Classroom face recognition attendance method and system driven by edge computing

Abstract

The application relates to the field of face recognition technology and provides a classroom face recognition attendance method and system driven by edge computing. The method comprises the following steps: a target classroom main camera is activated to collect and generate a face group image; a main edge node is used to perform individual segmentation screening and face recognition on the image, generate first face information, a blurred position and match auxiliary camera information; the main edge node shares information to an auxiliary edge node, the auxiliary camera is activated to perform secondary collection and recognition on the blurred position, and second face information is generated; and the first and second face recognition information is shared to a cloud center to record attendance. The application solves the technical problems of low accuracy of face recognition in a multi-person scene, low recognition rate and low recognition accuracy caused by blurred or blocked faces in the classroom attendance process, achieves high-precision and low-delay face recognition attendance through edge computing and cooperation of main and auxiliary cameras, and improves the recognition success rate of blurred or blocked faces.

Description

Technical Field

[0001] This application relates to the field of facial recognition technology, specifically to an edge computing-driven classroom facial recognition attendance method and system. Background Technology

[0002] With the acceleration of educational informatization, classroom attendance, as a crucial aspect of teaching management, is evolving from traditional manual roll call to intelligent and automated methods. Among these, facial recognition technology, due to its non-contact and efficient characteristics, has significant application potential in classroom attendance scenarios. However, the actual classroom environment is characterized by dense student populations, fixed seating but diverse individual postures (such as turning heads, looking down, or turning sideways). Traditional single-camera facial recognition solutions often face the following technical bottlenecks: First, due to limited viewing angles or insufficient resolution, cameras struggle to capture high-precision facial features of all individuals in a group image, especially when students' faces are obscured or not directly facing the camera, leading to missed detections or misjudgments. Second, cloud-based facial recognition models suffer from network latency, making it difficult to meet the real-time requirements of classroom attendance. Furthermore, statically configured cameras struggle to adapt to dynamically changing individual postures, resulting in some students' facial features not being effectively captured. Therefore, there is an urgent need for a method that can overcome the recognition bottlenecks in dense environments and achieve high-precision, low-latency automated attendance. Summary of the Invention

[0003] This application provides an edge computing-driven classroom face recognition attendance method and system, aiming to solve the technical problems of low accuracy and reduced recognition rate in multi-person scenarios during classroom attendance, as well as the problem of some faces being blurred or occluded, leading to low recognition accuracy.

[0004] The first aspect disclosed in this application provides an edge computing-driven classroom face recognition attendance method, the method comprising: activating a main camera in the target classroom to acquire a group of faces at a first moment, generating a group of faces image; using a main edge computing node corresponding to the main camera to perform individual segmentation and face recognition on the group of faces image based on face recognition accuracy, generating first face recognition information, blurred location, and matching auxiliary camera information; sharing the auxiliary camera information to an auxiliary edge computing node through the main edge computing node to activate the auxiliary camera to acquire auxiliary face images and perform face recognition on the blurred location, generating second face recognition information; and sharing the first face recognition information and the second face recognition information to a cloud center for attendance result recording.

[0005] Another aspect of this application discloses an edge computing-driven classroom face recognition attendance system, the system comprising: an image acquisition module: activating a main camera in the target classroom to acquire group face images at a first moment, generating a group face image; a face recognition module: using the main edge computing node corresponding to the main camera to perform individual segmentation and face recognition on the group face image based on face recognition accuracy, generating first face recognition information, blurred positions, and matching auxiliary camera information; an auxiliary face recognition module: sharing the auxiliary camera information to the auxiliary edge computing node through the main edge computing node to activate the auxiliary camera to acquire auxiliary face images and perform face recognition on the blurred positions, generating second face recognition information; and an attendance result recording module: sharing the first face recognition information and the second face recognition information to a cloud center for attendance result recording.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The aforementioned edge computing-driven classroom facial recognition attendance method works by first activating the main camera at the start of class to capture group facial images of all students. The corresponding main edge computing node then analyzes these images, selecting directly identifiable individuals based on recognition accuracy and verifying their identities. Simultaneously, it marks blurry areas with insufficient recognition accuracy and matches these areas with potential auxiliary cameras. Subsequently, the main edge computing node sends this auxiliary camera information to its corresponding auxiliary edge computing nodes, triggering the auxiliary cameras to re-capture and recognize the blurry areas, resulting in a second batch of supplementary facial recognition results. Finally, both batches of recognition information are uploaded to the cloud center to generate a complete classroom attendance record. This entire process, through distributed collaborative computing of edge nodes, effectively solves the problem of blind spots caused by partial occlusion in group scenarios while reducing cloud load, ensuring the completeness and accuracy of attendance results.

