An integrated employment service machine and method supporting both facial recognition and ID card login

By adaptively adjusting the face comparison weight and dynamically adjusting the authentication threshold, the problem of high false rejection rates caused by unstable lighting and outdated ID photos in the gig market and gig work stations has been solved, achieving efficient authentication in harsh environments.

CN123116248APending Publication Date: 2026-11-17LIUPANSHUI ZHUOCHENG INTELLIGENT TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611581663.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-13
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In public employment service venues such as gig markets and gig work stations, existing facial recognition and ID card verification technologies have a high false rejection rate under uncontrolled lighting and outdated ID photos, making it difficult to balance authentication pass rate and security.

Method used

An ambient light perception module and a face quality assessment module are used to obtain the illumination influence factor and the comprehensive face quality score. The face comparison weight is adaptively adjusted, and the authentication threshold is adjusted in real time through a dynamic threshold generation module. The face comparison and ID card authentication results are integrated.

Benefits of technology

It significantly improved the authentication pass rate in scenarios with poor lighting and outdated ID photos, while maintaining necessary security and reducing the risk of impersonation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116248A_ABST
    Figure CN123116248A_ABST
Patent Text Reader

Abstract

This invention relates to an integrated employment service machine and method supporting both facial recognition and ID card login. The machine includes an ambient light sensing module, a facial recognition and quality assessment module, an ID card reading and photo parsing module, a facial comparison module, an adaptive weight calculation module, a fusion authentication scoring module, a dynamic threshold generation module, and a decision and output module. The adaptive weight calculation module generates facial comparison weights based on the illumination influence factor and the comprehensive facial quality score, and subtracts the facial comparison weights from 1 to obtain the quality item weights. As the product of the illumination influence factor and the comprehensive facial quality score increases, the facial comparison weights increase from a preset lower limit to an upper limit. This invention solves the problem of low dual-modal authentication pass rates caused by changes in illumination and aging ID card photos in uncontrolled environments such as gig work stations. It enables the authentication trust level to adaptively adjust with the collection environment, improving the reliability of age-appropriate identity verification and user experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of public employment service terminal technology, specifically to a touch-screen all-in-one machine that supports dual-modal login using facial recognition and ID card, and its adaptive authentication method. Background Technology

[0002] Touchscreen kiosks deployed in public employment service locations such as gig economy markets and gig work stations need to provide convenient identity authentication channels for a group primarily composed of older workers. These users generally face difficulties operating smartphones and easily forget passwords; therefore, the kiosks typically offer both facial recognition and ID card login. Existing solutions generally employ a sequential process of "facial recognition first, then ID card verification if that fails," or require both facial recognition and ID card reading to pass. However, gig work stations are often located in temporary prefabricated houses, market sheds, or outdoor sunshades, where light intensity varies drastically throughout the day. Strong backlighting or insufficient illumination is common in the morning, evening, and on rainy days, leading to a sharp drop in facial image quality. Furthermore, older users' ID cards are often issued ten or even twenty years ago, resulting in limited resolution photos and noticeable changes in facial features due to natural aging. With the combined disadvantages of uncontrolled lighting and outdated ID photos, the false rejection rate of face matching with a fixed threshold increases sharply. Legitimate users cannot pass the authentication even after repeated attempts, while lowering the threshold rashly will introduce the risk of impersonation.

[0003] Therefore, there is an urgent need for an all-in-one solution that can dynamically adjust the dual-modal authentication trust mechanism based on the actual lighting conditions of the acquisition environment and the quality of the face on site. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes an integrated employment service machine supporting both face and ID card login, comprising: an ambient light perception module electrically connected to an adaptive weight calculation module, used to acquire the ambient light intensity of the environment in which the machine is located, and generate a light influence factor based on the ambient light intensity; a face acquisition and quality assessment module electrically connected to the adaptive weight calculation module and the fusion authentication scoring module, used to acquire the user's on-site face image and calculate a comprehensive face quality score; an ID card reading and photo parsing module electrically connected to a face comparison module, used to read the ID card photo stored in the user's ID card and extract the ID card photo features; a face comparison module electrically connected to the fusion authentication scoring module, used to compare the features of the on-site face image with the features of the ID card photo and output the original similarity; and an adaptive weight calculation module electrically connected to the fusion authentication scoring module, used to calculate the light influence factor based on the ambient light intensity. The system uses an influence factor and the comprehensive face quality score to generate a face comparison weight. Subtracting this weight from 1 yields a quality item weight. The face comparison weight increases from a preset lower limit to a preset upper limit as the product of the illumination influence factor and the comprehensive face quality score increases. A fusion authentication scoring module, electrically connected to the dynamic threshold generation module and the decision and output module, calculates a comprehensive authentication score based on the face comparison weight, the quality item weight, the original similarity, and the comprehensive face quality score. A dynamic threshold generation module, also electrically connected to the decision and output module, obtains the standard deviation of historical authentication score sequences and generates an adjusted judgment threshold based on the standard deviation. The decision and output module compares the comprehensive authentication score with the adjusted judgment threshold and allows login when the comprehensive authentication score is greater than or equal to the adjusted judgment threshold.

[0005] Preferably, the ambient light sensing module includes a light sensor and a light influence factor calculation unit. The light sensor is installed on the bezel of the all-in-one screen and is used to collect the ambient light intensity. The light influence factor calculation unit is used to calculate the light influence factor based on the ambient light intensity using the sigmoid function. The value range of the light influence factor is 0 to 1.

[0006] Preferably, the face acquisition and quality assessment module includes a camera and an image processing unit. The image processing unit is used to extract the clarity score and occlusion rate of the face region, and to perform a weighted summation of the clarity score and the occlusion rate to obtain the comprehensive face quality score.

[0007] Preferably, the adaptive weight calculation module generates the face comparison weights by: calculating the product of the illumination influence factor and the comprehensive face quality score; performing an exponential operation on the product, where the exponent is a preset coupling index; and linearly transforming the operation result to between the preset lower limit and the preset upper limit to obtain the face comparison weights.

[0008] Preferably, the preset lower limit is 0.3, the preset upper limit is 0.9, and the preset coupling index is 2; the ambient light sensing module is based on a reference critical illuminance of 200 lux and a steepness coefficient of 0.01 lux. -1 The illumination influence factor is generated; in the face acquisition and quality assessment module, the weight of the clarity score is 0.6 and the weight of the occlusion rate is 0.4; the dynamic threshold generation module is used to generate the adjusted judgment threshold with a base threshold of 0.6 and an adjustment coefficient of 0.8.

