Research On Driver Fatigue Detection Based On Eye State Recognition And Its Implement

As the society and economy continuing to develop, the national consumption capacity has gradually increased, and the number of cars is also increasing rapidly. However, China’s annual rate of traffic accidents remains high, which the proportion is as high as40%or more in major accidents due to traffic accidents caused by driver fatigue. Driver fatigue has become one of the most important factors threatening traffic safety, and the study of driver fatigue detection has important practical significance.Currently, the fatigue detection method based on computer vision is a hot research. Base-d on extensive research on domestic and related technologies, a non-contact driver fatigue de-tection system based on the state of the human eye is designed. A camera is placed in the car for real-time monitoring driver status and prompts when detecting the fatigue of drivers and the system have achieved good results either in the laboratory environment or in the actual cab environment. Firstly, the face detection technology is studied and in order to achieve a good face detection results in the complex background, we use the cascade Adaboost classifier based on Haar-like features, which meets the requirements and has a good detecting results in the room and in outdoor environments. Then the rough region of the human eye is got base on facial geometry, which is effectively reducing the detection time and improve the accuracy of positioning. In this paper, two location methods are studied. One of the method is based on LBP template matching. The template of LBP feature of the human eye samples is build. Eyes are located by matching the LBP histogram statistical characteristics of the detection area. The other method locates eyes by quickly locating the approximate point of interest through the detection zone to make fast radial symmetry transform. In the comparative analysis of the advantages and disadvantages of both, the paper proposes an improved method. In order to recognize the state of the human eye, this paper studied two states recognized methods based on the sparse representation and PCA principal component analysis. PERCLOS algorithms, which is internationally recognized most relevant with fatigue, and the eye closed time is used to decide whether the driver is fatigue. At last, our system have reached the expected results in indoor and outdoor driving videos. The average processing speed is46ms which meets real-time requirements.