Pain Classification and Intensity Estimation Through the Analysis of Facial Action Units

Federica Paolì, Andrea D'Eusanio, Federico Cozzi, Sabrina Patania, Giuseppe Boccignone

Abstract: This study focuses on using facial expressions to evaluate acute pain levels. We analyse videos by relying on an extended set of 17 Action Units (AUs) and head pose components. Multiple models are trained and compared to detect the presence of pain and classify its intensity on a 5-point scale, ranging from no pain to high pain. Validation studies were conducted on two publicly available datasets, evaluating both in within- and cross-dataset conditions. The experimental results show better pain classification performance when using both the extended AU set, instead of the restricted AU set related to pain expressions, and head pose information.