[0007] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1This is a flowchart illustrating an edge computing-driven classroom face recognition attendance method in one embodiment.

[0010] Figure 2 This is an architecture diagram of an edge computing-driven classroom face recognition attendance system in one embodiment.

[0011] Figure labeling: Image acquisition module 11, face recognition module 12, auxiliary face recognition module 13, attendance result recording module 14. Detailed Implementation

[0012] This application provides an edge computing-driven classroom face recognition attendance method and system, which solves the technical problems of low accuracy, blurred or occluded faces leading to decreased recognition rate and low recognition accuracy in classroom attendance scenarios with multiple people.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product, or device.

[0015] Example 1, as Figure 1 As shown, this application provides an edge computing-driven classroom face recognition attendance method, the method comprising: The main camera in the target classroom is activated to capture group facial images in the first instant, generating a group facial image.

[0016] In this embodiment, a start command is first sent to the main camera in the target classroom at a predetermined time, i.e., the scheduled attendance time. The main camera then enters the acquisition state according to pre-set operating parameters (including resolution, frame rate, focal length, white balance mode, etc.) and performs a panoramic shot of the entire student area within its coverage. This shooting process can be achieved by using a wide-angle lens, gimbal rotation, or multi-lens stitching to ensure that all seating areas in the classroom are completely included in the frame, thus containing as much student facial feature information as possible in a single frame. After acquisition, the main camera performs necessary local preprocessing on the captured raw image data, including brightness and contrast adjustment, distortion correction, image denoising, and format encoding, to improve the accuracy and efficiency of subsequent face detection and recognition. After the above preprocessing, one or more frames of high-quality image data containing the faces of all students in the classroom are obtained, serving as a group face image to provide the raw data foundation for subsequent individual segmentation, filtering, and face recognition steps.

[0017] Furthermore, this application provides configuration steps for the main camera, including: Collect seating arrangement information and classroom space characteristics of the target classroom; perform classroom modeling based on the seating arrangement information and classroom space characteristics to generate a classroom space twin model; collect camera parameters of a preset camera model, combine with the classroom space twin model, simulate the camera's field of view coverage, perform blind-spot-free monitoring simulation, and determine the configuration parameters of the main camera, including its configuration location and quantity.

[0018] Preferably, the seating arrangement information and spatial characteristics of the target classroom are first obtained through LiDAR scanning or CAD drawing import. The seating arrangement information includes data such as the number of seats in each row, spacing, number of rows, seat height, and layout information such as aisle location and podium location. Classroom spatial characteristic parameters include the classroom's length, width, height, window and lighting positions, and obstacle distribution. Subsequently, the seating arrangement information and spatial characteristic data are imported into modeling software (such as BIM software) to create a virtual 3D model consistent with the actual classroom scale, forming a classroom space twin model used to simulate the subsequent camera deployment effect. Then, detailed camera parameters of the preset main camera model are obtained, including horizontal and vertical viewing angles, resolution, lens focal length, effective imaging distance, and zoom range, and these parameters are loaded into the classroom space twin model. In the virtual environment, the width of the near-end rectangle and the width of the far-end rectangle are calculated based on the camera's horizontal viewing angle and maximum / minimum effective imaging distance, and the height of the near-end rectangle and the height of the far-end rectangle are calculated based on the camera's vertical viewing angle and maximum / minimum effective imaging distance. By placing the center points of the near and far rectangles along the optical axis of the camera at the minimum and maximum effective imaging distances, respectively, and connecting the vertices of the four rectangles, a field-of-view cone is formed for the camera. Then, a full-coverage layout simulation of the target classroom is performed based on this field-of-view cone. During this process, the monitoring coverage is gradually optimized by adjusting the installation height, angle, and position of the cameras to ensure there are no blind spots caused by obstruction, insufficient viewing angle, or distortion. If a blind spot is found, a new virtual camera is added, and the above adjustment and coverage simulation is repeated until all seating areas are within the line of sight of at least one camera. After the simulation, the final coordinates, installation height, orientation angle, and lens parameters of each virtual camera in the model are recorded. The total number of cameras required is calculated, and a main camera configuration parameter table is generated to guide the actual hardware deployment and ensure complete acquisition of student facial information within the panoramic view of the classroom.

