Sensors Behind Device Screens, Expression Tracking, and a New SDK: Facial Recognition Roundup

January 18, 2019 by Kate Smith

A look at some facial recognition-related info from around the industry.

Facial recognition is now a common fixture in mobile designs. Here's a look at some facial recognition-related info from around the industry.

ams Launches Sensor for Facial Recognition from Behind Device OLED Screens

Last week, ams launched a new color (RGB light) and IR proximity sensor IC, the TCS3701. One of the noteworthy factors of this sensor IC is that enables mobile devices to process facial recognition from behind OLED screens. This allows more design options as engineers may have more flexibility as to where they could place these sensors in a device. It also means that a smartphone's screen could effectively become an array of sensors.

According to a press release from ams, "Despite the constraint of operating behind an emissive OLED display screen, the TCS3701 senses the addition of the ambient light passing through the display to light emitted by the display’s pixels located just above the sensor." The company credits a non-predictive asynchronous algorithm with the device's ability to perform ambient light, regardless of screen display brightness and across a range of dim-to-bright environment light levels.


Schematic for the TCS3701. Image from ams


According to Reuters, ams also supplies the optical sensors for 3D facial recognition for the iPhone, though ams does not publically claim Apple as an official customer. This may suggest that such technology could end up in future iPhones. 

This is only one avenue through which ams is working with facial recognition. In the first week of January, ams announced a new partnership with Face++, a Chinese open AI platform that allows various forms of environment recognition—including facial recognition—to be integrated into designs more quickly via an API (application programming interface) and SDK (software development kit). 

The TCS3701 joins ams's suite of 3D sensing products. There is an eval board for the IC, though a datasheet for the device, itself, is not yet publically available.


Selected specs:

  • 2.0mm x 2.5mm x 0.5mm OQFN package
  • 1024X dynamic range
  • 1.7 - 2.0 V supply voltage
  • 1.8V I²C bus 
  • <50 cm recommended operating distance

Omron's B5T-007001 Human Vision Components (HVC-P2)

Switching gears to modules, Omron's HVC-P2 provides the components for recognition systems. First introduced in 2016, it consists of a small camera and circuit board that has advanced recognition capabilities. Models are offered with cameras for either long distance or wide-angle views. The component can be installed using USB (microUSB type B) or UART connections and includes ten different image sensing functions. 


HVC-P2 sensor


It can, for example, detect and recognize faces, estimate age, gender, expression, and determine what users are looking at. 

The image sensing functions included are:

  1. Human body detection
  2. Hand detection
  3. Face detection
  4. Face direction estimation
  5. Age estimation
  6. Gender estimation
  7. Blink estimation
  8. Expression estimation
  9. Face recognition*
  10. Gaze estimation

*For face recognition, registered users are identified while non-registered users are noted.


Imaging sensor capabilities image from B5T-007001 whitepaper


This sensor is currently configured for marketing applications—it can be used to determine user engagement with an advertisement or vending machine. The variety of functions provided, however, used alone or together, makes it suitable for many applications.

In marketing studies, expression estimates can provide more useful feedback than just having subjects answer survey questions.

In energy applications, human body detection provides more efficient systems. Lighting and temperature can be controlled to provide comfortable environments when needed and save energy if no one is around. If a lunch room or gym is not in use, for example, the systems can be turned off. When being used, the systems are turned on and as more people gather, they can be automatically adjusted to maintain comfortable surroundings.

Engineers could certainly apply the technology to other fields. For example, the device might be installed in an automobile to adjust the driver’s seat and mirror for different drivers, recognized before they enter the vehicle. Alternatively, it might be used in industrial settings and manufacturing plants to gauge worker attention and alertness, perhaps setting off an alarm before a machine operator falls asleep, or notifying a supervisor that crane operator is playing games on a smartphone.

In recent years, the HVC-P2 has reportedly found use in monitoring digital signage, as well as monitoring shopping districts in China (site in Chinese).

Neurotechnology SentiVeillance 7.0 Face Verification SDK

Also at the beginning of the year was the announcement of a new software development kit for facial recognition from Lithuanian company, Neurotechnology. 

One of the major challenges of the widespread use of facial recognition technology is developing reliable algorithms to parse large datasets. Neurotechnology's SentiVeillance platform utilizes a biometric facial identification algorithm that they claim allows the analysis and real-time creation of watch-lists. Imagine, for example, a watch-list created to assess security in an industrial facility.


Image from Neurotechnology


The algorithm reportedly allows the tracking of individual faces over a camera's field of view, even as they pass behind objects. According to a press release, the company also claims that the algorithm "can perform gender classification, evaluate a person's age, identify facial expressions (e.g., smile, open mouth, closed eyes)" and even identify if there are other factors such as glasses or facial hair. 

As sensor technologies become more advanced in 2019, it's important for software to follow suit to enable better data processing and, ultimately, better functionality of the hardware.


What are your predictions for the facial recognition sector in 2019? Share your thoughts in the comments below. 


Featured image used courtesy of ams.