Study On Detection Of Driver Distraction Based On I Binocular Vision And Face Orientation

With the continuous increasing of car sales, cars become people ‘s main traffic tools.Howerver, traffic accidents occur more frequently than ever before. According to incompletestatistics, the distraction of drivers is the main cause of traffic accidents. Therefore, in order toreduce the traffic accidents caused by distraction, we need to develop a system to detect thedriver ‘s attention that can help to estimate the driver’s driving state. In addition, when thedrivers are out of attention,it protects drivers from accident by sending warning signals.This paper is based on binocular vision technology. We use Adaboost algorithm to detectface and facial organs, and then we use the CamShift algorithm combined with Kalman filterfor target tracking,3D reconstruction method is also used to calculate the face orientation, andthen we measure the dispersion degree of attention according to the line of sight eccentricity.The main contents are as follows:This paper use Adaboost algorithm to detect the face and facial organs. A lot of foregroundimages and background images are collected and made into samples. Then we use positive andnegative samples to train acascade classifier, and find the three central positions of eyes andmouth with classifier. So we should find the location of face,and search the whole area offace, then we can find the eyes and mouth position quickly and accurately.We use the combination of CamShift and Kalman filter method to track the detected target,and get the target position each frame.This method overcomes the difficulty to identify targetdeformation and partial occlusionBased on OpenCV, we can calibrate the camera to get the internal and external parametersof cameras. Then, according to the obtained feature points and the camera internal and externalparameters, we can calculate the head orientation. The paper proposed a method to calculateface orientation which use the intersection of the normal vector of the face and front scenemapping plane.The car scenario is defined as three elliptic area, we can replace the line of sighteccentricity with the weighted value of elliptic area, during the whole time of driving, wecalculate the levelvalue of distraction by the continuous increasing or decreasing of the distraction.Validated in this paper, the proposed method can detect and track the driver‘s eyes andmouth effectively, and it also can calculate the face orientation, so we can detect the situation ofthe distraction of the drivers accurately.