Speaker
Description
Recent developments in psychological assessment have seen the rise of AI-based
automatic detection of emotional facial expressions, now widely implemented in
both commercial and open-source software. Despite its growing prominence, this
AI-based approach encounters practical, measurement, and diagnostic
challenges.
Our initial study (N = 18) involved a comparative analysis of OpenFace, an
AI-based system, and blenderFace, a non-AI-based system for facial expression
assessment. To facilitate a fair comparison, participants were recorded using
both a UV-sensitive Webcam for blenderFace, marked with trackable sunscreen,
and a standard Webcam for OpenFace. This methodology addressed potential
biases introduced by facial markers in AI-based recognition. The blenderFace
method, employing an optical, pattern-based tracking system, avoids the
pitfalls of AI-driven methods by not relying on a point distribution model or
predefined emotion categories.
In a larger study (N = 106), we conducted statistical analyses of raw facial
movement data from blenderFace. This approach offered a nuanced understanding
of facial expressions, using three-dimensional coordinates for in-depth
analysis, particularly beneficial for high-precision research such as
microexpressions or the Component Process Model.
Overall, our research provides insights into the comparative effectiveness of
AI-based and direct measurement methods in facial expression analysis. While
AI-based systems mark a technological advance, they also introduce certain
limitations. Our findings suggest a balanced approach that combines AI's
efficiency with the accuracy of direct movement data to improve psychological
assessment practices.
| Are you currently an Early Career Researcher? | No |
|---|