[0009] Preferably, the ambient light illuminance value obtained by the ambient light sensing module ranges from 0 lux to 1000 lux.

[0010] Preferably, the historical authentication score sequence obtained by the dynamic threshold generation module is composed of the comprehensive authentication scores of the users who have been authenticated on the same day by the all-in-one machine, arranged in chronological order of authentication time.

[0011] Preferably, when the comprehensive authentication score is less than the adjusted judgment threshold, the decision and output module triggers the all-in-one machine display screen to display the SMS verification code input interface, or sends a manual verification request to the on-site management terminal.

[0012] Preferably, an employment service authentication method supporting login via face and ID card is executed by an all-in-one machine. The all-in-one machine has an ambient light sensing module, a face acquisition and quality assessment module, an ID card reading and ID photo parsing module, a face comparison module, an adaptive weight calculation module, a fusion authentication scoring module, a dynamic threshold generation module, and a decision and output module. The method includes: S1: The ambient light sensing module acquires the ambient light intensity and generates an illumination influence factor based on the ambient light intensity;

[0013] S2: The face acquisition and quality assessment module acquires on-site face images and calculates a comprehensive face quality score;

[0014] S3: The ID card reading and photo parsing module reads the ID card photo and extracts the photo features;

[0015] S4: The face comparison module compares the on-site facial features with the features of the ID photo to obtain the original similarity.

[0016] S5: The adaptive weight calculation module generates face comparison weights based on the illumination influence factor and the comprehensive face quality score, and subtracts the face comparison weights from 1 to obtain the quality item weights. When the product of the illumination influence factor and the comprehensive face quality score increases, the face comparison weights increase from a preset lower limit to a preset upper limit.

[0017] S6: The fusion authentication scoring module calculates the comprehensive authentication score based on the face comparison weight, the quality item weight, the original similarity, and the comprehensive face quality score;

[0018] S7: The standard deviation of the historical authentication score sequence is obtained by the dynamic threshold generation module, and an adjusted judgment threshold is generated based on the standard deviation;

[0019] S8: The decision and output module compares the comprehensive authentication score with the adjusted judgment threshold. Login is allowed when the comprehensive authentication score is greater than or equal to the adjusted judgment threshold.

[0020] Preferably, the preset lower limit is 0.3, the preset upper limit is 0.9, and the preset coupling index is 2; the light influence factor is generated based on a reference critical illuminance of 200 lux and a steepness coefficient of 0.01 lux. -1 In the calculation of the comprehensive face quality score, the weight of the clarity score is 0.6 and the weight of the occlusion rate is 0.4. In the dynamic threshold generation step, the base threshold used to generate the adjusted judgment threshold is 0.6 and the adjustment coefficient is 0.8.

[0021] Technical effects:

[0022] The all-in-one machine of this invention obtains the illumination influence factor and the comprehensive face quality score through an ambient light perception module and a face quality assessment module, respectively. An adaptive weight calculation module, driven by the product of these two scores, continuously varies the face comparison weight, which represents the trust level of face comparison, between 0.3 and 0.9. Simultaneously, 1 is subtracted from this weight to obtain the quality item weight. This automatically reduces reliance on old ID photos for comparison in poor lighting or when faces are severely occluded, instead relying more on the observability of the face on-site for authentication scoring. A dynamic threshold generation module adjusts the judgment threshold in real time using the standard deviation of the comprehensive scores of authenticated users for the day, automatically tightening the threshold when there are large fluctuations in the user group, further suppressing the risk of impersonation. Therefore, the all-in-one machine significantly improves the authentication pass rate while maintaining necessary security in typical scenarios such as uncontrolled lighting and old ID photos of older users. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a block diagram showing the module connection of the employment service all-in-one machine that supports both facial recognition and ID card login according to the present invention.

[0025] Figure 2 This is a flowchart illustrating the steps of the employment service authentication method supporting facial recognition and ID card login according to the present invention.

[0026] Figure 3 This is a graph showing the change of the light influence factor with ambient light intensity in this invention.

[0027] Figure 4 This is a graph showing how the face comparison weight changes with the product of the illumination influence factor and the comprehensive face quality score in this invention.

[0028] Figure 5 This is a comparison chart of the changes in the comprehensive authentication score and the adjusted judgment threshold over time in this invention. Detailed Implementation

[0029] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0030] It should be understood that, when used in this application specification and the appended claims, the term includes indicating the presence of the described feature, integral, step, operation, element and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0031] Please see Figures 1-5 This implementation method is applied to the gig economy station scenario, where the existing dual-modal authentication method is difficult to balance the false rejection rate and security risks.

[0032] The employment service all-in-one machine supporting both face and ID card login includes an ambient light sensing module, a face acquisition and quality assessment module, an ID card reading and photo parsing module, a face comparison module, an adaptive weight calculation module, a fusion authentication scoring module, a dynamic threshold generation module, and a decision and output module. The output of the ambient light sensing module is connected to the input of the adaptive weight calculation module. The ambient light sensing module collects the ambient light illuminance value at the location of the all-in-one machine, generates a light influence factor based on the illuminance value, and outputs it to the adaptive weight calculation module. The output of the face acquisition and quality assessment module is connected to the inputs of both the adaptive weight calculation module and the fusion authentication scoring module. The face acquisition and quality assessment module collects the face image of the user on-site, processes the face image to obtain a comprehensive face quality score, and outputs it to both the adaptive weight calculation module and the fusion authentication scoring module. The output of the ID card reading and photo parsing module is connected to the input of the face comparison module. The ID card reading and photo parsing module reads the ID card photo image stored in the ID card chip, extracts the facial feature vector from the ID card photo image, and outputs it to the face comparison module. The output of the face comparison module is connected to the input of the fusion authentication and scoring module. The face comparison module extracts the feature vector of the face image at the scene, calculates the similarity between the feature vector of the face at the scene and the feature vector of the ID photo, obtains the original similarity, and outputs it to the fusion authentication and scoring module. The output of the adaptive weight calculation module is connected to the input of the fusion authentication and scoring module. The adaptive weight calculation module receives the illumination influence factor and the comprehensive face quality score, calculates the face comparison weight and the quality item weight based on the two input values, and outputs them to the fusion authentication and scoring module. The output of the fusion authentication and scoring module is connected to the input of the dynamic threshold generation module and the decision and output module. The fusion authentication and scoring module receives the face comparison weight, the quality item weight, the original similarity, and the comprehensive face quality score, calculates the weighted average to obtain the comprehensive authentication score, and outputs it to the dynamic threshold generation module and the decision and output module, respectively. The output of the dynamic threshold generation module is connected to the input of the decision and output module. The dynamic threshold generation module obtains the historical authentication score sequence, calculates the standard deviation of the sequence, generates an adjusted judgment threshold based on the standard deviation, and outputs it to the decision and output module. The decision and output module receives the comprehensive authentication score and the adjusted judgment threshold, compares the two values, and outputs a login permission or login denial instruction based on the comparison result.

