Research On The Multi-view Real-time Fatigue Detection Method For Train Drivers

With the rapid development of railway transportation in China,the train drivers’ operation has changed from traditional train control mode to long time monitoring of train real-time running information.Long-time monotonous driving leads to a significant increase in the degree of fatigue and lethargy of the train drivers.In the cause of train accidents,the drivers’ fatigue driving is the largest proportion.Therefore,it is very important to reduce the train accidents caused by fatigue driving and improve the driver’s driving safety.At present,most of fatigue detection algorithms require the train driver to face up to the camera,which makes the train driver inconvenient and causes the limitation of the fatigue detection method.To make the front face-based fatigue detection unaffected by train driver’s head posture changes,a machine vision-based method is used to analyze the head posture changes of a train driver and the algorithms of inverse projection transformation and eye gaze correction are used to correct the feature deformation caused by the head posture change.The main research contents include:(1)In order to meet the requirements of real-time,the Adaboost-based face detection algorithm is used to locate the face quickly and improve the convergence speed of the key feature point location algorithm.The supervised descent method is used to quickly located the key feature points of the face,and the key feature points are used to solve the head posture angle and calculate the ratio between the height and width of the eyes and mouth as the fatigue feature according to the position of the key feature points.(2)The relationship between head posture changes and facial fatigue characteristics was studied,and the inverse projection transformation algorithm of projection fatigue characteristics and real fatigue characteristics is given and the relationship between eye gaze direction and eye feature parameters is established.Through the attitude of the head,the inverse projection correction factor and the eye gaze correction factor are obtained,and the directly measured fatigue characteristic parameters are corrected to reduce the influence of the head posture on the fatigue detection result.Experimental results show that under multi-view environment,the algorithms can improve the accuracy of fatigue feature extraction 31.1%.(3)The characteristics of PERCLOS faigue degree judgment criterion and fuzzy reasoning method are studied,and a fatigue detection algorithm based on the combination of PERCLOS method and fuzzy reasoning method is designed.In this paper,the modified fatigue characteristics are input into the fuzzy reasoning system,and the fatigue level of the output is calculated using the PERCLOS principle.The fatigue levels are calculated at three different fatigue levels.The fatigue rate,which has the highest percentage,is taken as the output of the system to determine the train driver’s fatigue state.Experimental results show that this algorithm overcomes the disadvantages of the two independent algorithms and improves the resolution and robustness of fatigue detection.