Research On The Key Technologies Of Face Recognition In Classroom Environment

Face recognition technology is one of the main means of identity recognition and is widely used in security,finance and other fields.Research on the application of face recognition in the classroom environment is of great research value for improving teaching efficiency and promoting wisdom education.However,the factors such as shooting angle and light intensity in the classroom environment make face recognition in the classroom environment difficult.Aiming at this problem,this paper starts from three aspects: data set construction,face detection and face recognition.First,the lack of face data sets such as LFW and WIDER FACE is analyzed,and a data set for multi-face recognition research in the classroom environment is constructed.The data set construction includes several steps:(1)collect data from multiple orientations on different classroom environments and classes with different numbers of students;(2)process the pictures and calibrate the position of the face frame to generate corresponding label files;(3)Expand the number of data sets through data enhancement technology,and finally construct two types of data sets including 2782 images.Secondly,in the face detection section,the traditional face detection algorithms such as the Haar cascade classifier algorithm based on Open CV and the face detection algorithm based on dlib and the face detection algorithm based on deep learning are compared and analyzed.Through simulation experiments,the performance of algorithms in different face datasets(below 10 persons,11-20 persons,21-30 persons,31-40 persons,and 41 persons or more)is analyzed.The experimental results show that the traditional face detection algorithm It is difficult to accurately detect face pictures with a certain offset angle;the FCHD-based deep learning face detection algorithm has better performance for face pictures with a certain offset angle,and at the same time performs best in face data sets with less than 10 people(The average accuracy rate reaches 90%).Finally,in face recognition,unlike traditional recognition methods,this paper directly applies the target detection algorithm to face recognition.The paper comparatively analyzes several deep learning frameworks such as R-CNN,Fast R-CNN,and Faster R-CNN,and compares and analyzes Faster R-CNN detection algorithms.By adjusting multiple parameters such as related parameters,anchor points and non-maximum suppression thresholds in the RPN network,combined with the illumination invariance algorithm,a recognition result of 81.47% was obtained.