Research On Dangerous Driving Behavior Early Warning System

With the rapid development of the transportation industry in recent years,the number of traffic accidents has also increased.According to statistics,there are numerous causes of motor vehicle traffic accidents,with dangerous driving accounting for a significant proportion,posing a significant threat to traffic safety management.It is an effective measure for reasonably preventing traffic accidents using modern scientific and technological means to reasonably judge whether drivers have dangerous driving behaviors and detect and warn them in real time,and it is also an important topic of current road traffic safety management research.However,due to individual differences among drivers and the complex light changes,shading,and interference in the actual outdoor driving environment,there are still many technical bottlenecks in the research of driving fatigue detection technology with high real-time,high accuracy,and strong robustness.The research in this dissertation focuses on key technologies in the warning system for dangerous driving behavior.The deep learning algorithm is used to recognize dangerous driving behavior,and its image recognition effect is far superior to previous related technologies.The YOLOv5 s model is used to detect the face,and the improved YOLOv3 model is used to determine the state of fatigue in the eyes.The driver’s head posture is then judged based on the information of the key points of the face,and whether the driver is fatigued or not is judged based on the corresponding fatigue judgment methods of the two models.Finally,the improved YOLOv5 s model is used to determine whether the driver engages in the risky behavior of answering and making phone calls while driving.The following are the main points of this dissertation:(1)The following changes to the standard YOLOv3 are made: Reduce one YOLO layer(a detection branch)because the scale of the object to be detected is relatively uniform and does not require a large number of feature scales;Increase the feature scale because detecting small objects is easier with a larger feature scale.The model trained by this improved YOLOv3 algorithm is faster and more accurate.(2)The eyes’ and mouth’s fatigue judgment method is improved.The Open CV+Dlib method is used to detect the key points of the human face.To calculate and judge fatigue,the detector will detect the position coordinates of 68 key points on the human face and extract the coordinates of key points on the eyes and mouth.(3)The improved YOLOv5 s model is used to detect whether the driver makes phone calls while driving.The enhancement is the addition of an attention mechanism module to the YOLOv5 network.With the addition of the attention mechanism module,the model can be more accurate in feature extraction,improving precision.