Research on the Methods of Fatigue Driving Detection Based on Machine Vision
|Course||Control Theory and Control Engineering|
|Keywords||Driving Fatigue Skin Color Segmentation Moving Targets Detection TemplateMatching Area Complexity|
At present, the research on fatigue driving detection has drawn substantial attention. Domestic and foreign researchers have done a lot of work and got tremendous results on this field, some of which have been applied in practice. However, some drawbacks such as low correct rate, poor adaptability, weak real-time, and so on are exposed in usage. So some further researches are necessary.To remove those indistinguishable images that are caused by shadow or car lamps, this thesis binary the collected video images and calculate the size of target area in each frame. The underlying principle is that target areas in normal images hardly change and change dramatically in bad images.On the basis of particularity in fatigue driving detection which is mainly referred to the regularity of driver’s behaviors, the movement of driver’s face will be kept in some range. In this thesis, the method of skin color segmentation combined with moving target detection is used to find this range. Then edge detection and gray integration project are used to locate the driver’s face accurately in this range. Thus not only saves a lot of time, but also ruled out a number of interference. The driver’s face may be out of the bounds due to the change of driver’s sitting position with time goes by. In order to solve this problem, the method of CamShift is used to track driver’s face, which could adjust the range adaptively.Based on the above, this thesis uses prior knowledge to construct model for detecting driver’s eyes, meanwhile the method of calculating area complexity is used to correct the results of eyes detection, which not only shorten the computing time but also improve the precision of eyes detection. What’s more, the coordinate values gained by kalman filter are used to replace the values that fail to detect in serial images.Finally, in order to detect driver’s fatigue state, in view of that the sizes of non skin area in recognized eyes area are different when eyes are open or closed in binaryzation images, through counting the proportion of non skin area to whole recognized area on line, and combining the PERCLOS with the rule of people blink, the threshold to judge snoring can be calculated respectively. On the one hand, this method plays a good performance in complex environment. On the other hand, the adaptability in this method improved largely compared with a given threshold in advance. The simulation results also showed that the algorithm discussed in this thesis is of high practicability, good adaptability and fast speed.