A method and system for identity authentication of charging card dispensers based on multi-feature fusion of facial features

Abstract

This invention discloses a method and system for identity authentication in a charging card dispenser based on multi-feature fusion of facial features. Belonging to the field of card dispenser identity authentication technology, when a person is detected entering the acquisition area, the system automatically adjusts the camera parameters of the charging card dispenser after adaptive lighting supplementation. It automatically detects the position and angle of the person being acquired and acquires a facial image. Feature point detection is performed on the acquired facial image to extract facial features, and feature fusion is performed to generate a facial feature vector. The person being acquired is authenticated based on a set of legitimate user facial data. When the identity authentication is successful, a drive command is sent to the card dispenser's card magazine to eject the card, and the card identity is dynamically bound according to the user's identity information. This achieves an efficient, accurate, secure, and intelligent identity authentication and card issuance process for charging card dispensers.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of card issuing machine identity authentication, in particular to a face multi-feature fusion charging card issuing machine identity authentication method and system. BACKGROUND

[0002] In today's era of rapid development of digitization and intelligence, various places have put forward higher and higher requirements for the accuracy and efficiency of personnel identity authentication and management. Especially in some scenarios that require a large number of personnel to be issued cards, charged and identified, such as enterprise parks, large factories, schools, etc., the traditional identity authentication method has exposed many drawbacks. In the past, relying solely on IC / ID cards for identity recognition, it is easy to lose cards and be stolen, and the security cannot be guaranteed. Moreover, when facing a large number of personnel flow, the card management work becomes extremely cumbersome and inefficient.

[0003] Chinese patent with authorization publication number CN108921191B discloses a multi-biometric feature fusion recognition method based on image quality evaluation. First, the face and iris images of the user are collected by the image acquisition subsystem and are preprocessed accordingly. Second, the preprocessed images are sent to the corresponding recognition authentication subsystem for quality evaluation, feature extraction and template matching, etc. The respective matching scores and corresponding matching quality confidence scores are output. Finally, these scores are normalized and sent to the recognition fusion subsystem. The dynamic weighted fusion algorithm is used to obtain the recognition authentication result.

[0004] Although the prior art has higher recognition accuracy than single face or iris recognition algorithm, it can achieve high-precision personal identity recognition. At the same time, based on the existing mainstream fusion recognition algorithm, useful information of image quality is extracted and applied to the fusion recognition authentication process, further improving the system recognition performance.

[0005] However, it still cannot solve the problem of stable collection of high-quality face images under complex environmental light interference, which cannot ensure the start of efficient identity authentication process. The use of multi-feature fusion is not fine enough, which makes it difficult to accurately and quickly filter out target users in a large sample legal user database, affecting the authentication efficiency and accuracy. Therefore, in order to overcome these limitations, the present application provides a face multi-feature fusion charging card issuing machine identity authentication method and system. SUMMARY

[0006] In view of the deficiencies of the prior art, the purpose of the present application is to provide a face multi-feature fusion charging card issuing machine identity authentication method and system, which solves the technical problems of high-quality collection of face images under complex environment, fine fusion of multiple features for accurate and rapid authentication in a large sample legal user database, and dynamic binding of charging card and user identity after authentication.

[0007] To achieve the above object, the present application provides the following technical solutions:

[0008] The face multi-feature fusion charging card dispenser identity authentication system comprises a face collection module, a feature fusion module, an identity authentication module and a card dispensing control module.

[0009] The face collection module is used for acquiring the ambient light intensity through the light sensor of the charging card dispenser when detecting that the collector enters the collection area, automatically adjusting the camera parameters of the charging card dispenser after adaptive light source light compensation based on the ambient light intensity, and collecting the face image by automatically detecting the position and angle of the collector.

[0010] The feature fusion module is used for detecting the feature points of the collected face image, extracting the facial features of the face image based on the feature points, including geometric features and texture features, performing feature fusion on the facial features, and generating a facial feature vector.

[0011] The identity authentication module is used for performing identity authentication on the collector based on a legal user face data set, performing rapid screening on the collector through the geometric feature vector to obtain a preliminary legal user set, screening out an identity authentication set according to the step-by-step screening ratio of each down-sampling scale and the texture feature vector similarity order, then constructing a key facial feature vector, and screening out a target legal user to determine whether the identity authentication of the collector is successful.

[0012] The card dispensing control module is used for sending a driving instruction to the card cassette of the charging card dispenser when the identity authentication of the collector is passed, controlling the card cassette to eject the card, and dynamically binding the card identity according to the user identity information of the collector.

[0013] Specifically, the step of collecting the face image comprises:

[0014] The infrared sensor of the charging card dispenser continuously monitors the collection area to obtain the object features entering the collection area, and performs human body recognition based on the object features.

[0015] When the human body is recognized, a monitoring time window is configured to monitor the human body, the number of times of recognizing the human body is counted in the monitoring time window, and the human body recognition success rate is calculated.

[0016] A human body recognition threshold is configured, when the human body recognition success rate is greater than the human body recognition threshold, the face collection process is started, otherwise no operation is performed.

