InSight: Monitoring the state of the driver in low-light using smartphones
summary
Windshield-mounted smartphone captures near-infrared illumination using a retrofit 810nm NIR LED and repurposes standard RGB sensors to see the driver at 40–50 cm in low light. Labels transfer from an IR rig to NIR phone images via homography so facial landmarks and gaze models train without manual annotation. Evaluation on 15 drivers yields face detection ≈90%, landmark NRMSE <0.08, and event detection with F1 85% fatigue 93.8% distraction.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:36Visionbased system
- 0:48Challenges
- 2:02IR Filters
- 3:18Training Facial Models
- 4:11Homography Transformation
- 4:59Results
- 5:37Summary
related talks
46mDriver Assistant Solutions with
17mUnderstanding Face and Eye Visibility in Front-Facing Cameras of Smartphones used in the Wild
14mOddEyeCam: A Sensing Technique for Body-Centric Peephole Interaction using WFoV RGB and NFoV ...
26mSelf Driving cars with Opencv
09mAutomatic Frustration Detection Using Thermal Imaging
37mSeeing is Believing: Mobile Vision API Deep Dive
54mSurveillance Detection Scout – Your Lookout on Autopilot
41mComputer Vision for Remote AR