Intelligent attendance clock-in method and system based on face recognition

Abstract

The application discloses an intelligent attendance clock-in method and system based on face recognition and belongs to the technical field of image recognition, and comprises the following steps: S1, obtaining each functional area of a clock-in picture; S2, obtaining an optical feedback judgment result of the clock-in picture; S3, if the optical feedback judgment result of a certain clock-in picture is qualified, the clock-in picture is stored to an image recognition temporary storage area, if the optical feedback judgment results of all the clock-in pictures of continuous snapshots are all unqualified, image splicing processing is carried out, and the processed spliced image is stored to the image recognition temporary storage area; and S4, obtaining a face recognition comprehensive similarity, thereby obtaining an attendance clock-in judgment result, and issuing a prompt information, so that face recognition attendance can be quickly and accurately carried out under the condition of the influence of light environment, and the problem that the face recognition response speed is slow due to the influence of the light environment of attendance clock-in in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, in particular to an intelligent attendance clock-in method and system based on face recognition. BACKGROUND

[0002] The existing intelligent attendance clock-in system collects user facial images in real time through a camera, and then sequentially performs face detection, image preprocessing, feature extraction and other processing on the user facial images, and compares them with the registered templates in the database to complete identity matching and realize attendance clock-in.

[0003] For example, the classroom intelligent attendance method and system disclosed in Chinese patent CN112395950B includes pre-constructing a face library including the facial images of each student; triggering the electronic class board and the camera in the attendance classroom to enter the attendance mode; determining the set of students corresponding to the attendance classroom; performing clock-in attendance on the students through the electronic class board to obtain first attendance data; periodically collecting student images through the camera to obtain a set of student images, and performing face recognition on each image in the set of student images based on the face library to obtain second attendance data; and determining the attendance of each student in the set of students based on the first attendance data and the second attendance data.

[0004] For example, the multi-face tracking and recognition method, device, computer equipment and storage medium disclosed in Chinese patent CN113553990B include: acquiring face information of multiple targets in a current monitoring area; recording the time when the first face image of each target is detected; determining the acquisition degree of the face information of each target; when the acquisition degree does not satisfy a preset proportion, continuing to track the face of the target, supplementing and perfecting the face information of the target to obtain perfect target face information; performing face recognition based on the perfect target face information to obtain a recognition result; and taking the time of the first face image of the target as the clock-in time of the recognition result.

[0005] However, in the process of implementing the technical scheme of the present application, the present application has found that the above-mentioned technology at least has the following technical problems:

[0006] In the prior art, attendance clock-in is usually performed in an indoor or other well-lit environment. However, if it is performed outdoors or in a strong light environment, the problem of overexposure or severe shadow in part of the face image often occurs, which may cause the user to need to repeatedly perform face recognition multiple times, and in severe cases, the user may be identified as another person, which seriously affects the efficiency of the user's face recognition clock-in. Therefore, there is a problem of slow face recognition response speed caused by the light environment of attendance clock-in. SUMMARY

[0007] In order to solve the technical problem of slow response speed of face recognition caused by the light environment of attendance card punching in the prior art, the embodiments of the present application provide an intelligent attendance card punching method and system based on face recognition.

[0008] In one aspect, an intelligent attendance card punching method based on face recognition is provided, which comprises the following steps: S1, when receiving a user attendance card punching signal, obtaining continuous snapshots of the user during attendance card punching, counting to obtain card punching pictures, analyzing and dividing the card punching pictures to obtain each functional area of the card punching pictures; S2, obtaining optical feedback parameters of each dispersed area of the card punching pictures, analyzing to obtain optical feedback determination results of the card punching pictures, and the optical feedback parameters are used to reflect the imaging state of the card punching pictures; S3, based on the analysis of the optical feedback determination results of the card punching pictures, if the optical feedback determination result of a card punching picture is qualified, the card punching picture is stored in an image recognition temporary area, and if the optical feedback determination results of all the card punching pictures of the continuous snapshots are unqualified, image splicing processing is performed, and the processed spliced image is stored in the image recognition temporary area; S4, extracting the stored images in the image recognition temporary area, and analyzing to obtain a face recognition comprehensive similarity, thereby obtaining an attendance card punching determination result, and issuing a prompt information.

[0009] In another aspect, an intelligent attendance card punching system based on face recognition is provided, which applies the intelligent attendance card punching method based on face recognition, and the system comprises the following modules: a region division module, an optical feedback analysis module, a splicing processing module and an attendance card punching determination module; wherein the region division module is used to obtain continuous snapshots of the user during attendance card punching when receiving a user attendance card punching signal, count to obtain card punching pictures, analyze and divide the card punching pictures to obtain each functional area of the card punching pictures; the optical feedback analysis module is used to obtain optical feedback parameters of each dispersed area of the card punching pictures, analyze to obtain optical feedback determination results of the card punching pictures, and the optical feedback parameters are used to reflect the imaging state of the card punching pictures; the splicing processing module is used to analyze based on the optical feedback determination results of the card punching pictures, if the optical feedback determination result of a card punching picture is qualified, the card punching picture is stored in an image recognition temporary area, and if the optical feedback determination results of all the card punching pictures of the continuous snapshots are unqualified, image splicing processing is performed, and the processed spliced image is stored in the image recognition temporary area; the attendance card punching determination module is used to extract the stored images in the image recognition temporary area, and analyze to obtain a face recognition comprehensive similarity, thereby obtaining an attendance card punching determination result, and issuing a prompt information.

[0010] The technical scheme provided by the embodiments of the present application has at least the following beneficial effects:

[0011] The application provides an intelligent attendance clock-in method based on face recognition, which can automatically screen or optimize high-quality face images under complex light environments through region division, optical feedback parameter analysis, image splicing processing and comprehensive similarity calculation of clock-in pictures of different angles, thereby realizing face recognition attendance under the influence of light environment, and effectively solving the problem of slow face recognition response speed caused by the influence of light environment on attendance clock-in in the prior art.

[0012] 2. The application constructs a three-dimensional recognition model, and aligns the three-dimensional model to the clock-in picture by using face key point detection combined with a PnP pose estimation algorithm, so that the regions of the clock-in picture can be accurately divided, and one-to-one correspondence of multi-angle image regions can be realized combined with three-dimensional point cloud data, thereby realizing accurate matching of face regions under different angles and complex lighting conditions, and improving the accuracy of face feature extraction.

