Research On Fatigue Driving Detection Technology Based On Multi-feature Fusion

In recent years,my country has seen a high incidence of automobile traffic accidents caused by fatigue driving,and the number has increased year by year,which has seriously threatened people’s life,health and property safety.Predicting the fatigue state of the driver and warning it can fundamentally control the occurrence of traffic accidents.Whether the driver is fatigued or not can be directly reflected in the visual characteristics,physiological indicators and vehicle behavior/due to the fatigue state of the car driver.Reflected in visual characteristics,physiological indicators and vehicle behavior,it has gradually become an important technical research hotspot in the academic field and the automotive industry at home and abroad.Aiming at the shortcomings of current automobile driver fatigue assessment methods that are timeconsuming,low accuracy,and high hardware requirements,this paper proposes a method for fatigue driving detection based on multi-feature fusion.The main research work is:1.Preprocess the video image.The video image is collected,transmitted,and stored.During this process,its quality will be reduced due to noise interference,and the driver’s facial area will be affected by different light conditions during driving.Therefore,this article preliminarily performs the following on the video image Two kinds of processing ensure accurate face detection:(1)Remove image noise based on dual-neighborhood median filtering;(2)Combine histogram equalization algorithm and logarithmic transformation to uniform image illumination.2.Real-time face detection.Considering the two indicators of speed and accuracy,this article selects HOG feature inspection measure the face in the video image,that is,perform sliding scanning on the face image according to windows of different sizes,and at the same time extract the FHOG feature of the block,and use the face classifier trained by SVM to determine whether the current block belongs to a face,and the end of the After scanning the entire image,there is a situation where the same face area is detected multiple times.This phenomenon is processed with non-maximum suppression,and then the final result of face detection is obtained.Through experiments,this method is robust to occlusion and face changes.According to actual application needs,this article only performs maximum face detection.3.High-precision facial feature point positioning.Based on the correct detection of the face,this paper is based on the cascaded residual regression tree marks 68 facial feature points,and the coordinate information of these points will be used to calculate the fatigue feature parameters.Experiments show that this method is convenient and has a high recognition rate,and can provide accurate facial feature point coordinates for subsequent head pose estimation.In order to eliminate the interference of other face information,this paper only locates the facial feature points uniquely for the largest face detected.4.Eye and mouth fatigue detection.Calculate the eyes from the 12 feature points of the human eye based on the PERCLOS criterion aspect ratio EAR,according to the EAR threshold to identify the eye open and closed state,and then achieve blink detection,and then count the number of blinks in a unit time,so as to find the blink frequency;for the recognition of the mouth fatigue state,first according to the mouth 10 Calculate the mouth height-to-width ratio MAR,find a reasonable MAR threshold for the initial judgment of yawning,and then combine the mouth opening duration to verify whether you are yawning again.Similarly,the number of yawns per unit time is counted to find the frequency of yawning.5.Calculation of head attitude angle.Head pose estimation is the process of mapping from a two-dimensional video image to a three-dimensional space.This paper selects 14 representative points from the 68 2D face key points located,and uses these points to match the 3D standard face Model,the conversion relationship between the 3D model points and the corresponding 2D image points is solved,then the head posture angle can be obtained from the rotation matrix,so as to realize the head posture estimation.6.Multi-feature fusion to judge fatigue state.Extract five fatigue features,such as the degree of eyelid opening,blinking frequency,mouth opening degree,yawn frequency,and head posture angle,from which seven feature parameters are obtained.The support vector machine using RBF kernel function is selected to fuse the above features to establish a fatigue detection model,and then use the self-made fatigue driving detection video A comparative experiment on the data set proves that the method proposed in this paper has a faster running speed and can be used for real-time detection;on the other hand,the experimental results show that the fatigue driving detection algorithm based on multi-feature fusion correctly recognizes the fatigue state.The ability is significantly higher than the traditional method based on single fatigue feature,and compared with the other two existing methods of fatigue driving detection accuracy,it further demonstrates the excellent performance of the proposed method.