[0033] The ambient light sensing module includes a light sensor and a light influence factor calculation unit. The light sensor is mounted on the upper edge of the all-in-one screen bezel, with its photosensitive surface facing the user's standing direction and at the same angle to the screen plane. The output of the light sensor is connected to the input of the light influence factor calculation unit. The light sensor collects the current ambient illuminance value at a fixed frequency and outputs an analog or digital signal in lux to the calculation unit. The light influence factor calculation unit has a built-in processor and storage unit. The storage unit stores the calculation logic of the sigmoid function and two sets of parameters: a reference critical illuminance and a steepness coefficient. After receiving the illuminance value, the calculation unit substitutes the illuminance value into the sigmoid function for calculation. The result is a light influence factor with a value between 0 and 1. The physical meaning of the light influence factor is how favorable the current ambient lighting conditions are for face imaging. The closer the value is to 1, the more suitable the lighting conditions are for face capture; the closer the value is to 0, the worse the lighting conditions are. The sigmoid function's curve characteristics ensure that the light impact factor exhibits a continuous and smooth transition with changes in illuminance, without abrupt changes at any critical illuminance point. The light impact factor calculation unit outputs the calculation results to the adaptive weight calculation module via the data bus for subsequent weight allocation calculations.

[0034] Light Influence Factor The calculation formula is:

[0035]

[0036] in the formula The ambient illuminance, measured in lux, is obtained in real-time by the light sensor at a preset sampling frequency. The light sensor employs a silicon-based photodiode structure, with a photosensitive spectrum covering the visible light band and a measurement range from 0 lux to 1000 lux. Within this range, the output signal maintains a linear correlation with the actual illuminance. The sampling frequency is set to 5 times per second. After continuously acquiring 5 samples, the arithmetic mean is taken as the effective illuminance value at the current moment for calculation. The purpose of averaging is to suppress the interference of instantaneous changes in light and shadow in the environment on the measurement results. The reference critical illuminance is 200 lux. This value is derived from statistical analysis of a large amount of on-site measurement data from various workstations. During the statistical process, environmental illuminance samples were collected under different time periods and weather conditions. The error rate of the face recognition model under the corresponding illuminance conditions was recorded simultaneously. After plotting the curve of the error rate versus illuminance, the illuminance corresponding to the inflection point where the slope of the curve changes significantly is 200 lux. Below 200 lux, the error rate increases as illuminance decreases. Above 200 lux, the error rate tends to level off as illuminance increases and remains at a low level. Therefore, 200 lux was selected as the reference critical illuminance to distinguish between illuminance ranges that are conducive to imaging and those that are not. This represents the steepness coefficient, with a value of 0.01 per lux. This parameter controls the slope of the sigmoid function curve in the transition region. A larger steepness coefficient results in a narrower transition region and a more dramatic change in the curve near the critical point; a smaller steepness coefficient results in a wider transition region and a smoother curve. When a value of 0.01 per lux is selected, the light influence factor is considered within the range of illuminance increasing from 100 lux to 300 lux. The value increased from approximately 0.27 to approximately 0.73, with the entire transition range covering an illuminance range of 200 lux. This aligns with the actual characteristics of gradual illuminance changes in a gig economy environment, avoiding significant jumps in the light impact factor due to minor fluctuations in illuminance. In the formula... is a natural constant with a value of approximately 2.71828, and the exponent is... When the ambient light Illuminance greater than the reference critical level When the exponent is negative and the denominator is less than 2, the calculation result is... Greater than 0.5, when the ambient illuminance Less than the reference critical illuminance When the exponent is positive and the denominator is greater than 2, the calculation result is... Less than 0.5. When the illuminance value is much greater than 200 lux, the absolute value of the exponent increases, and the denominator approaches 1. The denominator approaches 1; when the illuminance value is much less than 200 lux, the denominator increases. Approaching 0. The output of the entire formula. It is a dimensionless numerical value, with its range strictly limited to an open interval between 0 and 1. It represents the imaging friendliness of the current ambient illuminance relative to the critical value, and provides a continuously differentiable input factor for subsequent adaptive weight calculation.

[0037] The face capture and quality assessment module includes a camera and an image processing unit. The camera is mounted centered above the all-in-one screen, with its lens optical axis aligned with the screen normal. The camera uses a visible light imaging sensor with a resolution meeting the minimum requirements for facial feature extraction. The camera's output is connected to the image processing unit's input. The camera captures a sequence of video frames of the user's face, selecting one or more frames for transmission to the image processing unit. The image processing unit incorporates face detection, sharpness calculation, and occlusion detection algorithms. The image processing unit first performs face region detection on the input image, locating the rectangular region containing the face and cropping it to obtain a sub-image of the face region. The image processing unit calculates a sharpness score for the face region sub-image. This score is obtained using the Laplacian gradient mean method. Specifically, the Laplacian operator convolution is performed on the face region sub-image to obtain a gradient image. The mean grayscale value of all pixels in the gradient image is calculated, and this mean is normalized to obtain the sharpness score. The score ranges from 0 to 1; a value closer to 1 indicates richer facial texture and a sharper image, while a value closer to 0 indicates a blurrier image. Simultaneously, the image processing unit performs occlusion detection on the face region sub-image. A semantic segmentation model is used to identify occluding objects such as masks, hats, sunglasses, and handkerchiefs within the face region. The proportion of occluded pixels to the total number of pixels in the face region is calculated to obtain an occlusion rate value, ranging from 0 to 1. A value of 0 indicates no occlusion, and a value of 1 indicates the face is completely occluded. The image processing unit then weights and sums the sharpness score and occlusion rate according to preset weights to obtain a comprehensive face quality score, ranging from 0 to 1. The comprehensive face quality score is output to the adaptive weight calculation module and the fusion authentication score module via the data bus, respectively, for weight calculation and fusion score calculation.

