Research On Real-time Fatigue Detection And Alarm System Based On Information Fusion

In recent years,the high demand for logistics efficiency forces the driver to deliver the goods within a limited time,causing the driver to drive for a long time.Therefore,the inevitable fatigue of the human race has become the main driver of accidents.However,the existing methods have high computational complexity,long time consuming and high difficulty in hardware environment.This paper designs and implements a multi-information fusion real-time face fatigue detection and pre-warning system.The system takes advantage of Android camera for driver's face video.Through improved CFAN,the coarse positioning of face key point can be realized quickly.Then,the analysis of human eyes and mouth is performed by means of deep convolution network and morphological method.Finally,fatigue status was analyzed based on the eyes and mouth condition.On the basis of this algorithm,this paper builds a platform for real-time fatigue detection and pre-warning system including Android and PC.The main work of this paper is as follows:(1)Research on the location of face key points.The characteristics of cascade regression model and depth model are compared and analyzed,and the CFAN algorithm in the depth model is improved to speed up the localization of key points.(2)The face key state discriminant.Based on the geometrical distribution of the "three-chamber five eyes" of human face organs,a more detailed segmentation is carried out for the detection area of human eyes and mouth.Then we first by the methods of morphology of human eyes and mouth to get basic elliptic contour of the human eye and mouth,and then using least square method to ellipse fitting ellipse length ratio of the shaft,thus to describe the fatigue of the key points of the face.However,through the experiment,it was less accurate and less robust when the external environment interfered.Based on the situation,this paper classifies and validates the state of human eyes using fusion convolution neural network and morphological methods.(3)Design and implementation of real-time fatigue detection and pre-warning system.In this paper,software for the server and android front end is developed,through the data exchange between the android car terminal and the server side,the function module of the display end and server side of the vehicle terminal and center control center are realized.