In order to develop and test the system we need some data to work on

What do we need:

• driver video (audio, head, face, eyes, hands)
• road video from driver perspective
• sensor data (GPS location, speed, acceleration)
• some tool to explore and view the collected data and processed results

Data quality and diversity (be aware of bias)

• Driving situations
• Light and weather conditions
• Driver types (sex, ethnicity, facial hair, clothing)
• all data needs to be time synchronized

Collecting data

Collect you own data:

• use an Android phone with TrackAddict and OBD2 to collect video and acceleration data
  • low cost
  • sometimes unreliable and drains battery fast
  • cannot catch driver face
• use a commercially available dashcam set that includes at least two camera's, GPS and acceleration sensor.
  • cost 150- 450 euro
  • front/rear/driver facing
  • low light / near infrared
  • reliable
  • gets all data

Note:

• Not all sensors are placed in the vehicle center of mass, how will that influence their measurements
• All sensors have drift and bias
• All sensors have to be calibrated
• Sensor fusion can help (Kalman, Madgwick or Mahony filter)
• GPS location is great but adds privacy issues

Driving data datasets

Use existing data.

TODO: Internet research find datasets assess their contents, quality and usage agreements

• DMD - Driving Monitoring Dataset
• The UAH-DriveSet
• The Warwick-JLR Driver Monitoring Dataset

Sensors

Learning:

• Matlab tutorials: Understanding Sensor Fusion and Tracking