[0038] Overall facial quality score The calculation formula is:

[0039]

[0040] in the formula The sharpness score, ranging from 0 to 1, is obtained by the image processing unit performing a Laplacian gradient operation on the face region sub-image and then normalizing it. The Laplacian operator uses a 3×3 convolution kernel, with the center element having a value of 8 and the eight surrounding elements having values ​​of -1. The convolution operation iterates through each pixel in the face region sub-image, calculating the sum of the grayscale differences between the center pixel and its eight neighboring pixels to obtain the gradient response value for that pixel. The arithmetic mean of the gradient response values ​​of all face region pixels is obtained. The gradient mean reflects the richness of edge texture in the face image; the sharper the image and the more edge details, the larger the gradient mean; conversely, the blurrier the image and the smoother the edges, the smaller the gradient mean. The normalization process employs a piecewise linear mapping method. Pre-defined lower and upper reference values ​​for the gradient mean are used. When the actual gradient mean is lower than the lower reference value, the sharpness score is mapped to 0; when the actual gradient mean is higher than the upper reference value, the sharpness score is mapped to 1. Gradient means between the upper and lower limits are mapped linearly to the 0-1 range. The lower and upper reference values ​​are determined statistically by collecting face sample images with different degrees of blur, ensuring that the mapped sharpness score can distinguish between different levels of blur. In the formula... The occlusion rate, ranging from 0 to 1, is obtained by the image processing unit using a semantic segmentation model. The semantic segmentation model classifies each pixel in the face region into two categories: facial skin regions and occluding object regions. The number of pixels classified as occluding objects is counted and divided by the total number of pixels in the face region to obtain the occlusion rate. The semantic segmentation model is trained on a face dataset containing various occlusions and can identify common occlusion types, covering various scenarios that users at gig work stations might encounter, such as wearing masks, hats, sunglasses, or holding objects that obscure their faces. In the formula... This represents the unobstructed proportion, which is the percentage of the face that is not obscured. This value also ranges from 0 to 1; the lower the occlusion rate, the higher the unobstructed proportion. In the formula... The weighting factor for the sharpness score is 0.6. The weighting coefficient representing the unoccluded proportion is 0.4, and the sum of the two weighting coefficients is 1, ensuring that the weighted sum remains within the range of 0 to 1. The weighting coefficients are determined based on the degree of influence of the two types of factors on the accuracy of face recognition. The impact of decreased sharpness on feature extraction accuracy is slightly greater than that of partial occlusion; therefore, the weighting coefficient for sharpness is set higher than that for the unoccluded proportion. Formula output results. The facial quality score is dimensionless and ranges from 0 to 1. It comprehensively reflects the reliability of the facial image collected on-site as a source of authentication information. The higher the score, the more suitable the facial image is for identity comparison. The lower the score, the worse the facial image quality and the lower the credibility of the comparison result.

[0041] The ID card reading and photo parsing module includes an ID card reader and a photo feature extraction unit. The ID card reader is installed in the reading area at the bottom of the all-in-one machine. The reader uses contactless radio frequency identification (RFID) technology, and its operating frequency conforms to the resident ID card chip reading standard. The output of the ID card reader is connected to the input of the photo feature extraction unit. When the user places their ID card in the reading area, the reader establishes a communication connection with the ID card chip and reads the personal information data stored in the chip, including the photo image data. The reader then transmits the read photo image data to the photo feature extraction unit. The photo feature extraction unit incorporates a face detection algorithm and a feature extraction algorithm. First, it locates the face region in the photo image, crops out a sub-image of the face, and then inputs this sub-image into a deep learning feature extraction model, outputting a fixed-dimensional feature vector. The photo feature extraction unit outputs the extracted feature vector to the face comparison module via a data bus for on-site similarity comparison between the face and the ID photo. The ID photo image is stored inside the ID card chip and written by the public security authority when it is issued. The image format and resolution follow a unified standard, and the ID photo image specifications are consistent across different ID cards.

[0042] The face comparison module includes an on-site face feature extraction unit and a similarity calculation unit. The input of the on-site face feature extraction unit is connected to the image output of the face acquisition and quality assessment module. The on-site face feature extraction unit receives the on-site face region image output by the face acquisition and quality assessment module and extracts the feature vector of the on-site face using the same deep learning model as the ID photo feature extraction. The dimension of the feature vector is consistent with that of the ID photo feature vector. The two inputs of the similarity calculation unit are connected to the output of the on-site face feature extraction unit and the output of the ID card reading and ID photo parsing module, respectively. The similarity calculation unit receives the on-site face feature vector and the ID photo feature vector and calculates the cosine similarity between the two vectors. The cosine similarity is calculated by dividing the dot product of the two vectors by the product of their magnitudes. The result ranges from -1 to 1; the closer the value is to 1, the closer the directions of the two feature vectors are, and the higher the similarity of the corresponding faces. The similarity calculation unit performs a linear mapping on the cosine similarity result, mapping the range of -1 to 1 to the interval of 0 to 1, obtaining the original similarity value. The original similarity score is a dimensionless value. The closer the value is to 1, the higher the similarity between the face at the scene and the face in the ID photo; the closer the value is to 0, the lower the similarity. The face comparison module outputs the calculated original similarity score to the fusion authentication scoring module via the data bus, where it participates in the weighted calculation of the comprehensive authentication score.

[0043] The adaptive weight calculation module receives the illumination influence factor from the ambient light perception module and the comprehensive face quality score from the face acquisition and quality assessment module. Based on these two input values, it calculates the face comparison weight and the quality item weight. First, the adaptive weight calculation module calculates the product of the illumination influence factor and the comprehensive face quality score. The product value reflects the overall credibility of the face comparison information source under the current acquisition conditions. Better lighting conditions and higher face quality result in a larger product value, making the face comparison result more trustworthy. Conversely, poorer lighting conditions or lower face quality result in a smaller product value, making the face comparison result less credible. The adaptive weight calculation module performs an exponential operation on the product result. The exponent is a preset coupling exponent. The exponential operation performs a non-linear transformation on the product value, resulting in slower weight growth when the product is in a lower range and faster weight growth when the product is in a higher range. The adaptive weight calculation module then linearly maps the exponential operation result to a preset lower and upper weight limit, obtaining the final face comparison weight value. The face comparison weight always stays between its lower and upper limits, ensuring it doesn't exceed these boundaries. This guarantees that the face comparison information source retains a certain level of participation even under extremely harsh conditions, and that the quality item information source retains a certain level of participation even under extremely ideal conditions. The adaptive weight calculation module subtracts the face comparison weight from 1 to obtain the quality item weight. The sum of the face comparison weight and the quality item weight is always equal to 1, providing a normalized weight coefficient for subsequent weighted fusion. The adaptive weight calculation module outputs the face comparison weight and the quality item weight to the fusion authentication scoring module.

