Research On Fatigue State Detection Of Train Drivers Based On Facial Image Feature Fusion

As one of the factors affecting the safe operation of the railroad,the state of the train driver has always been of great concern.With the continuous development of China’s railroad technology,the possibility of train driver fatigue driving is becoming more and more obvious in the face of the increasing workload and the double challenge of complex working conditions environment.How to accurately detect the fatigue of train drivers in real time is of great significance to the safe operation of trains.In order to cope with the problem of light variation of train driver’s facial images under complex working conditions,the light compensation method is used to process the collected train driver images.For the problem of switching the train driver’s vision back and forth between the track surface and the instrument panel during operation,the facial feature points are corrected by single-strain correction.Fuzzy inference is used to fuse the facial fatigue feature quantities to realize the fatigue detection of train drivers.The main research contents include:(1)Adaptive illumination compensation for the captured images of train driver images.The luminance calculation is performed on the captured driver images,and the compensation algorithm applicable to this environment is selected according to the driver image luminance to improve the detection inaccuracy caused by illumination and provide the basis for subsequent face detection and feature point localization.(2)Multi-feature point localization of faces for train driver fatigue detection.It is proposed to use MTCNN network combined with ERT algorithm to achieve face detection and feature point localization,improve the minimum detectable size of MTCNN network to speed up the detection speed,calculate the eye opening and closing degree and mouth opening and closing degree;establish the head pose model,and calculate the head pose angle by combining the head pose model and facial feature points.(3)Face feature point correction for train driver fatigue detection.In view of the fact that the detected eye opening and closing degree increases or decreases when the driver lowers or raises his head,but does not actually change,the single-strain transformation is used to correct the face feature points,so that the correction result matches the actual one.The adaptive threshold selection of the human eye opening and closing degree using k-means++makes each driver have his own eye opening and closing degree to improve the robustness of the detection method.(4)Fatigue level determination of train drivers.Using fuzzy inference as a tool,the fuzzy quantity of fatigue is quantified,fuzzy rules are formulated,and the driver’s fatigue level is used as the input of fuzzy inference with eye opening and closing degree,mouth opening and closing degree,and head posture angle,and the driver’s fatigue level is used as the output to achieve the driver’s fatigue determination.The real-time performance of the algorithm is about 17 fps,and the accuracy rate is 98.5% under normal environment and 86.8% under low light environment,which can be applied to practical application scenarios.