[0019] The main edge computing node corresponding to the main camera is used to perform individual segmentation and face recognition based on face recognition accuracy on the face group image, generating first face recognition information, blurred position and matching auxiliary camera information.

[0020] In one embodiment, after the group face image is acquired, it is transmitted to the main edge computing node connected to the main camera. This node then calls a locally deployed face detection and recognition algorithm to process the entire image. First, all suspected face regions are located in the image, and the corresponding face recognition accuracy is calculated for each region. For facial regions whose face recognition accuracy reaches or exceeds a preset face recognition accuracy constraint, they are segmented from the group image to form a high-quality set of individual facial images. These images are then compared with the feature vectors of a pre-built student face attendance database to confirm the student's identity and generate the first face recognition information. Facial regions below a threshold are marked as ambiguous locations. These regions are typically inaccurate due to pose deviation, partial occlusion, or insufficient lighting. Subsequently, the main edge computing node further analyzes the ambiguous locations. Specifically, based on the row and column of the ambiguous location and the classroom spatial twin model, it matches one or more auxiliary cameras most likely to clearly capture the student's face, generating corresponding auxiliary camera information to provide clear execution instructions for subsequent auxiliary acquisition and recognition steps. This not only ensures that students who can be directly identified can complete their attendance quickly, but also allows for the development of precise secondary data collection plans for individuals who are difficult to identify, thereby improving the overall integrity and accuracy of attendance records.

[0021] Furthermore, this application provides configuration steps for the auxiliary camera, including: Construct individual student simulation models; load the individual student simulation models into the classroom space twin model according to the seating arrangement information, and perform iterative adjustment simulation of facial posture for any individual based on a preset negative example posture set; identify the main camera shooting defects after facial posture adjustment based on the configuration parameters of the main camera; configure auxiliary cameras according to the main camera shooting defects and the classroom space characteristics.

[0022] Optionally, based on the real students' facial features (such as hairstyle, skin color, and facial proportions), body size differences (such as height and head size), and common clothing (such as hats, scarves, and other factors that may affect recognition), modeling software is used to construct individual student simulation models that match the actual students' body proportions. These models include independently adjustable head and facial posture parameters. Then, according to pre-obtained classroom seating information, these individual student simulation models are loaded one by one into the corresponding seat positions in the classroom space twin model, forming a virtual scene highly consistent with the real classroom. Next, a preset set of negative example postures is used to iteratively adjust the virtual students' facial postures. This set of negative example postures covers common non-frontal states in the classroom, such as turning the head left and right, tilting up and down, looking down at a book, and talking to a neighbor. After each posture adjustment, the scene is virtually filmed based on the determined main camera configuration parameters. The system automatically detects whether the main camera's field of view can clearly capture the students' facial features and identifies areas where the posture changes prevent the camera from meeting the required recognition accuracy, which are then considered as main camera shooting defects. To address these shortcomings, and considering the characteristics of the classroom space, the effect of deploying auxiliary cameras is simulated in a classroom space twin model. Their installation position, height, and shooting angle are adjusted to cover facial areas that the main camera cannot clearly capture. In actual operation, the main camera remains active, performing global monitoring, while multiple auxiliary cameras are triggered as needed based on the real-time shooting results from the main camera. The corresponding auxiliary camera is activated only when a student's face is detected as blurry or recognition fails, thus ensuring attendance integrity while reducing energy consumption and data redundancy.

