Research Of Highly Robust Driver Fatigue Detection

Traffic accidents not only cause serious economic loss , but also bring life safety danger. Compared to other causes of accident mortality,accident caused by fatigue driving mortality is higher. Therefore, the study of driver fatigue detection, and through the detection results to warn drivers are in state of fatigue to prevent the traffic accident, so to reduce casualties and property damage. The current fatigue detection has a low robustness.Therefore, this paper focuses on how to improve the robustness of driver fatigue detection.In this paper, we have studied and summarized the existing fatigue detection technology. Then, we mainly studied driver fatigue based on the visual fatigue detection method, and we analyzes the reason why the fatigue detection method based on vision is not robust, and proposes a kind of vision-based high-robust driving Fatigue detection method. The main features of the method are as follow : firstly, by detecting the face, to track the nose, and then get the head movement state; Secondly, combined with the nose position and head movement state to determine the location of the eyes and mouth, and then extract the eyes and mouth fatigue characteristics;finally, the eye fatigue characteristics, mouth fatigue characteristics and head movement characteristics were fused for the multi-characteristic fatigue detection.This paper is divided into six chapters to explain the above content.The first chapter introduces the significance of driver fatigue detection and the research status, then leads to the main research contents..At last, the organizational structure of this paper are given.In the second chapter, we focus on the existing fatigue testing methods in the key technology, and proposed the architecture of high robust fatigue detection system, and compared the difference between the architecture and the current fatigue detection architecture, and described the hardware structure and software flow of the system in detail.In the third chapter, we emphatically analyzed the design and implementation of face localization algorithm based on CNN and nose location algorithm based on CNN . This is the core work of the highly robust driver fatigue detection method proposed in this paper. The CNN model structure of face detection and the CNN model structure of nose detection in the most suitable subject is given, and then we compare them with Adaboost's face detection method and Adaboost's nose detection method in the obvious scene of illumination changed and the results are given.In the fourth chapter, we introduce the design and implementation of the method of locating the specific facial features and give the contrast of each part. Including the design and implementation of a nose tracking algorithm based on TLD and Kalman filtering, and comparing the tracking effect of the method with the nose tracking method based TLD. And the design and implementation of head movement based on the tracking. And the design and implementation of eyes and mouth localization algorithm based on experience and multi-condition, and compared with the method of eye and mouth positioning based on simple "three court five eyes".In the fifth chapter, we have analyzed the relationship between fatigue characteristics, and designed the method of fatigue judgment based on multi-feature fusion, then given the experimental results.In the sixth chapter, we have summarized the main work of the paper,and analyzed the shortcoming and further work.