A Visual-Based Driver Distraction Recognition and Detection Using Random Forest

Abstract

Driver distraction and fatigue are considered the main cause of most car accidents today. This paper compares the performance of Random Forest and a number of other well-known classifiers for driver distraction detection and recognition problems. A non-intrusive system, which consists of hardware components for capturing the driver’s driving sessions on a car simulator, using infrared and Kinect cameras, combined with a software component for monitoring some visual behaviors that reflect a driver’s level of distraction, was used in this work.

In this system, five visual cues were calculated: arm position, eye closure, eye gaze, facial expressions, and orientation. These cues were then fed into a classifier, such as AdaBoost, Hidden Markov Models, Random Forest, Support Vector Machine, Conditional Random Field, or Neural Network, in order to detect and recognize the type of distraction. The use of various cues resulted in a more robust and accurate detection and classification of distraction, than using only one. The system was tested with various sequences recorded from different users. Experimental results were very promising, and show the superiority of the Random Forest classifier compared to the other classifiers.