Video Analysis In The Application Of Staff-State Recognition In The Driving Key Position

The intermediate railway station is the railway transportation network unit cell, the station attendant makes the train running safely in each interval by arranging the train route.Security issues of the operation at the station related to the train running safety, is the key section of railway train department. In order to solve the problems such as lacking safety consciousness of some station attendant, dozing off or sleeping in the working gap and unauthorized undergo, the management department of station installed many camera to monitor the personnel state on working site. through artificial examining and analysis of videos find all kinds of illegal state and remind the worker to pay attention to. Due to the limited manpower and the too much camera points, to achieve a high quality of each lens is not realistic, it is difficult to achieve better results.With the development of image and video recognition technology, it is feasible to apply it to the automatic monitoring and control of the working. This paper discusses the method of issuing a warning immediately when staff dozing off, sleeping or unauthorized undergo by deploying the camera in a plurality of directions and using image processing technology.Staff undergo judgment is judged according to the future size of the area, if the foreground area is less than the threshold value, it is determined that the staff undergo. sleeping is judged by detecting the eyes closed with sleep posture. when the pupil is undetectable characters, eyes closed is judged. When foreground pixel differences fall below a specified threshold, we determine target person still or does not move. At last, combining with the recognition of human posture is consistent with several pre-defined position, we can judge whether or not the person to sleep. In algorithm design, for some of the problems encountered in the process,some improved methods are proposed.Firstly, aiming at the lacking accuracy problem of practical foreground extraction process due to the effects of light may lead to extract the foreground figure appears hollow,this paper proposes an improved method based on Canny operator that searches initial foreground contour, morphological closing operation after selecting the contours of the maximum area and polygonal approximation and filling. It obtains the ideal body contour finally.Secondly, for the problem that the general face detection method is very difficult to detect rotating face, this paper presents the method of the face rotation correction before detection. By using “YCr Cb color ellipse model”to find the face region, then making "and " operation with the contour of the human body to filter out background interferenceobjects, then by ellipse fitting to calculate the angle of rotation, executing anti rotation correction of the face, finally we use Haar cascade classifier for face detection. Experiments show that this method improves the accuracy of face detection.Thirdly, in order to locate pupil, this paper presents the gray gradient contribution matrix model. When the face region is detected, the eye region is calculated according to the distribution of facial features in the proportion of the eye, then the region image gray gradient contribution matrix is calculated, The conclusion is that matrix maximum element coordinate is the position of the pupil center. After experimental verification, the method can accurately find the pupil.Finally, in order to solve the interference recognition problem caused by the instability and diversity of human gesture contour, improve the accuracy figures for posture recognition.In this paper, the design of the statistical features based on the contour of the human body is made to describe posture. These characteristics include distribution area, compactness and convexity. After calculating the characteristic, by support vector machine(SVM) and BP neural network,we classify and compare it with a predetermined several pose features to recognize the possible human condition. Experiments show that the this method has a good robustness.Combine with the proposed method, we program with OpenCV. After the experiment,the system can detect most phenomenon about sleeping and undergoing,judge whether the person is in normal working condition. We verify the results of this test method, basically in line with the experimental expectations.