Research On Campus Security Pick-up Technology Based On Face Recognition

As a serious issue that maintains thousands of households,campus safety in primary and secondary schools has a close relationship with each teacher,student,and parent.In recent years,campus safety accidents have occurred frequently,especially accidental injuries of school-age children between 5 and 12 years old.The problem of campus safety has become the focus of discussion in the whole society.In order to facilitate parents and school teachers to grasp the situation of students entering and leaving school in time,this paper uses the safe transportation of primary and secondary schools as the research background.Aiming at the costly hardware equipment and the defects of single-person detection,a facial recognition technology using ordinary cameras is proposed.It can accurately identify the identities of the pick-up and drop-off personnel,prevent unidentified people from entering the campus,and avoid hidden dangers such as students being wrongly picked up.This dissertation mainly launched the research from the following directions:(1)Research and improvement of face detection model.This paper is based on the YOLOv3 target detection algorithm,in view of the large clustering deviation caused by the different selection of the initial clustering by the K-means algorithm,the K-means++ algorithm is used to improve the center position of the clustering priori box.Prediction makes the distance between the initial cluster points farther and improves the accuracy of the bounding box.At the same time,the Soft-NMS algorithm is used to improve the original non-maximum suppression method.The improved detection algorithm is used to train the face detection model of this article on the WIDER FACE dataset,which is used to detect the faces of pedestrians in the video stream.The accuracy of this model on the MOT16 multi-target tracking data set and the image data of the real scene reaches98%.(2)The facial image quality evaluation algorithm FIQUE.Pictures taken by ordinary camera equipment often face some complex real-world environments,such as insufficient light,occlusion of faces,and small face sizes,etc.,which will affect the quality of face images.For this reason,after the face detection is completed,this paper proposes a face image quality evaluation algorithm,which performs a mean normalization operation on the face image and fits it into an asymmetric generalized Gaussian distribution structure,and then passes it to classification in support vector machines.The algorithm can accurately evaluate the quality of face images,filter faces outside the set threshold score,filter out unqualified faces,and increase the proportion of high-quality face images.(3)Face feature extraction based on Inception-Res Net-v1 model.In this paper,the face filtered by the face image quality evaluation algorithm is sent to the Inception-Res Net-v1 model for face feature extraction,combined with the local face database,the two faces are judged by the Euclidean distance calculation of the face feature vector.The paper selects LFW,Age DB-30 and CFP face data sets,and conducts multiple sets of tests,and the accuracy of face recognition of this system is higher than 97%.At the same time,the paper collects video frames in real scenes for testing,and the experimental results show that the face recognition technology in this paper has good robustness.