[0044] Face comparison weight The calculation formula is:

[0045]

[0046] in the formula This indicates the lower limit of the face comparison weight, which is 0.3. This represents the upper limit of the face comparison weight, with a value of 0.9. Setting the lower and upper limits ensures that both information sources retain a minimum participation ratio in the fusion calculation. The face comparison weight will not be lower than 0.3, and the corresponding quality item weight will not be higher than 0.7; the face comparison weight will not be higher than 0.9, and the corresponding quality item weight will not be lower than 0.1. Setting weight boundaries aims to prevent a single information source from completely dominating the authentication result and to avoid authentication failure under extreme conditions. When both ambient lighting and face quality are at their worst, the face comparison weight takes the minimum value of 0.3, and the quality item weight takes the maximum value of 0.7. The authentication decision relies more on the face quality score than on low-confidence comparison results. When both ambient lighting and face quality are at their best, the face comparison weight takes the maximum value of 0.9, and the quality item weight takes the minimum value of 0.1. The authentication decision mainly relies on high-confidence face comparison results. In the formula... The illumination influence factor is calculated by the ambient light sensing module. The value represents the overall face quality score, which is calculated by the face acquisition and quality assessment module. Both values ​​range from 0 to 1. This represents the product of the illumination impact factor and the overall face quality score. The product value also ranges from 0 to 1. The product value comprehensively reflects the overall reliability of the face comparison information source under the current acquisition conditions. A decrease in either factor's value will lead to a decrease in the product value; a decrease in both factors simultaneously will result in a greater decrease in the product value. In the formula... The coupling index, with a value of 2, controls the sensitivity of the weights to changes in the product. A larger index value indicates a more pronounced non-linear relationship between the weights and the product. When the coupling index is 2, the product... With weight The relationship between them exhibits a quadratic function relationship. When When the value is 0, =0, face comparison weight It equals the lower limit of 0.3. When When the value is 0.5, Equal to 0.25, substituting into the formula yields... This equals 0.3 plus 0.6 multiplied by 0.25, resulting in 0.45. At this point, the face comparison weight is at a moderately low level, while the weight of the quality item is relatively high. When When the value is 1, =1, face comparison weight This equals the upper limit of 0.9. A coupling index of 2 keeps the face comparison weight low when the product value is small, giving greater decision weight to the quality item. Only when both lighting conditions and face quality are at a good level will the face comparison weight increase significantly, fully trusting the face comparison results. This non-linear weight allocation mechanism meets the needs of authentication security strategies, making conservative decisions under poor acquisition conditions, relying more on quality assessment results to reduce the risk of misjudgment, and fully utilizing comparison information to improve the pass rate under excellent acquisition conditions. (Formula output results) The weights for face comparison are dimensionless, ranging from 0.3 to 0.9, while the weights for the quality items are determined by... The calculated value ranges from 0.1 to 0.7.

[0047] The fusion authentication scoring module receives the original similarity from the face comparison module, the comprehensive face quality score from the face acquisition and quality assessment module, and the face comparison weight and quality item weight from the adaptive weight calculation module. The fusion authentication scoring module calculates the comprehensive authentication score using a weighted summation method. It multiplies the face comparison weight by the original similarity to obtain the face comparison score, and multiplies the quality item weight by the comprehensive face quality score to obtain the quality item score. The two scores are then added together to obtain the comprehensive authentication score. The comprehensive authentication score ranges from 0 to 1; a higher value indicates a greater likelihood of successful authentication, while a lower value indicates a greater likelihood of questionable user identity. The fusion authentication scoring module outputs the calculated comprehensive authentication score in two paths: one path is output to the decision and output module for current authentication determination, and the other path is output to the dynamic threshold generation module for updating the historical authentication score sequence. The weighted fusion mechanism allows the proportions of the two components in the overall authentication score to be dynamically adjusted according to the collection conditions. When the collection conditions are good, the proportion of the face comparison item is higher, and the overall score mainly reflects the degree of face similarity. When the collection conditions are poor, the proportion of the quality item increases, and the overall score reflects the quality status of the face image on site more.

[0048] The formula for calculating the overall certification score is:

[0049] S core =α··S raw +(1-α)·Q

[0050] in the formula The value represents the face comparison weight, which is calculated by the adaptive weight calculation module based on the illumination influence factor and the comprehensive face quality score, and ranges from 0.3 to 0.9. The original similarity score is calculated by the face comparison module using the cosine similarity between the features of the face at the scene and the features of the ID photo, and then mapped. The value ranges from 0 to 1. The original similarity score reflects the degree of similarity between the face image at the scene and the face in the ID photo, and is a core basis for identity authentication. In the formula... This represents the quality item weight, with a value ranging from 0.1 to 0.7. The sum of the quality item weight and the face comparison weight is always equal to 1. The overall face quality score, calculated by the face acquisition and quality assessment module, ranges from 0 to 1. This score reflects the recognizability of the face image acquired on-site. (Formula 1) The second term in the formula is for face comparison, which is the face comparison weight multiplied by the original similarity. The value of this term is determined by both the comparison similarity and the comparison weight. Higher similarity and a larger weight result in a higher value for this term, while lower similarity or a smaller weight results in a lower value. The quality item is calculated by multiplying its weight by the overall face quality score. The value of this item is determined by both the face quality and its weight; higher quality and a larger weight result in a higher value, and vice versa. The overall authentication score S is obtained by adding the two together. core These are dimensionless numerical values, ranging from 0 to 1. When data acquisition conditions are favorable, the face comparison weight... The weight of the quality item is close to the upper limit of 0.9. When the score is close to the lower limit of 0.1, the overall authentication score is mainly determined by the face comparison item. At this point, the authentication decision is primarily based on facial similarity, which aligns with the authentication logic under normal lighting conditions. However, when the data collection conditions are poor, the weight of face comparison decreases. The weight of the quality item decreased to near the lower limit of 0.3. As the score rises to near the upper limit of 0.7, the contribution of the quality factor to the overall authentication score increases. At this point, even if the original similarity value fluctuates significantly, the overall authentication score will be influenced by the quality factor, preventing erroneous authentication results due to low-quality images. The role of the quality factor in the fusion calculation is to correct comparison results under low-quality acquisition conditions. For users with outdated ID photos, facial similarity may fluctuate abnormally in poor lighting conditions. The facial quality score objectively reflects the current acquisition status. By dynamically increasing the weight of the quality factor, the impact of abnormal comparison results on the final authentication score can be reduced. The overall authentication score, as the final authentication criterion, integrates both identity similarity information and acquisition quality information. The weights of these two information sources adaptively adjust according to environmental conditions, balancing authentication pass rate and authentication security.

