Driver distraction behavior identification method based on bidirectional video stream

The invention provides a driver distraction behavior identification method based on a bidirectional video stream, overcomes the defect of contact type identification of driver distraction behaviors, and realizes identification of distraction behaviors such as yawning and calling of a driver, and the specific technical scheme comprises the following six parts. And region-of-interest selection: selecting the 1/3 position below the image to carry out driver head identification. And face feature point extraction: five accurate face feature points are obtained through an MTCNN algorithm. And head posture extraction: calculating a projection relation of the head by adopting a PnP algorithm through five human face feature points of the 2D human face to obtain a rotation matrix, and performing head posture estimation. And target action detection: carrying out call recognition by adopting a Mobilnet-SSD target detection algorithm. And lane departure early warning: carrying out edge detection by adopting a Canny operat