Design And Implementation Of Classroom Name Prompting System Based On Face Recognition And Image Stitching

In higher education,as the main place of teaching and communication between teachers and students,college classrooms are becoming more and more intelligent in recent years.But for college teachers,if they want to communicate with a certain classmate in the classroom,they can only point directly to him by the appearance feature or seat name,or find a classmate randomly according to the list.The lack of respect and human relations in the classroom communication between teachers and students may lead to poor learning efficiency.This situation is common in college classrooms.For this reason,this paper focuses on the problem of student name prompting in college classrooms.The main tasks are listed below:(1)Combined with the characteristics of college classroom,a set of student name prompting system schemes for teachers was designed and implemented on PC and ARM development board.This system scans through the pan / tilt camera to obtain clear images in the classroom.The image stitching technology is used to stitch the acquired images to obtain a classroom panorama.And use image recognition algorithm to identify the identity information of each student in the picture.Finally,it is displayed to the teacher of the class through the display operation interface,so that it can clearly and intuitively obtain the names of the students present in the classroom,so as to facilitate the interaction between the teacher and the students.(2)In the process of the implementation of the student name prompt system,a fast image registration algorithm based on deep learning is proposed.Firstly,feature points are extracted and descriptors are generated using Super Point network.The method of violence matching combined with KNN is used to describe the sub matching,and then RANSAC algorithm is used to remove the outer points to obtain the matching points.The algorithm is implemented and tested on an ARM development board.Experimental results show that the algorithm can quickly extract stable feature points from classroom images.(3)A fast human segmentation network is proposed.In this network,edge detection task and semantic segmentation task are combined to segment human instance.In the training of network,a new loss function of edge detection is proposed,which can get better edge detection effect.In the encoder part of the network,Mobile Net V2 is used for feature extraction,and in the decoder part,combined with the residual idea of RESNET network,the network obtains a good segmentation effect of human body instances.(4)Based on the proposed fast segmentation network of human body instances,a human body weight mapping algorithm is proposed to solve the problems such as stitching ghosting caused by human motion.Firstly,the stitching line is generated by using the stitching line finding function in Open CV to check whether the stitching line passes through the human body.If part of the stitching line passes through the human body,the human body instance is segmented near the stitching line to obtain the human body mask,that is,the pixel level position of the human body.The pixels outside the human mask are spliced first,and then the pixels inside the human mask are mapped back to the spliced image.The experimental results show the effectiveness of the algorithm.