[0023] Furthermore, this application provides that the preset negative example pose set includes multiple negative example poses, which are extracted from historical attendance negative example data.

[0024] Optionally, the preset negative example pose set is constructed by analyzing historical attendance negative example data. This data refers to image samples and their corresponding pose features from past classroom facial recognition attendance records where inappropriate student postures led to recognition failures, errors, or decreased accuracy. Examples include students looking down while reading, turning to talk to their deskmate, covering their face with their hands, turning to look at the blackboard or window, or resting on their desks. These actions can prevent the main camera from fully capturing facial features, resulting in biased attendance results. By labeling these negative example images and using a corresponding convolutional neural network (CNN, iteratively trained through forward propagation, loss calculation, backpropagation, and parameter optimization), a series of representative pose parameter combinations are identified, such as head horizontal yaw angle, head vertical pitch angle, and the angle between the face orientation and the camera's visual axis. Each combination is then organized into a negative example pose, forming the preset negative example pose set. This set is used to simulate potentially unfavorable recognition scenarios in actual classrooms, allowing for early identification of potential blind spots and optimization of the main and auxiliary camera configuration strategies during camera deployment.

[0025] Furthermore, this application provides a method for performing individual segmentation and face recognition based on face recognition accuracy on the group face image using the main edge computing node corresponding to the main camera, generating first face recognition information, blurred location, and matching auxiliary camera information, including: The system reads a pre-built student face attendance database; performs face detection on the group face image, extracts a set of individual face regions that meet the preset face recognition accuracy constraints, and identifies the locations of individuals whose face regions do not meet the preset face recognition accuracy constraints, generating the fuzzy location; iterates through the student face attendance database to compare features of each individual face region in the set of individual face regions to determine the individual's identity and generate the first face recognition information; identifies the individual's pose at the fuzzy location in the group face image, matches an auxiliary camera based on the individual's pose, and generates the auxiliary camera information.

[0026] Preferably, in the attendance recognition stage, the main edge computing node first calls the locally stored student face attendance database. This database stores one or more sets of high-quality facial feature vectors for each student, along with associated student ID, name, class, and other identity information. Subsequently, the main edge computing node performs face detection on the group face images from the main camera. That is, the group face images are input into a lightweight detector (such as an anchor / anchor-free CNN), which outputs candidate boxes, confidence scores, and multi-point keypoints. Then, non-maximum suppression is applied to the candidate boxes, that is, the box with the highest confidence score is retained, and other boxes with overlap exceeding a threshold are removed, forming the final face set. Next, for each face in the face set, the number of keypoints determined by the analysis of these faces is divided by the preset total number of keypoints. Then, the quotient is subtracted from 1 to obtain the degree of facial occlusion. By comparing the degree of facial occlusion of each face with the preset face recognition accuracy constraint, if the degree of facial occlusion is less than or equal to the preset face recognition accuracy constraint, it means that the face corresponding to the candidate box is not occluded. In this case, the face corresponding to the candidate box is segmented and added to the individual face region set. Conversely, if the degree of facial occlusion is greater than the preset face recognition accuracy constraint, it means that the face corresponding to the candidate box is occluded. In this case, the position coordinates corresponding to the candidate box are extracted to form a blurred position. Then, the individual face region set is traversed. In each traversal, the traversed individual face region image is input into the CNN for feature extraction, such as skin details (wrinkles, spots, pore distribution), light reflection characteristics (such as forehead highlights, nose shadows), edges and gradients (such as edge sharpness and gradient direction of facial organs), etc., thereby converting the individual face region image into the corresponding feature vector. By comparing this feature vector with all feature vectors in the student facial attendance database one by one, the identity is confirmed through cosine similarity and the corresponding similarity threshold, generating the corresponding first facial recognition information. In addition, the posture features of students corresponding to blurred positions in the group facial image are analyzed, such as facial orientation (turning left, turning right, looking down, looking up, etc.), head tilt angle, etc., and combined with the preset auxiliary camera configuration parameters, one or more auxiliary cameras that can obtain the best shooting effect under the current posture are selected to generate auxiliary camera information, providing the execution basis for subsequent secondary acquisition and recognition steps.

