Locomotive Driver’s Fatigue Detecting Based On Expression And Gestures

With the modernization of railway development, traditional transport systems will continue to face many new challenges. So locomotive driver’s normal driving occupy a position not to be ignored in the train’s safety. Traditional monitoring of fetigue driving technology is fetigue detection based on human eyes, based on head locations. These fetigue State detection existence some drawbacks. They are unable to overcome the fectors such as changes in light intensity. And other fetigue detection is based on EEG, EOG, ECG, and so on. EOG, ECG, EEG may able to detect locomotive driver’s state. But as we all know, EEG, EOG, ECG measurement needs contact with the human body which may affect the driver’s mood and against locomotive drivers of normal driving. In view of this, this article will Proposed a new study for focomotive driver’s fatigue detecting——locomotive driver’s fatigue detecting based on expression and gestures.After investigation, in the course of driving, train drivers need to make a series of prescriptive actions required by the provisions, and within a specified time interval to repeat these actions. Therefore, this article made a gesture-based condition monitoring technology for locomotive drivers’state monitoring. Simulation of locomotive driver’s gestures, the author get some video file similar to locomotive driver’s driving in a lab environment. The main algorithm of this article is shown in the following:Simulation of locomotive driver’s gesture in the process, we get the video files in a lab environment. After processing the video file, we can determine whether the gestures were normative or standard, as well as detection of locomotive driver’s driving status. The main algorithm of process is:firstly, locomotive driver video files must be preprocessed. By Gaussian mixture model, we can get background model and find moving targets (gestures). Background image is constantly updating and meantime we can judge the gesture occurs or not. When the gesture is detected the target appears, stop using Gaussian mixture model, and record the background image; And then use consecutive frames subtraction between five inter frame. Through compared between this difference and the set threshold, determining whether differences exist between neighboring five frames. If the difference is less than the threshold, the locomotive driver is doing the standard gesture at this time; and then through skin color model conversions, the more accurate gesture image is extracted from the image. At this time the resulting image may have some deficiency, such as a hole. Through corrosion and expansion method to make amendments, so as to be clear and precise gesture of binary image; lastly, the gesture of binary image target need to be recognition and judgment. Used shape context, HOG feature and AdaBoost classifier to comprehensive judgment. On one hand, HOG feature and AdaBoost classifier is used to classing and recognizing the Gestures. First need to train the classifier through AdaBoost classifier using gesture of HOG feature is training to be a strong category. Through the strong classifiers, the binary image can be classification. And we can determine whether the gesture is standard; On the other hand, using shape contexts algorithm to calculated gesture of binary images and templates to judging whether they are matched. A lot of experimental data suggests that through a series of algorithms of calculation, we can more accurate monitor the locomotive driver’s state.