Understanding Driving Distractions: A Multimodal Analysis on Distraction Characterization
summary
Dataset records 45 participants across 12 modalities including visual AUs and physiological signals (BVP, skin conductance, respiration, temperature). Four distractors are tested: texting (physical), N-back (neutral cognitive), emotional radio, and faulty GPS (affective cognitive). Models use AU features, BVP-derived temporal/spectral features, random forests with information gain and leave-one-subject-out and user-dependent schemes; physiological signals (especially BVP) outperform vision on cognitive distractions, multimodal fusion improves recall, and AU45 is strongly associated with texting.
AI-written summary · about
chapters (11)
- 0:00Introduction to research
- 1:15Research goals and dataset
- 2:08Experimental setup and distractors
- 3:25Sensors and modalities recorded
- 4:14Research questions and features
- 5:34Evaluation methods and results
- 6:28Modality performance comparison
- 7:40Multimodal results and benefits
- 8:29Action units and feature analysis
- 9:12Top features and model efficiency
- 10:23Main takeaways and conclusions
related talks
06mThe Empathetic Car: Exploring Emotion Inference via Driver Behaviour and Traffic Context
06mObjective Measures of Cognitive Load Using Deep Multi-Modal Learning — A Use-Case in Aviation
19mWhat Can Be Predicted from Six Seconds of Driver Glances?
29mAI & Neurobiology of Deception: How AI Helps Us Understand Ourselves
06mLearning When Agents Can Talk to Drivers using the INAGT Dataset and Multisensor Fusion
07mAssessing Cognitive Performance Using Physiological and Facial Features: Generalizing Across ...
17mWhat is multimodal design? An interview with John Alderman and Christine Park
06mExamining the Social Context of Alcohol Drinking in Young Adults with Smartphone Sensing