[0027] Furthermore, this application provides that the preset face recognition accuracy constraint is a face occlusion tolerance that satisfies the preset recognition accuracy.

[0028] Optionally, a preset face recognition accuracy constraint is used to determine whether the collected face image can be used for face recognition. One of the core indicators is the face occlusion tolerance, which is the range of facial features that can be partially occluded while ensuring that the face recognition result reaches the preset accuracy threshold. For example, the occlusion tolerance can be set to no more than 20% of the total area of ​​the face region. That is, when a student's face is partially occluded by hands, books, hairstyles, masks or other objects, but does not exceed this proportion, key feature points (such as eyes, nose bridge, mouth contours, etc.) can still be accurately extracted by the recognition algorithm and the identity recognition accuracy requirements can be met.

[0029] Furthermore, this application provides methods for identifying individual poses at blurred locations in the group face image, matching auxiliary cameras based on individual poses, and generating auxiliary camera information, including: The facial orientation features are identified based on the individual's posture; based on the facial orientation features, the auxiliary camera at the corresponding position is matched according to the configuration parameters of the auxiliary camera, and the auxiliary camera information is generated.

[0030] Optionally, after determining the blurred location, the main edge computing node will perform pose recognition processing on the individual student at that location. In this process, the CNN used to construct the preset negative example pose set will be called to identify the image corresponding to the blurred location in the face group image, extract the coordinates of the student's head and facial key points, as well as the facial orientation features corresponding to these key points, including the horizontal yaw angle (such as the specific degree of left turn or right turn), the vertical pitch angle (such as the specific degree of head down or head up), and the head tilt angle. These facial orientation features are used to accurately describe the spatial orientation of the student's face relative to the main camera at the current moment. Subsequently, the facial orientation features are matched with the configuration parameters of the auxiliary cameras, which include their installation location, shooting angle, and coverage area. The angle between the student's facial orientation vector and the auxiliary camera's lens orientation vector is calculated to determine if the angle falls within the camera's coverage area. This process selects one or more auxiliary cameras capable of acquiring a clear facial image in the current pose, generating auxiliary camera information including the required camera number, shooting command, and target coordinates. This information is then passed to the corresponding auxiliary edge computing node to trigger the auxiliary cameras for subsequent supplementary face acquisition and recognition. This ensures the targeted and efficient nature of the supplementary acquisition process, avoids unnecessary activation of auxiliary equipment, reduces energy consumption, and improves the overall recognition success rate.

[0031] The main edge computing node shares the auxiliary camera information with the auxiliary edge computing node to activate the auxiliary camera to perform auxiliary face image acquisition and face recognition at the blurred location, generating second face recognition information.

[0032] In one embodiment, after identifying blurred locations in a group of faces and matching auxiliary cameras, the main edge computing node sends the auxiliary camera information to the corresponding auxiliary edge computing nodes in the form of message packets. The message packets contain the target student's temporary tracking ID, blurred location, facial orientation features, and a list of candidate auxiliary cameras. Upon receiving the message packets, the auxiliary edge computing nodes issue immediate control to the selected auxiliary cameras, activating them for auxiliary image acquisition and face recognition to generate second face recognition information, ensuring the accuracy of the recognition.

[0033] Furthermore, this application provides a method for sharing the auxiliary camera information to the auxiliary edge computing node through the main edge computing node to activate the auxiliary camera for auxiliary face image acquisition and face recognition at the blurred location, generating second face recognition information, including: The auxiliary camera is activated to acquire an auxiliary face image at the blurred location, generating an auxiliary face image; the face region is extracted from the blurred location in the auxiliary face image, and the extraction result is compared with the features of the student face attendance database to generate the second face recognition information.

