Research On Invalid Face Filtering In Video Attendance

With the development of face recognition technology,video attendance technology based on dynamic face recognition has also achieved good results.Video attendance is more and more popular because of its non-contact and non-mandatory characteristics.This article has improved the performance of video attendance from many aspects based on the characteristics of practical problems.The three main contents of video attendance are image acquisition,face detection and face recognition.In this article,a motion detection method is designed to ensure the effectiveness of image acquisition.Face detection module needs to ensure accuracy and recall.Face recognition module must ensure accuracy.The development of deep learning solves one difficulty after another in the field of computer vision.The research and improvement of the convolution neural network has broken through various bottlenecks of traditional machine learning,so that it can accomplish the task of face detection and face recognition in video attendance.By comparison,MTCNN(Multi-task Cascaded Convolutional Networks)and FaceNet are selected in this paper.A lot of factors can make the image blurred such as Human motion,deviating from the focal length.The fuzzy face image will lead to the reduction of the correct rate of face recognition and the reduction of the reliability of the video attendance.Using the feature of having sufficient face image in video attendance,the method filtering fuzzy face image is used to improve the accuracy of face recognition.The fuzzy face image filtering method based on "extended two step degree" and the method based on CNN are designed.Another factor affecting the accuracy of face recognition is the excessive deflection face.When there are enough face images,the accuracy of face recognition can be significantly improved by retaining the frontal face.In this article,we separately use MTCNN output feature points and CNN features to design the face image filtering method.Through the analysis,a multi-task CNN is designed to complete two tasks at the same time by combining the fuzzy face image filtering and the excessive deflection angle face image filtering.In the process of validating the effectiveness of the filtering method,a ping-pong operation method is designed to realize the process similar to the pipelining process,and the timing of the ping-pong operation is realized by multithreading.In addition,this article compares the effects of various filtering methods through a test set.