Automatic Frustration Detection Using Thermal Imaging
summary
Paper presents an implemented ROS pipeline that collects synchronized RGB, thermal and EDA data from 18 participants, extracts landmarks with OpenFace, logs events and classifies frustration with a k-nearest neighbor model. Model reaches 89% accuracy on RGB, 87% on thermal, 84% on EDA and 86% when combined. Thermal signal needs ~7 s windows vs RGB ~3.5 s, and best thermal facial ROIs are nose, forehead and lower lip while cheek is discarded by the selector.
AI-written summary · about
related talks
15mC-Face: Continuously reconstructing facial expressions by deep learning contours of the face ...
15mRecognizing Emotional Expression in Game Streams
06mInSight: Monitoring the state of the driver in low-light using smartphones
57mFoundations of Facial Recognition
07mFingerTrak: Continuous 3D Hand Pose Tracking by Deep Learning Hand Silhouettes Captured by ...
13mEnhancing Object Detection & Avoidance Systems for Robots in Challenging Environments Using AI
21mLeveraging Python for Real-Time Image processing
31mFace Recognition