Dangerous Driving Behavior Detection Based on Multi-source Information Fusion

Abstract

Dangerous driving does great harm to traffic safety. A dangerous driving detection model was proposed based on multi-source information fusion in this paper. The information fusion model includes face fatigue features and distracted driving features, which are related to dangerous driving. Firstly, according to the characteristics of face fatigue, OpenCV and Dlib library are used to detect the blink rate, yawning behavior and head posture offset. The characteristics of face fatigue are comprehensively determined by calculating EAR (Eye Aspect Ratio), MAR (Mouth Aspect Ratio) and head posture offset. Then, distracted driving behavior includes making phone call, drinking water and looking at mobile phone, which are detected by the algorithm of improved YOLO-V5. Finally, the features of fatigue and distraction are fused, which is based on D-SET algorithm. The innovation of the paper is to fuse the features of different detection objects to improve the model detection effect. Compared with detecting two single indicators respectively, the accuracy of the algorithms fusion model for dangerous driving detection is improved by 4.8%, and the system response time is also faster by 0.53 seconds. Ultimately, the experimental results show that the model fusion algorithm effectively alleviates the problem of inaccurate detection caused by single information, and improves the accuracy of detection and the robustness of the system.