[0013] 3. The application can automatically distinguish qualified images and unqualified images by obtaining the optical feedback parameters of each scattered region of the clock-in picture and determining based on the optical feedback threshold, thereby realizing effective evaluation of light uniformity and exposure condition, ensuring the image quality stored in the image recognition temporary storage area, and improving the stability and reliability of subsequent face recognition.

[0014] 4. The application can improve the image quality when the local lighting condition is poor by splicing and optimizing the scattered regions, adjusting the to-be-optimized scattered region by using the brightness mean value and the exposure mean value of the adjacent preferred scattered region, thereby realizing adaptive optimization of the comprehensive face image, ensuring that the second clock-in picture or the third clock-in picture obtained by final splicing has high recognition usability, and effectively improving the precision of the comprehensive similarity calculation of face recognition. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0016] Figure 1 The flowchart of the intelligent attendance clock-in method based on face recognition provided by the embodiments of the application is shown in the figure.

[0017] Figure 2 The macro step flowchart of the intelligent attendance clock-in method based on face recognition provided by the embodiments of the application is shown in the figure.

[0018] Figure 3An image splicing processing flowchart of the intelligent attendance clock-in method based on face recognition provided in the embodiments of the present application is shown in the following figure;

[0019] Figure 4 A structure schematic diagram of the intelligent attendance clock-in system based on face recognition provided in the embodiments of the present application is shown in the following figure;

[0020] Figure 5 An adjacent area schematic diagram of the intelligent attendance clock-in method based on face recognition provided in the embodiments of the present application is shown in the following figure;

[0021] Figure 6 A clock-in record query schematic diagram of the intelligent attendance clock-in system based on face recognition provided in the embodiments of the present application is shown in the following figure;

[0022] Figure 7 A clock-in identity verification image quality schematic diagram of the intelligent attendance clock-in system based on face recognition provided in the embodiments of the present application is shown in the following figure;

[0023] Figure 8 A clock-in identity verification result schematic diagram of the intelligent attendance clock-in system based on face recognition provided in the embodiments of the present application is shown in the following figure. DETAILED DESCRIPTION

[0024] The technical solutions in the present application will be described below with reference to the drawings.

[0025] In the embodiments of the present application, the words such as “example”, “for example” are used to represent as an example, illustration or description. Any embodiment or design scheme described as “example” in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the word “example” is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by “and / or” can be both, or can be one of the two.

[0026] In the embodiments of the present application, “image” and “picture” can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. “Of”, “corresponding” and “corresponding” can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0027] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0028] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0029] As Figure 1 shown, the flow chart of the intelligent attendance clock-in method based on face recognition provided by the embodiment of the application, the method comprises the following steps: S1, when receiving the user attendance clock-in signal, obtaining the continuous snapshots of the user when performing attendance clock-in, and counting to obtain the clock-in pictures, analyzing and dividing the clock-in pictures to obtain the function areas of the clock-in pictures; S2, obtaining the optical feedback parameters of each scattered area of the clock-in pictures, and analyzing to obtain the optical feedback determination result of the clock-in pictures, the optical feedback parameters are used to reflect the imaging state of the clock-in pictures; S3, based on the optical feedback determination result analysis of the clock-in pictures, if the optical feedback determination result of a clock-in picture is qualified, the clock-in picture is stored to the image recognition temporary area, if the optical feedback determination results of all the clock-in pictures of the continuous snapshots are unqualified, image splicing processing is performed, and the processed spliced image is stored to the image recognition temporary area; S4, extracting the stored image of the image recognition temporary area, and analyzing to obtain the face recognition comprehensive similarity, thereby obtaining the attendance clock-in determination result, and issuing a prompt information.

[0030] In the embodiment, as Figure 2 shown, Figure 2 the macro step flow chart of the intelligent attendance clock-in method based on face recognition provided by the embodiment of the application, it can be seen from the figure that the overall process of the scheme first starts from obtaining the clock-in pictures from multiple angles, then extracts and analyzes the optical feedback parameters of the clock-in pictures to obtain the corresponding optical feedback determination result. If the determination result is qualified, the image is directly marked as the first clock-in picture; if the determination result is unqualified, the subsequent image splicing processing step is entered. The first clock-in picture and the clock-in picture obtained after splicing processing both need to perform face feature extraction, and calculate the face recognition comprehensive similarity based on different recognition schemes, finally obtain the attendance clock-in result and output, thereby completing the entire attendance recognition process.

[0031] The clock-in pictures of the user from different angles are obtained, specifically the clock-in pictures of the user when performing attendance clock-in, the attendance system (such as an attendance clock-in applet or an attendance machine) will issue a voice prompt to remind the user to make a specified action (such as left and right head shaking), at this time, the clock-in pictures of the user from multiple angles can be obtained. It needs to be explained that in this process, the continuous snapshots when performing attendance clock-in are obtained, so it is not necessary to consider the bad behaviors of the user when performing clock-in (such as rapidly switching expressions, making strange expressions, etc.). The scheme mainly considers the people who perform outdoor work, part-time work or clock-in work in an environment with large changes in lighting conditions (exposure), such as summer camp part-time workers, etc.

[0032] According to the attendance card punching judgment result, a prompt information is sent out, if the attendance card punching judgment result is card punching success, a card punching success prompt is sent out, if the attendance card punching judgment result is card punching failure, a card punching failure prompt is sent out, and re-punching is reminded.

[0033] An attendance card punching judgment result is obtained, and the specific method is as follows: a preset face recognition comprehensive similarity threshold in a database is obtained, and the face recognition comprehensive similarity is compared with the face recognition comprehensive similarity threshold, if the face recognition comprehensive similarity is above the face recognition comprehensive similarity threshold, the attendance card punching judgment result is card punching success, otherwise, the attendance card punching judgment result is card punching failure.

[0034] Further, each functional area of the card punching picture is obtained, and the specific method is as follows: the card punching picture is input into a three-dimensional recognition model, the three-dimensional recognition model performs face key point detection on the card punching picture and aligns the three-dimensional recognition model on the card punching picture based on a PnP pose estimation method, thereby analyzing each scattered area of the card punching picture, and each scattered area of the card punching picture is numbered to obtain the number of each scattered area of the card punching picture, wherein the card punching pictures with the same number are the same functional area, thereby obtaining each functional area of the card punching picture.