[0034] Preferably, upon receiving a message packet from the main edge computing node, the corresponding auxiliary edge computing node activates its auxiliary camera and, based on the blurred location in the message packet, controls the auxiliary camera to perform position calibration, aligning its field of view with the student's face area requiring additional capture. Subsequently, the auxiliary camera enters acquisition mode, taking one or more shots of the blurred location at a preset resolution, frame rate, and shutter speed to obtain a high-quality auxiliary face image. After acquisition, the auxiliary edge computing node performs the same preprocessing as the main edge computing node locally, such as brightness and contrast adjustment and distortion correction, to ensure the accuracy of subsequent recognition. Then, precise face region extraction is performed on the blurred location in the auxiliary face image, using the same extraction method as described above, to obtain the feature vector of the blurred location. By comparing the extracted feature vector with the feature vectors stored in the student face attendance database one by one, second face recognition information is formed to supplement the individual records that the main camera failed to recognize, improving the overall attendance completeness and recognition accuracy.

[0035] The first and second facial recognition information are shared to the cloud center for attendance record recording.

[0036] In one embodiment, after obtaining the first and second facial recognition information, this information is integrated into a complete attendance record, packaged into a data packet using an encryption protocol, and sent to the cloud center via a secure communication channel. Upon receiving the data packet, the cloud center further verifies and stores the transmitted attendance information. During this process, it checks whether the record's timestamp matches the class schedule to ensure the timeliness of the attendance data. Then, it uses an identity verification mechanism to confirm that the recognition information matches the records in the student database, ensuring the accuracy of the identity. If the information is correct, the cloud center saves the attendance record to the cloud database and generates a corresponding attendance statistics report for teachers or administrators to query and audit. This process also ensures the traceability of the entire attendance process; all recognition results and related data are completely recorded, facilitating future queries and audits.

[0037] Furthermore, this application provides that after sharing the first facial recognition information and the second facial recognition information to the cloud center for attendance result recording, it also includes: The main camera in the target classroom captures group facial images at the second moment. The group facial images at the first moment and the group facial images at the second moment are compared to determine the consistency of the dynamic mapping between students and seats. If the result is consistent, the attendance record at the first moment is used as the attendance record at the second moment.

[0038] Optionally, at the second moment, the main camera in the target classroom again captures group facial images. These images are compared with the images from the first moment using a dynamic mapping algorithm to determine if the students in each seat are consistent. Specifically, by analyzing the seating layout in the images from the first and second moments, the position of each seat in the image is automatically marked. By comparing whether there is someone at the same position in the images from the first and second moments, if facial images are present at the same position in both moments, it indicates that the relative positions of these areas have not changed significantly. In this case, it is considered that the students in the target classroom have not changed, i.e., their relative positions are consistent, and the attendance record from the first moment is directly copied as the attendance result for the second moment, without needing to re-perform facial recognition, thus saving computational resources and processing time. If a position changes at the second moment—that is, if a face was present at that position in the first moment but not in the second moment—the facial recognition process is restarted to process the changed position, ensuring the accuracy of the attendance record. This consistency judgment method based on seat mapping avoids re-recognizing the entire image in most cases, effectively saving computational resources while ensuring the efficiency and accuracy of attendance records.

[0039] In summary, the embodiments of this application have at least the following technical effects: This embodiment first activates the main camera in the target classroom to capture group facial images at the first moment, generating a group facial image. Then, using the main edge computing node corresponding to the main camera, it performs individual segmentation and face recognition based on facial recognition accuracy on the group facial image, generating first facial recognition information, blurred locations, and matching auxiliary camera information. Next, the main edge computing node shares the auxiliary camera information to the auxiliary edge computing node to activate the auxiliary camera to capture auxiliary facial images and perform face recognition on the blurred locations, generating second facial recognition information. Finally, the first and second facial recognition information are shared to the cloud center for attendance recording. These technical effects collectively solve the technical problems of low accuracy and reduced recognition rate due to blurred or occluded faces in multi-person scenarios during classroom attendance, achieving high-precision, low-latency facial recognition attendance through edge computing in collaboration with main and auxiliary cameras, and improving the success rate of recognizing blurred or occluded faces.