[0051] The dynamic threshold generation module includes a score sequence storage unit, a standard deviation calculation unit, and a threshold generation unit. The score sequence storage unit stores the comprehensive authentication scores of users who have completed authentication that day. The data is arranged in chronological order of authentication completion, and the corresponding comprehensive authentication score is appended to the end of the sequence each time authentication is completed. The storage space of the score sequence storage unit has an upper limit. When the sequence length reaches the upper limit, the oldest score data is discarded to accommodate new score data, ensuring that the storage capacity remains within a controllable range. The input of the standard deviation calculation unit is connected to the output of the score sequence storage unit. The standard deviation calculation unit reads all historical authentication score data at the current moment and calculates the standard deviation value of the score sequence. The standard deviation reflects the dispersion of historical authentication scores. A larger standard deviation value indicates a wider fluctuation range of authentication scores for the day, suggesting a higher complexity of the authentication environment or greater differences in the user group, thus increasing the risk of impersonation. A smaller standard deviation value indicates a more concentrated authentication score for the day, suggesting a stable authentication environment or consistent user group characteristics, thus reducing the risk of impersonation. The input of the threshold generation unit is connected to the output of the standard deviation calculation unit. The threshold generation unit reads the standard deviation value and adds an increment proportional to the standard deviation to the base threshold to calculate the adjusted judgment threshold. The adjusted judgment threshold dynamically changes with the daily certification environment. When the environment complexity increases, the threshold increases to raise the security judgment standard; when the environment complexity decreases, the threshold decreases to maintain a reasonable pass rate level. The dynamic threshold generation module outputs the adjusted judgment threshold to the decision and output module for comparison in the current certification.

[0052] Adjusted judgment threshold The calculation formula is:

[0053]

[0054] in the formula The base threshold, set to 0.6, is the authentication decision threshold under standard environmental conditions, corresponding to the score value at the equal error rate equilibrium point. The equal error rate refers to the threshold position corresponding to the point where the false acceptance rate and false rejection rate are equal. Under standard lighting conditions and standard user group conditions, the equal error rate equilibrium point for the overall authentication score corresponds to a value of 0.6; therefore, the base threshold is set to 0.6. The base threshold is determined by collecting a large number of authentication samples from both legitimate and unauthorized users, statistically analyzing the distribution of the overall authentication scores of the two types of samples, and plotting curves showing the change in false acceptance rate and false rejection rate with the threshold. The threshold corresponding to the intersection of the two curves is the equal error rate threshold. In the formula... This represents the adjustment coefficient, with a value of 0.8. The adjustment coefficient controls the magnitude of the threshold's response to changes in standard deviation. A larger adjustment coefficient results in a greater threshold adjustment for the same increase in standard deviation, while a smaller adjustment coefficient results in a smaller threshold adjustment. When the adjustment coefficient is 0.8, the threshold's response to changes in standard deviation is within a reasonable range, responding to environmental changes without causing a significant drop in pass rate due to excessive adjustment caused by fluctuations. In the formula... The standard deviation, representing the historical authentication score sequence, is calculated by the standard deviation calculation unit based on the comprehensive authentication score sequence of authenticated users for the current day. The standard deviation is a dimensionless value reflecting the dispersion of the score sequence. The standard deviation is calculated by dividing the sum of the squares of the differences between each score and the sequence mean by the sequence length and then taking the square root. The calculation process covers all completed authentication score data for the current day. The standard deviation is recalculated for each new authentication score to ensure the timeliness of threshold adjustments. (Formula 2) This is the threshold increment, which is the adjustment coefficient multiplied by the standard deviation. The increment value increases as the standard deviation increases and decreases as the standard deviation decreases. When the authentication environment is stable and the user group characteristics are consistent on that day, the standard deviation... The value is small, the threshold increment is small, and the adjusted judgment threshold is... Approaching the baseline threshold of 0.6, the standard judgment criteria are maintained. However, when the daily authentication environment is complex and variable, such as drastic changes in lighting during the morning and evening, or when the user group has a wide age range and significant differences in the condition of their ID photos, the distribution range of the comprehensive authentication score widens, and the standard deviation increases. As the value increases, the threshold increment increases accordingly, and the adjusted judgment threshold... The threshold exceeds the baseline, raising the security threshold for authentication. The dynamic threshold mechanism allows the authentication system's security standards to automatically adjust according to the actual situation of the day. It proactively tightens the judgment standards when environmental complexity increases, reducing the probability of successful impersonation, and maintains lenient standards when the environment is stable, ensuring a smooth experience for legitimate users. (Formula output results) The adjusted threshold is dimensionless and serves as the criterion for determining whether certification is passed or not.

[0055] The decision and output module includes a comparison unit and an output interface. The two input terminals of the comparison unit are connected to the comprehensive authentication score output terminal of the fusion authentication scoring module and the judgment threshold output terminal of the dynamic threshold generation module, respectively. The comparison unit receives the comprehensive authentication score and the adjusted judgment threshold, and compares the two values. When the comprehensive authentication score is greater than or equal to the adjusted judgment threshold, the comparison unit outputs a login permission signal; when the comprehensive authentication score is less than the adjusted judgment threshold, the comparison unit outputs an authentication failure signal. The output interface converts the comparison result into a control command executable by the all-in-one machine. The login permission command is sent to the all-in-one machine's business system, triggering the login process and entering the employment service interface. When authentication fails, the decision and output module triggers an auxiliary authentication process. The all-in-one machine's display shows an SMS verification code input interface. After the user enters the SMS verification code received on their bound mobile phone number, they complete the identity verification. Alternatively, the all-in-one machine sends a manual verification request to the on-site management terminal, where on-site management personnel manually confirm the user's identity. The auxiliary authentication process supplements the main authentication path, covering users who cannot pass the main authentication, ensuring that all legitimate users can complete identity verification. SMS verification code is suitable for users who have a mobile phone number and can operate the mobile phone, while manual verification is suitable for special groups who cannot complete the authentication themselves. Both methods can be configured and enabled according to the actual scenario.