[0035] In the embodiment, it should be noted that the three-dimensional recognition model is a three-dimensional geometry and texture template in a unified face coordinate system, which is usually a dense three-dimensional grid (vertex / triangle) and is divided into several face functional areas (for example, left eye area, right eye area, nose bridge area, left cheek area, right cheek area, mouth area, and chin area) on the model according to semantics. It is used to map two-dimensional card punching pictures of different angles to the same semantic coordinate system, to ensure the consistency and comparability of the “same area” in different pictures, and its output form is a parameterized three-dimensional grid (vertex coordinate array, triangle index), and the area index of each vertex. The face coordinate system is a local three-dimensional coordinate system (origin, axial direction fixed) bound with the three-dimensional recognition model, which is used to define the region position and UV mapping (UV represents two-dimensional texture coordinates) in three-dimensional space. The face functional area is a semantic scattered area (for example, the “left eye area” is composed of several vertices / patches on the model) defined in advance on the three-dimensional model or its UV expansion plane. The face key point detection specifically detects a group of predefined face landmark points (such as two eye corners, nose tip, and mouth corner) on a two-dimensional image, which can be realized by a heat map regression method (FAN—Face Alignment Network) based on a convolutional neural network (CNN).

[0036] The specific method of point cloud matching is: obtaining the three-dimensional point cloud data of the target face, preprocessing the point cloud data by using the PCL library, including using the Voxel Grid Filter to reduce the sampling, and removing the noise points by using the Statistical Outlier Removal to improve the data quality, extracting the key points of the point cloud by using the ISS (Intrinsic Shape Signatures) algorithm in the PCL, and calculating the FPFH (Fast Point Feature Histograms) descriptor of the key points. By using the RANSAC (Random Sample Consensus) algorithm in the Open3D library, the feature points of the two point clouds are matched, so as to obtain the initial rigid transformation matrix, complete the coarse registration, use the ICP (Iterative Closest Point) algorithm in the PCL to perform fine registration on the point clouds after coarse registration, and further optimize the alignment effect of the point clouds by iteratively optimizing the distance between the point clouds, so as to realize high-precision point cloud matching. The same numbered region of the same point cloud data is added to the same region of the identified clock-in picture (continuous snapshot picture), for example, the left eye region is numbered as 01, the right eye region is numbered as 02, the nose bridge region is numbered as 03, and the mouth region is numbered as 04, so that the same region in the clock-in picture (continuous snapshot picture) is marked as a number.

[0037] Further, the optical feedback determination result of the clock-in picture is obtained, and the specific method is: obtaining the optical feedback parameters of each scattered region of the clock-in picture, the optical feedback parameters including the brightness mean value, the shadow coverage rate, the overexposure area proportion and the brightness range; based on the optical feedback parameters of each scattered region of the clock-in picture and the preset optical feedback calibration set in the database, the optical feedback values of each scattered region of the clock-in picture are obtained; the optical feedback calibration set includes the brightness mean value calibration value, the shadow coverage rate calibration value, the overexposure area proportion calibration value and the brightness range calibration value; the preset optical feedback threshold value in the database is obtained, and compared with the optical feedback values of each scattered region of the clock-in picture, to obtain the optical feedback determination result of the clock-in picture, if the optical feedback values of each scattered region of a clock-in picture are all above the optical feedback threshold value, the optical feedback determination result of the clock-in picture is qualified, and the clock-in picture is marked as the first clock-in picture, otherwise the optical feedback determination result of the clock-in picture is unqualified.

[0038] In the embodiment, the optical feedback parameters including the brightness mean value, the shadow coverage rate, the overexposure area proportion, the shadow area proportion and the brightness range can be obtained by using the OpenCV image processing tool library (Open Source Computer Vision Library).

[0039] The optical feedback value is obtained by analyzing the optical feedback parameters, considering the mutual influence relationship between the parameters. For example, when the average brightness is moderate, it can provide a clear baseline for face recognition. If the average brightness is too high, the proportion of overexposed area will increase, resulting in loss of facial feature details. If the average brightness is too low, the shadow coverage rate and the shadow area proportion will increase, causing the feature of the local area to be blurred. At the same time, the brightness range can reflect the uniformity of the brightness distribution of the image. If the range is too large, it often means that there are both serious shadows and overexposure, resulting in unstable overall image features. Therefore, the formation of the optical feedback value is the result of the mutual action of the above parameters. The higher the optical feedback value, the better the image quality, which is more conducive to the stable extraction and recognition of facial features.

[0040] By obtaining the optical feedback parameters (average brightness, shadow coverage rate, overexposure area proportion and brightness range) of each dispersed area of the punch-in picture, and calculating the optical feedback value of each area, the local image quality of the face can be quantitatively evaluated, ensuring that the feature differences of the face area under different lighting conditions are analyzed in detail, thereby improving the stability of the face feature extraction. By comparing the optical feedback value with the preset optical feedback calibration set and the optical feedback threshold, the optical feedback determination result is obtained, realizing the classification of qualified and unqualified punch-in pictures, effectively ensuring the image quality entering the subsequent face recognition process, and improving the overall recognition accuracy and response speed.

[0041] Further, the optical feedback value of each dispersed area of the punch-in picture is obtained. The specific method is: the difference degree analysis is performed on the average brightness and the average brightness calibration value to obtain the difference degree determination result; the shadow coverage rate, the reflection area proportion, the shadow area proportion and the brightness range are respectively analyzed with the shadow coverage rate calibration value, the reflection area proportion calibration value, the shadow area proportion calibration value and the brightness range calibration value to obtain the proportion analysis result; based on the difference degree determination result and the proportion analysis result, the corresponding weighting factor is introduced for coupling processing to obtain the optical feedback value of each dispersed area of the punch-in picture. The optical feedback value is used to represent the usability of the punch-in picture in face feature extraction.

[0042] In this embodiment, the optical feedback value of each dispersed area of the punch-in picture is obtained. The specific method is:

[0043]

[0044] In the formula, G i represents the optical feedback value of the i-th dispersed area of the punch-in picture, LY i represents the average brightness of the i-th dispersed area, LB represents the average brightness calibration value, YY i represents the shadow coverage rate of the i-th dispersed area, YB represents the shadow coverage rate calibration value, YSi SB represents the overexposure area proportion calibration value, YH represents the brightness mean calibration value, and ω1 represents the brightness mean weighting factor. i HB represents the brightness range calibration value, ω1 represents the brightness mean weighting factor, ω2 represents the shadow coverage rate weighting factor, ω3 represents the overexposure area proportion weighting factor, and ω4 represents the brightness range weighting factor.