[0040] Example 2, based on the same inventive concept as the edge computing-driven classroom face recognition attendance method in the aforementioned examples, such as... Figure 2 As shown, this application provides an edge computing-driven classroom face recognition attendance system. The system includes: an image acquisition module 11: activating the main camera in the target classroom to acquire group face images at the first moment, generating a group face image; a face recognition module 12: using the main edge computing node corresponding to the main camera to perform individual segmentation and face recognition on the group face image based on face recognition accuracy, generating first face recognition information, blurred positions, and matching auxiliary camera information; an auxiliary face recognition module 13: sharing the auxiliary camera information to the auxiliary edge computing node through the main edge computing node to activate the auxiliary camera to acquire auxiliary face images and perform face recognition on the blurred positions, generating second face recognition information; and an attendance result recording module 14: sharing the first face recognition information and the second face recognition information to the cloud center for attendance result recording.

[0041] Furthermore, the image acquisition module 11 is also used to perform the following method: Collect seating arrangement information and classroom space characteristics of the target classroom; perform classroom modeling based on the seating arrangement information and classroom space characteristics to generate a classroom space twin model; collect camera parameters of a preset camera model, combine with the classroom space twin model, simulate the camera's field of view coverage, perform blind-spot-free monitoring simulation, and determine the configuration parameters of the main camera, including its configuration location and quantity.

[0042] Furthermore, the face recognition module 12 is also used to perform the following methods: Construct individual student simulation models; load the individual student simulation models into the classroom space twin model according to the seating arrangement information, and perform iterative adjustment simulation of facial posture for any individual based on a preset negative example posture set; identify the main camera shooting defects after facial posture adjustment based on the configuration parameters of the main camera; configure auxiliary cameras according to the main camera shooting defects and the classroom space characteristics.

[0043] Furthermore, the auxiliary face recognition module 12 is also used to perform the following methods: The preset negative example pose set includes multiple negative example poses, which are extracted from historical attendance negative example data.

[0044] Furthermore, the face recognition module 12 is also used to perform the following methods: The system reads a pre-built student face attendance database; performs face detection on the group face image, extracts a set of individual face regions that meet the preset face recognition accuracy constraints, and identifies the locations of individuals whose face regions do not meet the preset face recognition accuracy constraints, generating the fuzzy location; iterates through the student face attendance database to compare features of each individual face region in the set of individual face regions to determine the individual's identity and generate the first face recognition information; identifies the individual's pose at the fuzzy location in the group face image, matches an auxiliary camera based on the individual's pose, and generates the auxiliary camera information.

[0045] Furthermore, the face recognition module 12 is also used to perform the following methods: The preset face recognition accuracy constraint is the tolerance for facial occlusion that meets the preset recognition accuracy.

[0046] Furthermore, the face recognition module 12 is also used to perform the following methods: The facial orientation features are identified based on the individual's posture; based on the facial orientation features, the auxiliary camera at the corresponding position is matched according to the configuration parameters of the auxiliary camera, and the auxiliary camera information is generated.

[0047] Furthermore, the auxiliary face recognition module 13 is also used to perform the following methods: The auxiliary camera is activated to acquire an auxiliary face image at the blurred location, generating an auxiliary face image; the face region is extracted from the blurred location in the auxiliary face image, and the extraction result is compared with the features of the student face attendance database to generate the second face recognition information.

[0048] Furthermore, the attendance result recording module 14 is also used to perform the following method: The main camera in the target classroom captures group facial images at the second moment. The group facial images at the first moment and the group facial images at the second moment are compared to determine the consistency of the dynamic mapping between students and seats. If the result is consistent, the attendance record at the first moment is used as the attendance record at the second moment.

[0049] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0050] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0051] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. An edge computing-driven classroom face recognition attendance method, characterized in that, include: The main camera in the target classroom is activated to capture group facial images in the first instant, generating a group facial image. The main edge computing node corresponding to the main camera is used to perform individual segmentation and face recognition based on face recognition accuracy on the face group image, generating first face recognition information, blurred position and matching auxiliary camera information; The auxiliary camera information is shared with the auxiliary edge computing node through the main edge computing node to activate the auxiliary camera to perform auxiliary face image acquisition and face recognition at the blurred position, and generate second face recognition information; The first facial recognition information and the second facial recognition information are shared to the cloud center for attendance record recording; The configuration steps for the auxiliary camera include: Construct individual student simulation models; The individual student simulation model is loaded into the classroom space twin model according to the seating arrangement information, and the facial posture is iteratively adjusted for any individual based on a preset set of negative example postures. The main camera shooting defects after facial posture adjustment are identified based on the configuration parameters of the main camera. The auxiliary camera is configured based on the shooting defects of the main camera and the characteristics of the classroom space.

