Toward dynamic pain expressions in avatars: Perceived realism and pain level of different action unit orders

an article by Marie-Hélène Tessier,  Chloé Gingras and Philip L. Jackson (Université Laval, Québec, Canada; Centre Interdisciplinaire de Recherche en Réadaptation et Intégration Sociale (CIRRIS), Québec, Canada; CERVO Research Center, Québec, Canada) and Nicolas Robitaille (Centre Interdisciplinaire de Recherche en Réadaptation et Intégration Sociale (CIRRIS), Québec, Canada) published in Computers in Human Behavior Volume 96 (July 2019)

Highlights

• Fine facial movements can affect the perception of virtual facial expressions.
• Perceived realism and pain level change according to the order of pain AUs.
• Avatars' gender affects the perceived realism and pain level of the order of AUs.
• A specific combination of AU yields the highest ratings on realism and pain level.
• AUs order should be considered in the creation of virtual humanoid.

Abstract

The facial expression of pain can be decomposed in three sets of Action Units (AUs), the smallest discriminating facial movements: Brow lowering (B), Nose wrinkling + Upper lip raising (N), and Orbit tightening + Eyelid closure (O).

This study compared the perception of realism and pain level from different onset orders of AUs in avatars. Seven videos of facial expressions of pain were created with four different avatars (2 women): six sequential onsets combining the three sets of AUs and one synchronized onset. 45 healthy adults (22 women; aged 23.6 ± 5.2 years) rated the realism of facial movements, and the level of intensity and unpleasantness of perceived pain.

A more realistic expression was associated with the onset of O before or at the same time as N, a more intense expression was associated when B occurred last, and a higher level of unpleasantness was associated with the onset of N before B. Therefore, the sequence ONB yielded the highest ratings on both measures of realism and pain levels.

These findings describe the perceived content of different orders of facial movements that could contribute to the creation of realistic pain-expressing virtual agents designed to study human-computer interactions.

---