[0045] It should be noted that the brightness mean weighting factor, the shadow coverage rate weighting factor, the overexposure area proportion weighting factor, and the brightness range weighting factor can be obtained from a database. For example, the brightness mean weighting factor can be obtained by comparing the historical brightness mean stored in the database with the brightness mean, marking the historical brightness mean that is within a preset range of the difference between the historical brightness mean and the brightness mean as each historical control brightness mean, obtaining the weighting factor of each historical control brightness mean and performing mean processing to obtain the mean value of the weighting factor of each historical control brightness mean, and performing standard deviation analysis on the weighting factor of each historical control brightness mean to obtain the standard deviation of the weighting factor of each historical control brightness mean. A preset standard deviation reference value is obtained from the database and compared with the standard deviation of the weighting factor of each historical control brightness mean. If the standard deviation of the weighting factor of each historical control brightness mean is below the mean value of the standard deviation, the mean value of the weighting factor of each historical control brightness mean is taken as the brightness mean weighting factor. Otherwise, the standard deviation of the weighting factor of each historical control brightness mean is subtracted from the standard deviation reference value to obtain a standard deviation adjustment coefficient, and the standard deviation adjustment value of the weighting factor of each historical control brightness mean is obtained by multiplying the standard deviation adjustment coefficient by the mean value of the weighting factor of each historical control brightness mean. The mean value of the weighting factor of each historical control brightness mean is adjusted by subtracting the standard deviation adjustment value of the weighting factor of each historical control brightness mean to obtain an adjusted weighting factor mean, and the adjusted weighting factor mean is taken as the weighting factor of the third region recognition similarity. The other weighting factors, such as the shadow coverage rate weighting factor, the overexposure area proportion weighting factor, and the brightness range weighting factor, are obtained in the same way as the brightness mean weighting factor.

[0046] Further, image stitching processing is performed, and the specific method is as follows: the scattered areas of the check-in picture are summarized to obtain the scattered areas of each functional area of the check-in picture; the optical feedback values of the scattered areas of each functional area of the check-in picture are compared with the optical feedback threshold to obtain the stitching optional determination result of each scattered area, if the optical feedback value of a certain scattered area is above the optical feedback threshold, the stitching optional determination result of the scattered area is qualified, and the scattered area is marked as an optional qualified scattered area, if the optical feedback value of a certain scattered area is less than the optical feedback threshold, the stitching optional determination result of the scattered area is unqualified, and the scattered area is marked as an optional scattered area; each functional area attribute stitching determination result is obtained based on the analysis of the scattered areas, and corresponding scattered area optimization processing is performed based on the functional area attribute stitching determination result to obtain each preferred scattered area of the check-in picture, and thus the image stitching processing is performed to obtain the check-in picture after the stitching processing, wherein if the check-in picture after the stitching processing is composed of the optional qualified scattered areas, the check-in picture is a second check-in picture, and if the check-in picture after the stitching processing contains the optional scattered area, the check-in picture is a third check-in picture.

[0047] In the embodiment, Figure 3 The image stitching processing flowchart of the intelligent check-in and clock-out method based on face recognition provided in the embodiment is as follows: the optical feedback values of the scattered areas in the check-in picture are obtained, and compared with the preset optical feedback threshold to obtain the stitching optional determination result. If a certain scattered area is determined to be qualified, it is marked as an optional qualified scattered area, and if it is determined to be unqualified, it is marked as an optional scattered area. Then, the functional area attribute stitching determination is performed based on the scattered area result, if there is a qualified scattered area in the functional area, the scattered area with the maximum feedback value is directly selected as the preferred area, and if all the functional areas are unqualified areas, the scattered area with the maximum feedback value is selected as the to-be-optimized area. The brightness and exposure of the to-be-optimized area need to be further obtained, and the brightness and exposure of the adjacent preferred area are adjusted to obtain the optimized preferred scattered area. All the preferred scattered areas are stitched to obtain the stitched picture, wherein if the stitched picture is composed of qualified scattered areas, the stitched picture is a second check-in picture, and if the stitched picture contains the optimized area, the stitched picture is a third check-in picture.

[0048] It should be noted that the image stitching processing of different functional areas to obtain a complete face recognition image has a significant advantage in extracting and analyzing the features of different images to calculate the comprehensive similarity compared to not stitching. Without stitching, different functional areas may be scattered in multiple images, for example, the forehead area in the upper left corner and the corner of the mouth area in the lower right corner may exist in different images (here referring to the clear image of the area), which leads to the inability to calculate the Euclidean distance between these far distance feature points on the same image, so that only limited local similarity can be obtained, and the comprehensive similarity evaluation is limited. Through stitching processing, each scattered area is integrated into the same image according to the functional area, so that all key areas are uniformly mapped to the same coordinate system, which can calculate the feature Euclidean distance between any areas on the same image, and obtain more and more complete feature similarity information. The calculation of the comprehensive similarity is more comprehensive, the feature coverage is more sufficient, and the overall facial similarity between different images can be more accurately reflected, thereby significantly improving the accuracy and stability of face recognition. In addition, the stitched image can effectively avoid the recognition error caused by uneven illumination, local occlusion or low-quality area of a single image, improve the stability of the comprehensive similarity of face recognition, and significantly improve the recognition performance and accuracy of attendance and clock-in under different illumination and posture conditions.

[0049] When stitching multiple clock-in pictures (including multiple images continuously taken), due to different shooting angles, part of the images may be side faces and part of the images may be front faces, which will cause the same functional area to appear shape stretching or scaling in different images, thereby affecting the accuracy of subsequent face recognition. Therefore, the stitching correction can be performed based on three-dimensional point cloud data, and the specific process is as follows: first, the face three-dimensional point cloud data corresponding to each picture is obtained by the clock-in device, and the three-dimensional recognition model is used to align each point cloud to a unified face coordinate system; second, the normal vector and depth of the point cloud corresponding to the functional area of each picture are analyzed to determine the degree of deflection or inclination caused by different angles; then, the point cloud of each functional area is standardized by rigid transformation (rotation, translation, scaling) to keep the size and proportion of the functional area at different angles consistent in the unified coordinate system; finally, the corrected two-dimensional image is stitched according to the point cloud mapping position to ensure that each functional area maintains the correct proportion and position in the stitched image, thereby avoiding local enlargement or reduction caused by angle difference and improving the face recognition accuracy of the stitched image.