2. The edge computing-driven classroom face recognition attendance method as described in claim 1, characterized in that, The configuration steps for the main camera include: Collect information on the seating arrangement and spatial characteristics of the target classroom; Based on the seating arrangement information and classroom space characteristics, a classroom space twin model is generated. The camera parameters of a preset camera model are collected and combined with the classroom space twin model to simulate the camera's field of view coverage, conduct a no-blind-spot monitoring simulation, and determine the configuration parameters of the main camera, including its configuration location and quantity.

3. The edge computing-driven classroom face recognition attendance method as described in claim 1, characterized in that, The preset negative example pose set includes multiple negative example poses, which are extracted from historical attendance negative example data.

4. The edge computing-driven classroom face recognition attendance method as described in claim 1, characterized in that, The system utilizes the main edge computing node corresponding to the main camera to perform individual segmentation and face recognition on the group of faces image based on face recognition accuracy, generating first face recognition information, blurred location, and matching auxiliary camera information, including: Read the pre-built student facial recognition attendance database; Face detection is performed on the group face image to extract the set of individual face regions that meet the preset face recognition accuracy constraints, and the locations of individuals whose face regions do not meet the preset face recognition accuracy constraints, and the fuzzy locations are generated. Each individual face region in the set of individual face regions is traversed through the student face attendance database for feature comparison to determine the individual's identity and generate the first face recognition information; Identify the individual pose at the blurred location in the group face image, match the auxiliary camera based on the individual pose, and generate the auxiliary camera information.

5. The edge computing-driven classroom face recognition attendance method as described in claim 4, characterized in that, The preset face recognition accuracy constraint is the tolerance for facial occlusion that meets the preset recognition accuracy.

6. The edge computing-driven classroom face recognition attendance method as described in claim 5, characterized in that, Identify the individual pose at the blurred location in the group face image, match the auxiliary camera based on the individual pose, and generate the auxiliary camera information, including: Facial orientation features are identified based on the individual's posture. Based on the facial orientation features, and according to the configuration parameters of the auxiliary cameras, the corresponding auxiliary cameras are matched to generate the auxiliary camera information.

7. The edge computing-driven classroom face recognition attendance method as described in claim 1, characterized in that, After sharing the first and second facial recognition information to the cloud center for attendance record recording, the process also includes: The main camera in the target classroom captures group facial images at the second moment, and the group facial images at the first moment and the second moment are compared to determine the consistency of the dynamic mapping between students and seats. If the judgment results are consistent, the attendance record result at the first moment shall be used as the attendance record result at the second moment.

8. The edge computing-driven classroom face recognition attendance method as described in claim 1, characterized in that, The primary edge computing node shares the auxiliary camera information with the auxiliary edge computing node to activate the auxiliary camera for assisted facial image acquisition and facial recognition at the blurred location, generating second facial recognition information, including: Activate the auxiliary camera to acquire auxiliary facial images of the blurred location and generate an auxiliary facial image; In the auxiliary face image, the blurred area is extracted, and the extraction result is compared with the features of the student face attendance database to generate the second face recognition information.

9. An edge computing-driven classroom facial recognition attendance system, characterized in that, The system is used to execute the edge computing-driven classroom face recognition attendance method according to any one of claims 1-8, including: Image acquisition module: Activate the main camera in the target classroom to acquire group facial images in the first instant and generate a group facial image; Face recognition module: Utilizes the main edge computing node corresponding to the main camera to perform individual segmentation and face recognition on the face group image based on face recognition accuracy, generating first face recognition information, blurred position, and matching auxiliary camera information; Assisted face recognition module: The auxiliary camera information is shared with the auxiliary edge computing node through the main edge computing node to activate the auxiliary camera to perform auxiliary face image acquisition and face recognition at the blurred position, and generate second face recognition information; Attendance Result Recording Module: Shares the first facial recognition information and the second facial recognition information to the cloud center for attendance result recording.

Images