Research On Fatigue Detection Based On Analysis Of Facial Image

With continuous advancement of the transportation industry, human society is facingincreasingly serious traffic safety problem. The survey showed that one of killer of traffic accidents isdriver fatigue. If we can develop a real-time effective driver fatigue detection system which can alarmwhen detecting fatigue occurs, then we can reduce a lot of accidents, accordingly reduce casualtiesand economic losses. So it has very important practical significance and economic value.In this paper, we make in-depth research on fatigue detection method based on facial imageanalysis from the point of machine vision based on analysis of existing fatigue detection methods.Simultaneously, we use the PERCLOS principle which is currently considered most effective toconduct driver fatigue status analysis. The main contents of the present paper include:1. We use AdaBoost algorithm for face detection. After the face detection, we adopt enhancedpictorial structures method to achieve eye localization.2. Reseach on eyes closeness detection using appearance based methods.we present an extensivecomparison on several state of art appearance-based eye closeness detection methods, with emphasizeon the role played by each crucial component, including eye alignment, feature extraction, andclassification. Three conclusions are highlighted:1) fusing multiple cues significantly improves theperformance of the detection system;2) the AdaBoost classifier with difference of intensity of pixelsis a good candidate scheme in practice due to its high efficiency and good performance;3) eyealignment is important and influences the detection accuracy greatly.3. Research on eye closeness detection in a multi-scale and noise case. Taking into account theinterference of noise, we propose a feature extraction method based on covariance matrix, and theexperiment results on two databases vertify that the method has a better anti-interference of Gaussiannoise. In addition, for the test samples multi-scale case, we propose multi-scale fusion technologybased HOG. Experiments on two databases at the same time prove a big role played by the method forperformance improvement.4. We make fatigue state analysis based PERCLOS principle, calculating eyes closed frames oftotal number of frames in60s as a percentage, then determine appropriate threshold to the fatigue statejudge. Many of video clips are used on driver fatigue system to do simulation experiments. The resultsprove that this method has good practicality.