[0050] By comparing each scattered region of the punch card picture with the optical feedback threshold, and dividing the result into optional qualified scattered regions and alternative scattered regions, the quality screening of the local region can be realized, and then the qualified region is preferentially used in image stitching to ensure that the overall image quality after stitching is significantly better than the original image. Based on the stitching determination result of each functional region, differential optimization processing is performed on scattered regions of different qualities, and after obtaining the preferred scattered region, stitching is performed, which avoids the loss of some face feature due to poor single image quality, and realizes the comprehensive retention of the complete face feature, and ensures the accuracy of subsequent feature extraction.

[0051] In the present application, the difference in optical feedback value during the stitching process is dynamically determined and regionally optimized, which can effectively weaken the influence of uneven illumination, local overexposure or shadow coverage on the overall image recognition effect, realize the stability of the stitched image in different light environments, and improve the recognition robustness. By distinguishing the second punch card image (stitched entirely from qualified scattered regions) and the third punch card image (stitched with alternative scattered regions), the present application can provide clear quality level identification for the subsequent identification link, thereby supporting the identification module to adjust the feature extraction and comparison strategy according to the level of the stitched image, and improving the recognition response speed and success rate of face recognition under different quality conditions. In the prior art, the overall image quality is often not up to standard, resulting in waste of information with good quality in some regions, which affects the effect of face recognition. Through regional level stitching and optimization processing, the present application can generate a usable face image even if there is a local low-quality region in the overall image, effectively improving the image utilization rate and avoiding the problems of repeated punch card and delay caused by invalid collection.

[0052] Further, the attribute stitching determination result of each functional region is obtained, and the specific method is as follows: based on the analysis of each scattered region of each functional region, if the optical feedback value of the scattered region of a certain functional region is an optional qualified scattered region, the attribute stitching determination result of the functional region is qualified, and if each scattered region of a certain functional region is an alternative scattered region, the attribute stitching determination result of the functional region is unqualified.

[0053] In the present embodiment, the present application comprehensively analyzes the optical feedback value of each scattered region at the functional region level, avoids misjudgment of the overall functional region due to abnormal individual scattered region, and realizes more robust quality determination of the functional region, thereby improving the accuracy of the stitching determination. In the present application, part of the scattered regions are allowed to be unqualified in the determination of the functional region, as long as there is a qualified scattered region in the functional region, the functional region is determined to be qualified, and the qualified image is selected for stitching, thereby effectively avoiding the influence of small range optical abnormality on the overall image stitching effect.

[0054] Further, the preferred dispersion region of the punch card picture is obtained by the following method: based on the analysis of the splicing judgment result of the attribute of each functional region, if the splicing judgment result of the attribute of a certain functional region is qualified, the dispersion region corresponding to the maximum optical feedback value of the functional region is selected as the preferred dispersion region of the functional region; if the splicing judgment result of the attribute of a certain functional region is unqualified, the dispersion region corresponding to the maximum optical feedback value of the functional region is selected and marked as a dispersion region to be optimized, and the dispersion region to be optimized is processed to obtain an optimized dispersion region, which is marked as a preferred dispersion region. In this way, the preferred dispersion region of each functional region is obtained, and the processed spliced image is obtained by splicing, which is marked as a third punch card picture.

[0055] In this embodiment, by selecting the dispersion region corresponding to the maximum optical feedback value as the preferred dispersion region when the functional region is qualified, the optimal region selection of the functional region is realized, and the image quality of the spliced image in the key region is always in the best state. When the functional region is unqualified, the dispersion region with the maximum feedback value is not directly discarded, but is converted into a preferred dispersion region by optimization processing, thereby ensuring the integrity of the functional region during splicing and avoiding the problem of insufficient facial feature information caused by local missing. By directly selecting the optimal region for the qualified region and selecting the optimal region after optimization for the unqualified region, the overall consistency of the spliced image can be ensured, the visual discord caused by local over-brightness or over-darkness and the like can be effectively reduced, and the accuracy and stability of subsequent feature extraction are improved. Thus, under complex conditions such as different illuminations, different angles, and even makeup and obstruction, the dispersion region can be dynamically selected or optimized based on the optical feedback value, and a reliable spliced image is formed, thereby significantly enhancing the adaptability and robustness of the system under complex working conditions. Since the spliced image is always composed of preferred dispersion regions, the overall image quality tends to be optimal, thereby improving the accuracy of facial feature extraction, effectively reducing the false recognition rate caused by poor image quality, speeding up the recognition response speed, and improving the punch card efficiency.

[0056] Further, the dispersion region optimization processing is performed, and the specific method is as follows: acquiring each preferred dispersion region adjacent to the dispersion region to be optimized, and marking each adjacent preferred dispersion region as an adjacent preferred dispersion region; acquiring the brightness value and exposure of each adjacent preferred dispersion region, and performing mean value processing to obtain the brightness mean value and exposure mean value of the adjacent preferred dispersion region; and adjusting the brightness and exposure of the dispersion region to be optimized based on the brightness mean value and exposure mean value of the adjacent preferred dispersion region.

[0057] In this embodiment, as shown in FIG. 8, the dispersion region optimization processing is performed on the dispersion region to be optimized, and the specific method is as follows: Figure 5 Figure 5 ​An adjacent area schematic diagram of the intelligent attendance clock-in method based on face recognition provided by the embodiments of the present application is shown in the figure. As shown in the figure, if area A is a to-be-optimized dispersion area, then areas B, C, D and E adjacent to area A are analyzed. If there is a preferred dispersion area in areas B, C, D and E, then the preferred area in areas B, C, D and E is the adjacent preferred dispersion area, thereby obtaining each adjacent preferred dispersion area. If there is no preferred area in areas B, C, D and E, then area A and areas B, C, D and E are regarded as a whole area, marked as area combination ABCDE, and the adjacent areas thereof are analyzed again. At this time, the adjacent areas of the area combination ABCDE are areas F, G, H, I, J, K, L and M, and it is continued to be judged whether there is a preferred area in areas F, G, H, I, J, K, L and M, until the preferred dispersion area is found.

[0058] The brightness and exposure of the to-be-optimized dispersion area are adjusted based on the brightness mean value and the exposure mean value of the adjacent preferred dispersion area. The specific method is as follows: the brightness value of the to-be-optimized dispersion area is adjusted to the brightness mean value of the adjacent preferred dispersion area, and the exposure of the to-be-optimized dispersion area is adjusted to the exposure mean value of the adjacent preferred dispersion area.

