An Improved Random Forest Algorithm-Based Fatigue Recognition With Multiphysical Feature

Fatigue driving is one of the crucial factors to cause traffic accidents. This article presents a novel fatigue recognition method with a multiphysical feature based on an improved random forest (RF) algorithm, which has significant merits in detection reliability and judgment accuracy. Key coordinate points of the face are obtained and then used to evaluate the eye aspect ratio (EAR), mouth aspect ratio (MAR), and head pitch angle (HPA) after preprocessing the frame-by-frame images in the video. Afterward, these evaluation metrics are used to further calculate the percentage of eyelid closure (PERCLOS) over pupil time, blink frequency (BF), yawn frequency (YF), and head nod frequency (HNF), of which the four parameters are used as input for the improved RF model. Different from the conventional RF model, our proposed model mainly optimizes the three parameters of the maximum number of features, the number of decision trees, and the maximum depth of decision trees in the scenarios of fatigue evaluation. The testing results show that the detection accuracy of our model using multiphysical features reaches up to 95%, and its reliability and robustness completely meet the standards of practical fatigue recognition.