Research On Driver's Fatigue Status Detection Technology And Realization In Engineering

It is well known that driver fatigue is one of main causes of the traffic accident. Thus,the real-time detection of the driver fatigue has the vital practical significance in reducing the accidents caused by driver fatigue.Some beneficial improvement have been achieved in this field by employing the machine vision technology to analysis the facial image,then extracting many effective driver fatigue characters from these informations,and finally using the fuzzy logic and the artificial neural networks technology to detect and provide the early warning about the state of fatigue.The main task and contribution of this paper lie in:(1) The fatigue and the process that causes the driving fatigue are analyzed.Then based on the person-vehicle-environment systems,the driver fatigue analysis system has been established.Furthermore,the model of the driver behavior is also proposed based on the S-O-R theory.With these preparations,the person-focused model is established to analyze the driver fatigue behavior.(2) The mechanism of driver fatigue is studied from a psychological,physiological and behavioral science point of view.The actions employed by the driver to control the vehicles are named driver behavior.And the thought and feeling of the driver during the driving is called the driver psychology.It is proved that the time distribution of the traffic accident caused by the driver fatigue is just the same as the level of the sleeping or the wake of human beings.(3) We mainly study the influence of the following factors such as the driving time, speeding,driving environment,the background of the driving,social environment, health,road conditions etc.,to the driving fatigue.It is proved that the influence of these factors to the driver fatigue is greatly augmented as the driving time increases.However, such an influence can be resumed if the driving is stopped.(4) A new algorithm is proposed to locate the eyes,which is based on the projection and the intersected complexity.Specifically,a coarse-to-fine two level localization strategy is first adopted to approximately locate the area of the eyes based on the projection of the face image.Then,this area is divided into several small regions and some of them with high complexity are extracted.Finally,the regions that do not correspond to eyes are excluded by using the judging rules,and the rest is just in relation to the eyes.(5) Based on the color and texture compound characterization,the double state eye's tracking algorithm is proposed.The main idea of this algorithm lies in adding an additional determination criterion to the weight which denotes the open degree of the human eyes.More specifically,if the probability of the biggest granule is smaller than a given threshold P_min as the granule updates in the epicycle.We take P_min=0.65 on trial and achieved good tracking effect.(6) Based on the Gabor wave filterings,an algorithm is proposed to extract the texture of eyes.The obtained characteristics is imported into RBF neural network to learn and its output is classified into five types as 0,1,2,3,4 which correspond to the value of the parameters PERCLOS and AECS 0%,25%,50%,75%,100%,respectively.(7) Many of face detection and identification methods are studied.Firstly we introduce many face detection and face recognition method.Then the Hopfield neural network is adoptted to carry out the facial recognition.The experimental results illustrate that such a recognition method has higher identification accuracy in small sample set than BP neural networks.(8) An experimental platform is estabilished to real time detect the fatigue of the driver,in which the DM642 is the main processor.By employing the fuzzy neural network technique,the parameters that denote the degree of the driver fatigue are fused and then a new algorithm is presented to detect the driver fatigue.The main contributions of this paper are as follows:(1) A new algorithme,which is based on the projection and the intersected complexity,is proposed to locate the eyes.(2) A double state eye's tracking algorithm which is based on the color and texture compound characterization is proposed.(3) An algorithm based on the Gabor wave filterings is proposed to extract the texture of eyes,by which the PERCLOS and ECS can be obtained through the output of the RBF neural network.(4) Fuzzy neural network is introduced to detect the driver fatigue.The computer vision technology is first used to collect the changes of the driver's expression,blinks, movement of the eyes,and the line of sight etc.And then the four fatigue characteristic parameters named PERCLOS,AECS,NodFreq,and YawnFreq are extracted from these image informations.Finally,these obtained datas are fused by the fuzzy neural network and the output is just corresponded to the classification of the fatigue.Compared these value with the PVT,the degree of the driver fatigue is decided and then the corresponding warning degree is confirmed.The results of the experiment have indicated that our method is more efficient in real-time fatigue detection.Furthermore, the proposed algorithm is also valuable for the other related fields.