[0059] The brightness value and the exposure of each preferred dispersion area adjacent to the to-be-optimized dispersion area are obtained, and the mean value is processed, and then the to-be-optimized dispersion area is adjusted based on this, so that the adaptive brightness and exposure optimization of the dispersion area is realized, the visual effect of the local area is reasonably processed, and the spliced image is more natural on the face structure information, which is beneficial to the subsequent face key point detection and feature extraction.

[0060] Further, a face recognition comprehensive similarity is obtained, and the specific method is as follows: the optimized clock-in picture of the image recognition temporary storage area is extracted and analyzed through a preset face recognition scheme to obtain the face recognition comprehensive similarity. The optimized clock-in picture includes the first clock-in picture, the second clock-in picture and the third clock-in picture. The preset face recognition scheme includes the first face recognition scheme, the second face recognition scheme and the third face recognition scheme. If the optimized clock-in picture is the first clock-in picture, the first face recognition scheme is used to analyze the face recognition features of the first clock-in picture to obtain the face recognition comprehensive similarity. The first face recognition scheme is specifically as follows: the first face recognition features are obtained by performing overall image feature extraction on the first clock-in picture, the pre-stored face recognition feature set in the database is obtained, and the feature similarity analysis is performed on the first face recognition features to obtain the similarity between the first clock-in picture and the face recognition feature set, and the similarity is marked as the face recognition comprehensive similarity. If the optimized clock-in picture is the second clock-in picture, the second face recognition scheme is used to analyze the face recognition features of the second clock-in picture to obtain the face recognition comprehensive similarity. The second face recognition scheme is specifically as follows: the second face recognition features of each optimal dispersed area of the second clock-in picture are obtained by performing feature extraction on each optimal dispersed area, and the feature similarity analysis is performed on the face recognition feature set to obtain the face recognition similarity between each optimal dispersed area and the face recognition feature set. The face recognition similarity between each optimal dispersed area and the face recognition feature set is processed by mean value to obtain the face recognition comprehensive similarity. If the optimized clock-in picture is the third clock-in picture, the third face recognition scheme is used to analyze the face recognition features of the third clock-in picture to obtain the face recognition comprehensive similarity. The third face recognition scheme is specifically as follows: the third face recognition features of each optimal dispersed area of the third clock-in picture are obtained by performing feature extraction on each optimal dispersed area, and the feature similarity analysis is performed on the face recognition feature set to obtain the face recognition similarity between each optimal dispersed area and the face recognition feature set. The face recognition similarity between each optimal dispersed area and the face recognition feature set is processed by mean value to obtain the third area recognition similarity. The third clock-in picture is subjected to overall image feature extraction, and the feature similarity analysis is performed on the face recognition feature set to obtain the similarity between the third clock-in picture and the face recognition feature set, and the similarity is marked as the third overall recognition similarity. The third area recognition similarity and the third overall recognition similarity are coupled and analyzed to obtain the face recognition comprehensive similarity.

[0061] In the embodiment, by adopting the hierarchical face recognition scheme of overall feature extraction, dispersed region feature extraction and region and overall coupling analysis respectively for the optimization of punch card pictures of different qualities, the matching recognition mode can be adaptively selected according to the quality difference of the first punch card picture, the second punch card picture and the third punch card picture, and stable and accurate face recognition comprehensive similarity can be obtained in the case of different image qualities. The method not only ensures fast recognition under high-quality images, but also improves the robustness of recognition through regional feature mean and region-global coupling in the case of low-quality or partial region optimization, effectively improving the accuracy and reliability of face recognition under complex illumination and diversified scenes.

[0062] The face recognition feature analysis is performed through the second face recognition scheme to obtain the face recognition comprehensive similarity, and the specific method is as follows:

[0063] ;

[0064] In the formula, XS represents the face recognition comprehensive similarity, S j represents the face recognition similarity of the jth preferred dispersed region, j represents the number of the preferred dispersed region, j = 1, 2,..., j max max represents the total number of the preferred dispersed region.

[0065] The face recognition feature analysis is performed through the third face recognition scheme to obtain the face recognition comprehensive similarity, and the specific method is as follows:

[0066] ;

[0067] In the formula, XS represents the face recognition comprehensive similarity, S j represents the face recognition similarity of the jth preferred dispersed region, j represents the number of the preferred dispersed region, j = 1, 2,..., j max max represents the total number of the preferred dispersed region, SA represents the third overall recognition similarity, ε1 represents the weighting factor of the third regional recognition similarity, and ε2 represents the weighting factor of the third overall recognition similarity.

[0068] ​​The weighting factor of the third regional recognition similarity and the weighting factor of the third overall recognition similarity can be obtained from the database. For example, the historical third overall recognition similarities stored in the database are obtained, and are compared with the third overall recognition similarity. The historical third overall recognition similarities whose difference with the third overall recognition similarity is within a preset range are marked as historical control third overall recognition similarities. The weighting factors of the historical control third overall recognition similarities are obtained and are processed by mean value to obtain the mean value of the weighting factors of the historical control third overall recognition similarities. Meanwhile, the standard deviation of the weighting factors of the historical control third overall recognition similarities is analyzed to obtain the standard deviation of the weighting factors of the historical control third overall recognition similarities. A preset standard deviation reference value in the database is obtained, and is compared with the standard deviation of the weighting factors of the historical control third overall recognition similarities. If the standard deviation of the weighting factors of the historical control third overall recognition similarities is below the mean value, the mean value of the weighting factors of the historical control third overall recognition similarities is taken as the weighting factor of the third regional recognition similarity. Otherwise, the standard deviation of the weighting factors of the historical control third overall recognition similarities is subtracted from the standard deviation reference value to obtain a standard deviation adjustment coefficient, which is multiplied by the mean value of the weighting factors of the historical control third overall recognition similarities to obtain a standard deviation adjustment value of the weighting factors of the historical control third overall recognition similarities. The mean value of the weighting factors of the historical control third overall recognition similarities is subtracted from the standard deviation adjustment value of the weighting factors of the historical control third overall recognition similarities to obtain an adjusted mean value of the weighting factors, which is taken as the weighting factor of the third regional recognition similarity.

[0069] As shown in Figure 6 , Figure 6 The carding record query schematic diagram of the intelligent attendance carding system based on face recognition provided by the embodiment of the application can query the carding time, carding place, carding time consumption and carding state of the carding personnel, and can also see the carding picture of the carding personnel and the carding personnel picture in the system template.

