Driver Fatigue Detection Based On Multi-algorithm Fusion

The rapid development of transportation，but it also brings a lot of negativeeffects, accidents caused by driver fatigue is increasingly becoming an importantreason. Including that more and more design about automobile meets human comfortrequirements, the driver fatigue is more likely to occur. In order to reduce theoccurrence of such incidents, it becomes a necessary mean to find an accurate andtimely fatigue detection algorithm. In this paper, based on driving comfort, detectionstate is detected by non-contact fatigue detection algorithm. Specific detectionalgorithm in terms of fatigue separately with the following components:Face detection, in the selection of the right face database, based on the colorfeature in the color space. Face detection algorithm based on color features isproposed. After the skin pretreatment, then give the pretreatment the integralprojection to separate the face detection region.Eye detection, feature extraction using adaptive edge detection algorithm tolocate the human eye. First, with Robert adaptive edge detection operator to extractface image, after seeking the gradient and computing complexity post, the position ofthe human eye can be basically determined. This paper presents a second featureextraction, and the complexity of calculating the gradient after processing again togive an eye positioning results. Experimental results show that this method cangreatly improve the detection rate of the human eye detection and can improve thedetection rate.Eye state judging, the common international PERCLOS state recognitionmethod is used to judge the human eye. By calculating the size of the eye area, a eyestate is judged with standard degree in comparative judgments about eyes open orclosed.Face detection algorithm, as well as eye fatigue detection algorithm todetermine the positioning algorithm is implemented in MATLAB image library.Experiments show that the improved algorithm for face detection and eye detection rate has significantly improved, and to some extent, it improved real-time, so thealgorithm used in this paper is feasible and has certain practical significance.