage here is just how quickly a system integrator can select an appliance, connect it to edge AI cameras and back-end cloud networks, and start delivering visual recognition services for their customers. The FaceView user interface also makes it easy to manage an entire database of registered end customers or participants on an opt-in basis, while generating useful CRM-oriented data to help understand key behaviors. For example, in a retail setting, FaceView users can see when customers last visited a store location, add them to VIP groups to help prioritize service delivery, or even differentiate between customers and employees on the sales floor.

Flexible AI Vision for Multiple Scenarios

The Advantech FaceView solution is a versatile addition to any system integrator portfolio of offerings and helps ensure quick deployment, time to value, and ease of use for businesses. FaceView offers substantial performance and efficiency for visual recognition with its inclusion of the industry-recognized CyberLink FaceMe® SDK. Advantech designs its hardware configurations with the ideal combination of Intel Atom® processors with Intel® Movidius™ Myriad™ X VPUs to deliver power-efficient performance that’s optimized for vision AI workloads. The net result of these technologies and efforts is a simple yet powerful facial object recognition solution that businesses can now access.

Notices and Disclaimers

Intel is committed to respecting human rights and avoiding complicity in human rights abuses. See Intel’s Global Human Rights Principles. Intel® products and software are intended only to be used in applications that do not cause or contribute to a violation of an internationally recognized human right.

Intel does not control or audit third-party data. You should review this content, consult other sources, and confirm whether referenced data is accurate.

Performance varies by use, configuration, and other factors. Learn more at www.intel.com/performanceindex​.

No product or component can be absolutely secure.

Your costs and results may vary.

Intel® technologies may require enabled hardware, software, or service activation.

© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.

Product and Performance Information

1

Source: NIST March 2020 Wild FVRT 1:1 test result measured for the CyberLink FaceMe® SDK.

2

Source: Internal CyberLink performance data.

3

Source: Internal Advantech performance data. Primary configuration is built with an Intel Atom® processor E3950, vs. a system built with an Intel Atom® processor E3950 and an Advantech VEGA-320 module with an Intel® Movidius™ Myriad™ X VPU.