[0070] As shown in Figure 7 , Figure 8 , Figure 7 The carding identity verification image quality schematic diagram of the intelligent attendance carding system based on face recognition provided by the embodiment of the application, Figure 8The card punching identity verification result schematic diagram of the intelligent attendance card punching system based on face recognition provided by the embodiment of the application can query the quality of each functional area image of the card puncher, and the similarity between the card puncher picture and the card puncher picture in the system template. The recognition duration of the card punching operation, the total time duration of the card punching, and the face recognition verification result, the positioning verification result, the network state, and the server response state during card punching, etc.

[0071] As shown in Figure 4 , Figure 4 The structure schematic diagram of the intelligent attendance card punching system based on face recognition provided by the embodiment of the application, the intelligent attendance card punching system based on face recognition provided by the embodiment of the application comprises a region division module, an optical feedback analysis module, a splicing processing module, and an attendance card punching judgment module. The region division module is used to obtain continuous snapshots when a user performs attendance card punching after receiving a user attendance card punching signal, and obtain card punching pictures by statistics. The card punching pictures are analyzed and divided into regions to obtain each functional region of the card punching pictures. The optical feedback analysis module is used to obtain optical feedback parameters of each dispersed region of the card punching pictures, and analyze to obtain optical feedback judgment results of the card punching pictures. The optical feedback parameters are used to reflect the imaging state of the card punching image. The splicing processing module is used to analyze the optical feedback judgment results of the card punching pictures. If the optical feedback judgment result of a card punching picture is qualified, the card punching picture is stored in an image recognition temporary area. If the optical feedback judgment results of all the card punching pictures of the continuous snapshots are unqualified, image splicing processing is performed, and the processed splicing image is stored in the image recognition temporary area. The attendance card punching judgment module is used to extract the stored image of the image recognition temporary area, and analyze to obtain a face recognition comprehensive similarity. Thus, an attendance card punching judgment result is obtained, and a prompt information is sent.

[0072] In summary, the embodiment can automatically screen or optimize high-quality face images under complex light environments through region division, optical feedback parameter analysis, image splicing processing, and comprehensive similarity calculation of card punching pictures of different angles. Thus, face recognition attendance can be quickly and accurately performed under the influence of light environment, and the problem of slow face recognition response speed caused by the influence of light environment on attendance card punching in the prior art is effectively solved.

[0073] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0074] The present application is described in reference to the flowchart and / or block diagrams of the methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for carrying out the function specified by the flowchart block or blocks.

[0075] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart and / or block diagram block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for carrying out the function specified by the flowchart block or blocks.

[0076] The computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flowchart and / or block diagram block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for carrying out the function specified by the flowchart block or blocks.

[0077] While preferred embodiments of the application have been described, modifications and variations can be apparent to those skilled in the art once aware of the general underlying concepts. Therefore, it is intended that the scope of the application be governed by the following claims and their equivalents.

[0078] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A smart attendance tracking method based on facial recognition, characterized in that, Includes the following steps: S1. After receiving the user's attendance check-in signal, obtain continuous snapshots of the user's attendance check-in, collect the check-in images, analyze the check-in images and divide them into regions to obtain the functional regions of the check-in images. S2. Obtain the optical feedback parameters of each dispersed area of ​​the check-in image, and analyze the optical feedback judgment result of the check-in image. The optical feedback parameters are used to reflect the imaging state of the check-in image. S3. Based on the optical feedback judgment result analysis of the check-in image, if the optical feedback judgment result of a certain check-in image is qualified, the check-in image is stored in the image recognition temporary storage area. If the optical feedback judgment result of all consecutive check-in images is unqualified, image stitching is performed, and the processed stitched image is stored in the image recognition temporary storage area. S4. Extract the stored image from the image recognition temporary storage area, analyze it to obtain the comprehensive similarity of the face recognition, thereby obtaining the attendance check-in judgment result and issuing a prompt message; The specific method for image stitching processing is as follows: The scattered areas of the check-in image are summarized to obtain the scattered areas of each functional area of ​​the check-in image. The optical feedback values ​​of each scattered area in each functional area of ​​the check-in image are compared with the optical feedback threshold to obtain the splicing selection result of each scattered area. If the optical feedback value of a certain scattered area is above the optical feedback threshold, the splicing selection result of the scattered area is qualified and the scattered area is marked as a qualified scattered area. If the optical feedback value of a certain scattered area is less than the optical feedback threshold, the splicing selection result of the scattered area is unqualified and the scattered area is marked as a candidate scattered area. Based on the analysis of each dispersed region, the splicing judgment results of each functional region attribute are obtained. Based on the splicing judgment results of each functional region attribute, the corresponding dispersed region optimization processing is performed to obtain each preferred dispersed region of the check-in image. Then, image splicing processing is performed to obtain the spliced ​​check-in image. If the spliced ​​check-in image is composed entirely of selectable qualified dispersed regions, then the check-in image is the second check-in image. If there are candidate dispersed regions in the spliced ​​check-in image, then the check-in image is the third check-in image.

2. The intelligent attendance and check-in method based on face recognition as described in claim 1, characterized in that: The specific method for obtaining the functional areas of the check-in image is as follows: The check-in image is input into a 3D recognition model. The 3D recognition model performs facial landmark detection on the check-in image and aligns the 3D recognition model with the check-in image based on the PnP pose estimation method. This analysis yields the scattered regions of the check-in image, and each scattered region is numbered. Check-in images with the same number belong to the same functional region, thus statistically determining the functional regions of the check-in image.

3. The intelligent attendance and check-in method based on face recognition as described in claim 1, characterized in that: The specific method for obtaining the optical feedback determination result of the check-in image is as follows: Obtain optical feedback parameters for each dispersed region of the check-in image, including average brightness, shadow coverage, overexposure area ratio, and brightness range. The optical feedback values ​​of each dispersed region of the check-in image are obtained by analyzing the optical feedback parameters of each dispersed region of the check-in image with the preset optical feedback calibration set in the database. The optical feedback calibration set includes average brightness calibration value, shadow coverage calibration value, overexposure area ratio calibration value, and brightness range calibration value; Obtain the preset optical feedback threshold from the database and compare it with the optical feedback values ​​of each dispersed area of ​​the check-in image to obtain the optical feedback judgment result of the check-in image. If the optical feedback values ​​of each dispersed area of ​​a certain check-in image are all above the optical feedback threshold, the optical feedback judgment result of the check-in image is qualified and the check-in image is marked as the first check-in image; otherwise, the optical feedback judgment result of the check-in image is unqualified.

