Design And Implementation Of A Real-time Multi-feature Fatigue Detection System Based On Android

With the rapid development of the social transportation, more and more long-distance transport drivers are busy at work. However, the driving accidents occur frequently due to the fatigue driving. Therefore, it is significant to research the fatigue driving detection. The detection method based on computer vision has become a hot topic because of no interference to the drivers. The existing methods have the problems of high computational complexity and time consume so that it is difficult to achieve the real-time detection on mobile devices.In this paper, we implement the real-time fatigue detection system based on Android system. Our system includes the mobile detection terminal and the remote network monitoring system. The face alignment algorithm is used to get the fatigue characteristics of the eyes, the mouth and the head. This paper mainly has the following research results:We improve the explicit shape regression face alignment algorithm and propose the novel face alignment algorithm based on the shape parameter regression algorithm using the second order regression framework. We regulate face shapes by low dimensional face shape parameters which are obtained by PCA. In the framework of the second regression, only the low dimensional parameters are calculated for replacing the original high dimension parameters. For different rotation angles and face scales, the explicit shape feature index algorithm is used to replace the local feature index, which can improve the comparability of features. We employ the multiple random feature selection method to select better features. Through improving the algorithm, the computation complexity and data storage capacity are greatly reduced, and the alignment speed and alignment effect are improved.During the implementation of the Android system, the face alignment is divided into two steps, the alignment of the human eyes and the alignment of the human face contours. In the first step, we just process the human eye areas with the size of 80*40 pixels. 20 feature points whose dimensions are 20 are used for aligning the eye shape. By this way, we reduce the image information and data dimensions for the single processing. In the second step, 27 feature points whose dimensions are also 20 are selected to represent the shape of the face contours. According to the results of the human eye alignment, we optimize the initial shape of the face contours. Thus, the face contour alignment is faster and more accurate.When analyzing the characteristics of the fatigue, according to the alignment of the human eye shapes and the coordinates of the feature points in the shape of the face contours, we use a simple four operations to judge the open and close of eyes, the open and close degrees of the mouth, and the up and down movement of the head. The coordinates of three feature points of the nose and cheek are selected to form the characteristic triangle. According to the change of the triangle shape, we judge the deflection directions of the head. The detection results are given by the overall fatigue features, which reduce errors detected by single fatigue characteristic.Finally, we implement our detection program on the general Android mobile phone and the tablet PC, achieving good detection results and real-time detection speed. Furthermore, we realize the network monitoring server system.