MACHINE LEARNING SYSTEMS FOR DETECTING DRIVER DROWSINESS

MACHINE LEARNING SYSTEMS FOR DETECTING  DRIVER DROWSINESS

June 29, 2009  

Drowsy  driver  detection  is  one  of  the  potential  applications  of  intelligent vehicle systems. Previous approaches to drowsiness  detection primarily make pre-assumptions  about  the  relevant  behavior,  focusing  on  blink  rate,  eye closure,  and  yawning.  Here  we  employ  machine   learning  to  datamine  actual human behavior during drowsiness episodes.  

Automatic classifiers for 30 facial actions from the facial action coding system  were   developed   using   machine   learning   on   a   separate   database   of  spontaneous  expressions.  These  facial  actions   include  blinking  and  yawn  motions, as well as a number of other facial movements. These measures were  passed to learning-based classifiers such as Adaboost and multinomial ridge  regression.  

Head motion information was collected through automatic eye tracking and an accelerometer. The system was able to predict sleep and crash episodes on a simulator with  98% accuracy across subjects. It is the highest prediction rate reported  to  date  for  detecting  drowsiness.  Moreover,  the  analysis  revealed new information about human facial behavior for drowsy drivers.  

Keywords   Driver fatigue - Drowsiness - Machine learning - Facial expressions  -  Facial  action  unit -  Head  movements  -  Multinomial  logistic  regression - Support vector machine (SVM) - Coupling - Driver behavior