Driver's Abnormal Behavior Detection Based On Machine Vision

With the substantial increase in car ownership,traffic accidents are becoming more frequent,which seriously threatens the safety of people’s property and lives.The vast majority of traffic accidents are caused by abnormal driving behavior.In order to reduce the occurrence of traffic accidents,the research of safety assisted driving system has great value.The detection method based on machine vision has the advantages of noncontact and low interference.This topic studies the detection of driver’s abnormal behavior,including fatigue driving detection and telephone behavior detection.The human face displays important status information of a person,and the detection of multiple features such as eyes and ears becomes the key to determining whether the driver is in an abnormal behavior.In video detection,the principle of "big to small,gradually positioning" is adopted.The thesis first uses the Adaboost algorithm to determine the initial position of the driver’s face,then uses the KCF tracking algorithm to track the face,and then uses the SDM algorithm to align the facial feature points to obtain the specific positions of the eyebrows,eyelids,nose,mouth and so on.Finally,the eye area is extracted and sent to a CNN neural network for processing.The CNN neural network is trained on the "open and closed eyes" dataset to classify and recognize the opening and closing of eyes.The thesis uses the fusion method of PERCLOS fatigue parameters and blink frequency to determine the driver’s fatigue status.During the call detection,the SDM algorithm is used to locate the positions of the face organs,and the "three courts and five eyes" prior knowledge is used to locate the "calling action" area of interest.When the driver calls,his hand is close to the ear,there will be a large number of skin-like pixels in the area of interest.The skin color has a good clustering characteristic in the YCb Cr space,and a large number of skin-like pixels in the ear region can be detected,thereby realizing the driver.Calling behavior detection.The thesis detection algorithm is developed in MATLAB environment.The experimental verification results show that the driver’s abnormal behavior detection method proposed in the thesis has high practical value.The face localization time of each frame of video is only 10.95 ms,and the accuracy of detecting fatigue driving and calling is 92.5% and 93.5%,respectively.