[0056] The ambient light sensing module collects ambient light illuminance values ​​ranging from 0 lux to 1000 lux, covering all typical lighting scenarios at the work station, from indoors at night to the entrance during the day. The light sensor exhibits linear output characteristics within the 0 lux to 1000 lux range, with measurement errors controlled within acceptable limits, ensuring stable and accurate illuminance data. The lower limit of the illuminance range, 0 lux, corresponds to a completely dark nighttime environment, while the upper limit, 1000 lux, corresponds to the diffuse light intensity at the entrance of the work station on a sunny day, covering all lighting conditions that the work station may encounter during its 24 / 7 operation. The sensor's range is matched to the actual scene, preventing saturation distortion under high illuminance due to insufficient range, and avoiding insufficient resolution under low illuminance due to excessive range.

[0057] The historical authentication score sequence obtained by the dynamic threshold generation module is composed of the comprehensive authentication scores of users authenticated on that day, arranged in chronological order of authentication time. At midnight each day, the score sequence is automatically cleared and reset, and authentication score data accumulates from zero for the new day. Each time a user authentication process is completed, regardless of whether authentication is successful or not, the corresponding comprehensive authentication score is added to the sequence in chronological order. The sequence length increases with the number of authentications per day, and the standard deviation is calculated in real time as the sequence is updated. The standard deviation is recalculated after each new data entry, and the adjusted judgment threshold is updated synchronously. Using daily data to construct the sequence ensures that the threshold adjustment remains synchronized with the actual environmental conditions of that day. Lighting conditions and user demographics differ across dates, and mixing data from different dates reduces the specificity of the threshold adjustment. Using only daily data ensures that the standard deviation reflects the authentication status of the current date, avoiding interference from historical data in current threshold decisions.

[0058] When the overall authentication score is less than the adjusted judgment threshold, the decision-making and output module triggers the all-in-one machine to display an SMS verification code input interface or sends a manual verification request to the on-site management terminal. In the SMS verification code authentication process, after the user enters their ID card number or scans their ID card, the system queries the user's registered mobile phone number and sends a short message containing a random verification code to that number. The user enters the verification code on the all-in-one machine interface, and the system allows login after verifying the verification code is correct. In the manual verification process, the all-in-one machine pushes the user's identity information and a captured image to the management terminal. After the management personnel verify that the information matches the on-site personnel, they click "confirm" on the management terminal, and the all-in-one machine allows login after receiving the confirmation instruction. Both auxiliary authentication methods can be selected and enabled based on the station's staffing and user characteristics. Stations with sufficient staff can focus on manual verification, while stations with limited staff can focus on SMS verification codes.

[0059] The employment service authentication method supporting both face and ID card login is executed by an all-in-one machine. This machine is equipped with an ambient light sensing module, a face acquisition and quality assessment module, an ID card reading and photo parsing module, a face comparison module, an adaptive weight calculation module, a fusion authentication scoring module, a dynamic threshold generation module, and a decision and output module. After the authentication process begins, the ambient light sensing module acquires the current ambient light level and generates a light influence factor based on that level. The face acquisition and quality assessment module acquires the face image of the user on-site, processes the image, and calculates a comprehensive face quality score. The ID card reading and photo parsing module reads the photo image from the ID card and extracts the facial features from the photo. The face comparison module extracts features from the on-site face image and compares these features with the ID card photo features to obtain the original similarity value. The adaptive weight calculation module receives the illumination impact factor and the comprehensive face quality score, generates a face comparison weight based on these two values, and subtracts the face comparison weight from 1 to obtain the quality item weight. As the product of the illumination impact factor and the comprehensive face quality score increases, the face comparison weight increases from a preset lower limit to a preset upper limit. The fusion authentication scoring module receives the face comparison weight, quality item weight, original similarity, and comprehensive face quality score, and calculates a weighted average to obtain the comprehensive authentication score. The dynamic threshold generation module obtains the historical authentication score sequence, calculates the standard deviation of the sequence, and generates an adjusted judgment threshold based on the standard deviation. The decision and output module compares the comprehensive authentication score with the adjusted judgment threshold. When the comprehensive authentication score is greater than or equal to the adjusted judgment threshold, it outputs a login permission command.

[0060] In the authentication method, the preset lower limit of weight is 0.3, the preset upper limit of weight is 0.9, and the preset coupling index is 2. The generation of the illumination influence factor is based on a reference critical illuminance of 200 lux and a steepness coefficient of 0.01 per lux. In the calculation of the comprehensive face quality score, the weight of the clarity score is 0.6, and the weight of the occlusion rate is 0.4. In the dynamic threshold generation step, the base threshold used to generate the adjusted judgment threshold is 0.6, and the adjustment coefficient is 0.8. The selection of the above parameters is determined based on on-site measured data. During deployment, the preset parameters can be used directly without additional on-site parameter adjustment. After the parameter configuration is fixed, the execution flow of the authentication method remains consistent, and the all-in-one machine deployed in different gig economy stations can achieve uniform authentication performance.

[0061] To verify the feasibility of the technical solution of this invention, computer simulation was used to simulate the authentication process. The simulation parameters were set as follows: reference critical illuminance. Steepness coefficient Weight lower limit Weight cap Coupling index Comprehensive facial recognition quality score Original similarity Basic threshold adjustment coefficient The simulation environment simulates the continuous authentication process at a gig economy station from 8:00 AM to 10:15 AM, with authentication occurring every 15 minutes. Illuminance gradually increases from 80 lux to 500 lux. Based on these parameters, the illumination impact factor, face comparison weight, comprehensive authentication score, and dynamic threshold based on the standard deviation of the score sequence are calculated for each time point. The simulation results are presented as the illumination impact factor-illuminance relationship curve and the face comparison weight variation curve. The changes are presented in the form of curves and a comparison chart of the comprehensive certification score and the dynamic threshold.

