Study On Driver Fatigue Detection Based On Eye Features

With the rapid economic development, the number of vehicles nationwideincreases rapidly, and traffic safety has become increasingly important. In thetraffic safety issue, the impacts of the factors such as drivers, vehicles and the roadenvironment are different, among which the driver accounts for the main role intraffic safety. In accidents mainly caused by drivers, driver fatigue is a major cause.Driver fatigue detection technology is therefore an important research direction inthe safety-critical monitoring and control technology. Detection of driver fatigue inthe video detection and analysis of the human eye characteristics has greatsignificance.This paper conducts research on adaptive driver fatigue monitoring systembased on the video analysis in a variety of lighting conditions. The contributions ofthe paper are as follows: we apply the technology of face detection based on skincolor model in the nonlinear piecewise color transform space; and Kalman filtertracking method to improve the speed of face detection. We study the clusteringproperties of the color space, skin color model and the characteristics of vulnerablenature of the impact of light intensity changes. On this basis, we design an adaptivelight intensity judgment. We achieve illumination correction using the method ofleast-squares fitting surface images collected in order to solve the driver fatiguedetection problem under the harsh glare light conditions. According to the weak,the strong three light conditions, the adaptive processing by the system to completethe rough location of the region of the human eye; On this basis, according to theregion growing, morphological operations and combined with prior knowledge offacial features to establish the precise positioning of the rule seteye position.Alsowe study the eye winking detection method, using PERCLOS as a standard to judgethe status of driver fatigue and fatigue-related parameters.We implement the system on TMS320C6000DSP. The design of the facedetection, tracking and expression analysis algorithms is ported to DSP and deepoptimization on the program is conducted, which handles20per second andachieves a preliminary real-time effect. The experiments show that the accuracy ofour system is good, where the correct detection rate is above85%. This generallymeets the basic fatigue testing requirement under a variety of light conditions.