Automatic Estimation of Self-Reported Pain by Trajectory Analysis in the Manifold of Fixed Rank Positive...

Area of Science:

• Computer Vision
• Biomedical Signal Processing
• Pain Assessment

Background:

• Accurate pain assessment is crucial in clinical settings.
• Objective pain measurement methods are needed to complement self-reporting.
• Facial expressions are key indicators of pain intensity.

Purpose of the Study:

• To develop an automated system for estimating self-reported pain intensity from facial movements in videos.
• To model facial landmark dynamics on Riemannian manifolds for pain estimation.
• To validate the proposed method against state-of-the-art techniques on public datasets.

Main Methods:

• Facial landmark extraction and decomposition into four regions.
• Representation of landmark trajectories using Gram matrices on a Riemannian manifold.
• Curve fitting for trajectory smoothing and temporal alignment for similarity computation.
• Support Vector Regression (SVR) for encoding trajectories into pain levels.
• Late fusion of regional estimations for final pain prediction.

Main Results:

• The proposed method accurately estimates pain intensity based on facial landmark dynamics.
• The approach demonstrates competitiveness against existing state-of-the-art methods.
• Validation on the UNBC-McMaster Shoulder Pain Archive and Biovid Heat Pain datasets confirms efficacy.

Conclusions:

• Automated facial landmark analysis offers a promising avenue for objective pain assessment.
• The Riemannian manifold approach effectively captures facial movement dynamics related to pain.
• The developed method provides a robust and competitive solution for video-based pain estimation.