What Can Be Predicted from Six Seconds of Driver Glances?
summary
Uses a 100-car naturalistic study with ~5,000 six-second epochs (30 Hz) of hand-annotated macro-glances to train supervised gaze classifiers (CNN) and model temporal dynamics with an HMM + CNN pipeline. Evaluates 27 binary tasks and achieves ~80% for glance-related behaviors (radio tuning, distraction), ~60% for some environment signals (proximity to intersection, day/night, urban vs rural, age groups), and poor accuracy for gender, weather, and speeding. Results inform multi-sensor driver-state fusion.
AI-written summary · about
chapters (5)
- 0:00Introduction to research
- 3:02Glance classification approach
- 7:34Predicting external environment
- 12:15Why this matters
- 13:23Q&A session
related talks
06mLearning When Agents Can Talk to Drivers using the INAGT Dataset and Multisensor Fusion
38mWhere are we looking? Predicting human gaze using deep networks.
06mThe Empathetic Car: Exploring Emotion Inference via Driver Behaviour and Traffic Context
10mHAGI: Head-Assisted Gaze Imputation for Mobile Eye Trackers
10mVEmotion: Using Driving Context for Indirect Emotion Prediction in Real-Time
28mMachine Learning Inside the Car
07mAssessing Cognitive Performance Using Physiological and Facial Features: Generalizing Across ...
17mLearning Cooperative Personalized Policies from Gaze Data