4. The intelligent attendance and check-in method based on face recognition as described in claim 3, characterized in that: The specific method for obtaining the optical feedback values ​​of each dispersed region of the check-in image is as follows: The difference between the mean brightness value and the calibrated mean brightness value is analyzed to obtain the result of the difference determination. The shadow coverage, reflective area ratio, shadow area ratio, and luminance range were compared with the calibration values ​​of shadow coverage, reflective area ratio, shadow area ratio, and luminance range, respectively, to obtain the ratio analysis results. Based on the difference degree determination results and the ratio analysis results, corresponding weighting factors are introduced for coupling processing to obtain the optical feedback values ​​of each dispersed region of the check-in image. The optical feedback values ​​are used to represent the usability of the check-in image when extracting facial features.

5. The intelligent attendance and check-in method based on face recognition as described in claim 1, characterized in that: The specific method for obtaining the attribute splicing determination results of each functional area is as follows: Based on the analysis of the scattered areas of each functional area, if the optical feedback value of a scattered area of ​​a certain functional area is a selectable qualified scattered area, then the attribute splicing judgment result of that functional area is qualified; if all scattered areas of a certain functional area are candidate scattered areas, then the attribute splicing judgment result of that functional area is unqualified.

6. The intelligent attendance and check-in method based on face recognition as described in claim 1, characterized in that: The specific method for obtaining the preferred dispersed regions of the check-in image is as follows: Based on the analysis of the splicing judgment results of each functional area attribute, if the splicing judgment result of a certain functional area attribute is qualified, then the dispersed area corresponding to the maximum optical feedback value of that functional area is selected as the preferred dispersed area of ​​that functional area. If the splicing result of a certain functional area is unqualified, the scattered area corresponding to the maximum optical feedback value of that functional area is selected and marked as the scattered area to be optimized. The scattered area to be optimized is then processed to obtain the optimized scattered area, which is then marked as the preferred scattered area. Thus, the preferred scattered areas of each functional area are obtained and spliced ​​together to obtain the processed spliced ​​image, which is then marked as the third check-in image.

7. The intelligent attendance and check-in method based on face recognition as described in claim 6, characterized in that: The specific method for performing the decentralized region optimization processing is as follows: Obtain each preferred dispersion region adjacent to the dispersion region to be optimized, and mark them as each adjacent preferred dispersion region. Obtain the brightness value and exposure of each adjacent preferred dispersion region, and perform mean processing to obtain the mean brightness value and mean exposure value of the adjacent preferred dispersion regions. The brightness and exposure of the area to be optimized are adjusted based on the average brightness and average exposure of adjacent preferred dispersion areas.

8. The intelligent attendance and check-in method based on face recognition as described in claim 1, characterized in that: The method for obtaining the comprehensive facial recognition similarity is as follows: Optimized check-in images are extracted from the image recognition temporary storage area and analyzed through a preset face recognition scheme to obtain the comprehensive face recognition similarity. The optimized check-in images include a first check-in image, a second check-in image, and a third check-in image. The preset face recognition scheme includes a first face recognition scheme, a second face recognition scheme, and a third face recognition scheme; If the optimized check-in image is the first check-in image, then the first check-in image is subjected to facial recognition feature analysis using the first facial recognition scheme to obtain the comprehensive facial recognition similarity. The first face recognition scheme is as follows: extract the overall image features of the first check-in image to obtain the first face recognition features, obtain the face recognition feature set pre-stored in the database, and perform feature similarity analysis with the first face recognition features to obtain the similarity between the first check-in image and the face recognition feature set, and mark it as the face recognition comprehensive similarity. If the optimized check-in image is the second check-in image, then the second check-in image is subjected to facial recognition feature analysis using the second facial recognition scheme to obtain the comprehensive facial recognition similarity. The second face recognition scheme is as follows: each preferred dispersed region of the second check-in image is subjected to feature extraction to obtain the second face recognition feature of each preferred dispersed region, and feature similarity analysis is performed with the face recognition feature set to obtain the face recognition similarity between each preferred dispersed region and the face recognition feature set. The face recognition similarity between each preferred dispersed region and the face recognition feature set is averaged to obtain the comprehensive face recognition similarity. If the optimized check-in image is the third check-in image, then the third check-in image is subjected to facial recognition feature analysis using a third facial recognition scheme to obtain the comprehensive facial recognition similarity. The third face recognition scheme is as follows: extract features from each preferred dispersed region of the third check-in image to obtain the third face recognition features of each preferred dispersed region, and perform feature similarity analysis with the face recognition feature set to obtain the face recognition similarity between each preferred dispersed region and the face recognition feature set. Then, average the face recognition similarity between each preferred dispersed region and the face recognition feature set to obtain the third region recognition similarity. The overall image features of the third check-in image are extracted and compared with the face recognition feature set to obtain the similarity between the third check-in image and the face recognition feature set, and it is marked as the third overall recognition similarity. By coupling the third region recognition similarity and the third overall recognition similarity, the comprehensive face recognition similarity is obtained.

9. A system applying the intelligent attendance and check-in method based on face recognition as described in any one of claims 1-8, characterized in that, include: The module includes a region division module, an optical feedback analysis module, a stitching processing module, and an attendance check-in / check-out judgment module. The region division module is used to obtain continuous snapshots of the user's attendance check-in after receiving the user's attendance check-in signal, to obtain the check-in images, to analyze the check-in images and divide them into regions, and to obtain the functional regions of the check-in images. The optical feedback analysis module is used to acquire the optical feedback parameters of each dispersed area of ​​the check-in image, and analyze them to obtain the optical feedback judgment result of the check-in image. The optical feedback parameters are used to reflect the imaging state of the check-in image. The stitching processing module is used to analyze the optical feedback judgment result based on the check-in image. If the optical feedback judgment result of a certain check-in image is qualified, the check-in image is stored in the image recognition temporary storage area. If the optical feedback judgment result of all consecutive check-in images is unqualified, the image stitching processing is performed, and the processed stitched image is stored in the image recognition temporary storage area. The attendance check-in determination module is used to extract the stored image in the image recognition temporary storage area, analyze it to obtain the comprehensive similarity of the face recognition, thereby obtaining the attendance check-in determination result and issuing a prompt message.

Images