[0062] Unless otherwise specified, the equipment components involved in the above embodiments are all conventional equipment components, and the connection methods and control methods involved are all conventional connection methods and control methods unless otherwise specified.

[0063] The present invention has been described in detail above with reference to the embodiments. However, those skilled in the art will understand that, without departing from the spirit of the present invention, various specific parameters in the above embodiments can be changed to form multiple specific embodiments, all of which are common variations of the present invention, and will not be described in detail here.

Claims

1. An all-in-one employment service machine that supports login via facial recognition and ID card, characterized in that, include: Ambient light sensing module: electrically connected to the adaptive weight calculation module, used to obtain the ambient light intensity of the environment in which the all-in-one machine is located, and generate a light influence factor based on the ambient light intensity; Face capture and quality assessment module: Electrically connected to the adaptive weight calculation module and the fusion authentication scoring module, used to capture the user's face image on site and calculate the comprehensive face quality score; ID card reading and photo parsing module: electrically connected to the face comparison module, used to read the photo stored in the user's ID card and extract the photo features; Face comparison module: electrically connected to the fusion authentication scoring module, used to compare the features of the on-site face image with the features of the ID photo, and output the original similarity; Adaptive weight calculation module: electrically connected to the fusion authentication scoring module, used to generate face comparison weights based on the illumination influence factor and the comprehensive face quality score, and subtract the face comparison weights from 1 to obtain the quality item weights. When the product of the illumination influence factor and the comprehensive face quality score increases, the face comparison weights increase from a preset lower limit to a preset upper limit. Fusion authentication scoring module: electrically connected to the dynamic threshold generation module and the decision and output module, used to calculate the comprehensive authentication score based on the face comparison weight, the quality item weight, the original similarity and the comprehensive face quality score; Dynamic threshold generation module: electrically connected to the decision and output module, used to obtain the standard deviation of the historical authentication score sequence and generate an adjusted judgment threshold based on the standard deviation; Decision and output module: Used to compare the comprehensive authentication score with the adjusted judgment threshold, and to allow login when the comprehensive authentication score is greater than or equal to the adjusted judgment threshold.

2. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 1, characterized in that, The ambient light sensing module includes a light sensor and a light influence factor calculation unit. The light sensor is installed on the bezel of the all-in-one screen and is used to collect the ambient light intensity. The light influence factor calculation unit is used to calculate the light influence factor based on the ambient light intensity using the sigmoid function. The value range of the light influence factor is 0 to 1.

3. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 1, characterized in that, The face acquisition and quality assessment module includes a camera and an image processing unit. The image processing unit is used to extract the clarity score and occlusion rate of the face region, and to perform a weighted summation of the clarity score and the occlusion rate to obtain the comprehensive face quality score.

4. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 1, characterized in that, The adaptive weight calculation module generates the face comparison weights by: calculating the product of the illumination influence factor and the comprehensive face quality score; performing an exponential operation on the product, where the exponent is a preset coupling index; and linearly transforming the operation result to between the preset lower limit and the preset upper limit to obtain the face comparison weights.

5. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 4, characterized in that, The preset lower limit is 0.3, the preset upper limit is 0.9, and the preset coupling index is 2; the ambient light perception module generates the illumination influence factor based on a reference critical illuminance of 200 lux and a steepness coefficient of 0.01 lux⁻¹; in the face acquisition and quality assessment module, the weight of the clarity score is 0.6, and the weight of the occlusion rate is 0.4; the dynamic threshold generation module uses a base threshold of 0.6 and an adjustment coefficient of 0.8 to generate the adjusted judgment threshold.

6. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 5, characterized in that, The ambient light illuminance value obtained by the ambient light sensing module ranges from 0 lux to 1000 lux.

7. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 5, characterized in that, The historical authentication score sequence obtained by the dynamic threshold generation module is composed of the comprehensive authentication scores of the users who have been authenticated by the all-in-one machine on that day, arranged in chronological order of authentication time.

8. The employment service all-in-one machine supporting facial recognition and ID card login as described in claim 1, characterized in that, When the comprehensive authentication score is less than the adjusted judgment threshold, the decision and output module triggers the all-in-one machine display screen to show the SMS verification code input interface, or sends a manual verification request to the on-site management terminal.

9. A method for employment service authentication supporting both face and ID card login, executed by an all-in-one machine, the all-in-one machine comprising an ambient light sensing module, a face acquisition and quality assessment module, an ID card reading and photo parsing module, a face comparison module, an adaptive weight calculation module, a fusion authentication scoring module, a dynamic threshold generation module, and a decision and output module, characterized in that, The method includes: S1: The ambient light sensing module acquires the ambient light intensity and generates a light influence factor based on the ambient light intensity; S2: The face acquisition and quality assessment module acquires on-site face images and calculates a comprehensive face quality score; S3: The ID card reading and photo parsing module reads the ID card photo and extracts the photo features; S4: The face comparison module compares the on-site facial features with the features of the ID photo to obtain the original similarity. S5: The adaptive weight calculation module generates face comparison weights based on the illumination influence factor and the comprehensive face quality score, and subtracts the face comparison weights from 1 to obtain the quality item weights. When the product of the illumination influence factor and the comprehensive face quality score increases, the face comparison weights increase from a preset lower limit to a preset upper limit. S6: The fusion authentication scoring module calculates the comprehensive authentication score based on the face comparison weight, the quality item weight, the original similarity, and the comprehensive face quality score; S7: The standard deviation of the historical authentication score sequence is obtained by the dynamic threshold generation module, and an adjusted judgment threshold is generated based on the standard deviation; S8: The decision and output module compares the comprehensive authentication score with the adjusted judgment threshold. Login is allowed when the comprehensive authentication score is greater than or equal to the adjusted judgment threshold.

10. The employment service authentication method supporting facial recognition and ID card login according to claim 9, characterized in that, The preset lower limit is 0.3, the preset upper limit is 0.9, and the preset coupling index is 2; the generation of the illumination influence factor is based on a reference critical illuminance of 200 lux and a steepness coefficient of 0.01 lux⁻¹. In the calculation of the comprehensive face quality score, the weight of the clarity score is 0.6 and the weight of the occlusion rate is 0.4; in the dynamic threshold generation step, the base threshold used to generate the adjusted judgment threshold is 0.6 and the adjustment coefficient is 0.8.