Page 47 - EE Times Europe Magazine - June 2025

EE|Times EUROPE 47

AUTONOMOUS DRIVING | IN-CABIN MONITORING SYSTEMS
Face Value: AI That Knows When You’re

Too Tired to Drive

By Rebecca Pool
As vehicles edge closer to full autonomy, in-cabin monitoring is swerving from gaze

tracking to full emotion insight.

t U.K.-based Blueskeye AI,
a team of facial expression
analysts spends a lot of
A their time reviewing images
of faces, tagging tiny nuances in
muscle movement with meticulous
precision. It could be a curl of the lip
or a slightly raised eyebrow; either
way, the high-quality, annotated data
will be used to train facial recognition
and emotion AI models.
The Facial Action Coding System
(FACS) annotator—also known as
muscle action coder—isn’t a new
concept. Developed by psychologists
back in the 1970s, FACS is now a glob-
ally recognized tool to measure visual
facial changes caused by muscle
actions and thereby analyze the full
range of human expression. What is
new is the application of FACS anno-
tation in software development for
in-car monitoring systems designed
to monitor driver behavior and detect
signs of distraction, fatigue, and
intoxication. Smart Eye’s technology combines driver and cabin monitoring to track eye gaze, body key points,
Blueskeye AI is among the com- and other status data for all vehicle occupants. (Source: Smart Eye)
panies working on the technology.
“These micro-expressions have been studied for a long time, and very To establish a firm foundation for the AI to understand human
many experimental psychologists use [FACS] all the time,” said Michel behavior, images and audio from millions of people are used to train
Valstar, co-founder and chief scientific officer of Blueskeye AI. “But the company’s models to detect factors such as eye gaze direction,
20 years ago, we’d [think], ‘Oh, this is in the future.’ The advent of deep head posture estimation, facial point locations, and tone of voice. From
neural networks really changed things.” there, the models can be refined to recognize fatigue, depression, and
Indeed, the rapid advance of artificial intelligence twinned with apparent emotion.
increasingly stringent automotive safety targets has pushed the field Blueskeye AI’s software is based on a transformer network—a neural
of facial expression analysis well and truly into the public eye. The network that excels at processing sequential data—that Valstar and his
EU recently mandated that all newly registered vehicles be capable of colleagues modified to capture probability. “This allows us to accumu-
detecting driver drowsiness and distraction by 2026. At the same time, late and communicate the certainty of different parts of a network. If
the European New Car Assessment Programme (Euro NCAP) requires face tracking is fine but there are uncertainties about the head pose,
that new cars have a driver monitoring system that can track behavior then inferences won’t be made based on that. By embedding probabil-
to qualify for a five-star safety rating. ity on [network] components, we can propagate that throughout the
“We’re now building [AI] algorithms to predict, for example, how entire network.”
sleepy someone is, before they fall asleep at the wheel,” Valstar said. Valstar noted that the neural networks were developed many years ago
“Basically, every car in Europe is going to have a driver-facing camera, but added that his team has been “refining and refining and refining,”
and once that’s in, we can really help with these safety elements.” especially since the launch of Blueskeye AI and the start of product
development. Today, the models not only can detect whether someone is
BUILDING THE MODEL blinking and smiling, for example, but they can determine why.
University of Nottingham spinout Blueskeye AI was founded by Valstar “We use temporal dynamics … so our algorithms can infer the dif-
in 2019 to capitalize on his two decades’ worth of research into human ference between somebody blinking in the sunlight versus an altered
behavior understanding. And the intensities of facial muscle actions [slower] blink pattern because of fatigue,” Valstar said. “We can look
merely scratch the surface of the data that Blueskeye AI uses to at the acceleration of a smile; is its trajectory smooth or stilted? It can
develop its software. really infer the difference between a fake or a real smile.

www.eetimes.eu | JUNE 2025