[0017] A collection trigger threshold is configured, the collection distance between the collector and the camera of the charging card dispenser is obtained through the distance sensor, the collection distance change rate is obtained, and when the collection distance change rate is less than the collection trigger threshold, the light compensation device parameter adjustment and the camera parameter adjustment are triggered.

[0018] After the light supplement device parameters and the camera parameter adjustment, the body contour information of the collector is captured by the camera, and the key point coordinates of the human body contour are extracted;

[0019] According to the key point coordinates of the human body contour, the horizontal offset and the vertical offset of the head relative to the image center are calculated, and the tilt angle of the collector's head is calculated;

[0020] The position deviation range and the angle deviation range are configured. When the horizontal offset of the head relative to the image center, the vertical offset of the head relative to the image center, and the tilt angle of the head are all within the position deviation range and the angle deviation range, the face image collection is performed. Otherwise, according to the horizontal offset of the head relative to the image center, the vertical offset of the head relative to the image center, and the tilt angle of the head, the voice prompt is given to guide the collector to adjust the position and the angle.

[0021] Specifically, the step of collecting the face image further includes:

[0022] The collected face image is preliminarily evaluated to obtain the face image clarity, the face image integrity, and the face image brightness uniformity;

[0023] The face image clarity is obtained by calculating the variance of the image Laplacian operator;

[0024] The face image integrity is obtained by calculating the completeness and the missing proportion of the face contour key points;

[0025] The face image brightness uniformity is obtained by calculating the brightness standard deviation of different regions of the face image;

[0026] The clarity threshold, the integrity threshold, and the brightness threshold are configured. When the face image clarity is less than the clarity threshold, or the face image integrity is less than the integrity threshold, or the face image brightness uniformity is less than the brightness threshold, the image collection is retriggered. Otherwise, no operation is performed;

[0027] The re-collection threshold is configured, and the number of times of retriggering the image collection is monitored. When the number of times of retriggering the image collection is greater than the re-collection threshold, the collection process is suspended, and a collection warning is given.

[0028] Specifically, the step of adjusting the light supplement device parameters includes:

[0029] The light intensity threshold is configured, including the upper limit threshold of the light intensity and the lower limit threshold of the light intensity. The ambient light intensity is obtained by the light sensor of the charging card machine;

[0030] When the ambient light intensity is less than the lower light intensity threshold, the fill light brightness and fill light intensity are increased, the initial fill light brightness is adjusted according to the proportion of the current ambient light intensity to the lower light intensity threshold, and the fill light brightness is optimized in combination with the position of the face in the image;

[0031] When the ambient light intensity is greater than or equal to the upper light intensity threshold, the fill light brightness is reduced, and the fill light brightness is reduced in proportion to the degree that the light intensity exceeds the upper light intensity threshold on the basis of the initial fill light intensity until it is reduced to the minimum fill light brightness.

[0032] After adjusting the fill light brightness, the fill light angle is adjusted according to the offset direction of the face center relative to the image center on the basis of the initial fill light angle.

[0033] Specifically, the step of adjusting the camera parameters comprises:

[0034] The ambient light intensity after adjusting the fill light device parameters is obtained, and the camera parameters including the sensitivity, focal length, aperture and shutter speed are adjusted;

[0035] If the ambient light intensity obtained by the light sensor of the charging card machine is still greater than the upper light intensity threshold at the fill light brightness reduced by the fill light device, the camera exposure parameters are adjusted, and the camera sensitivity is reduced in proportion to the degree that the ambient light intensity exceeds the upper threshold;

[0036] The camera focal length is adjusted according to the distance between the collector and the camera, and the camera focal length is increased or decreased according to the change of the distance on the basis of the basic camera focal length;

[0037] According to the ambient light intensity, the initial aperture is adjusted according to the difference between the current ambient light intensity and the reference light intensity, and when the ambient light intensity is higher than the reference light intensity, the camera aperture is increased; otherwise, the camera aperture is reduced;

[0038] The camera shutter speed is adjusted according to the ambient light intensity and the collection distance, and on the basis of the initial shutter speed, when the ambient light intensity is greater than the reference light intensity, the shutter speed is reduced in proportion, and the collection distance is inversely proportional to the shutter speed.

[0039] Specifically, the step of generating the face feature vector comprises:

[0040] The face image is obtained, and the feature points of the face image are detected, including eye feature points, nose feature points, mouth feature points and face contour feature points;

[0041] The geometric features of the face image are extracted based on the feature points of the face image, including distance features, angle features and proportion features;

[0042] The diagonal length of the face image is selected as a scale factor, and the distance feature in the geometric feature is normalized based on the scale factor to eliminate the influence of the face image size on the distance feature.

[0043] The normalized distance feature, angle feature and proportion feature are sorted in the order of the face image from top to bottom and from left to right, respectively, to construct a geometric distance feature vector, a geometric angle feature vector and a geometric proportion feature vector.

[0044] The geometric features are fused, and the geometric distance feature vector, the geometric angle feature vector and the geometric proportion feature vector are sequentially spliced to construct a geometric feature vector.

[0045] The face image is subjected to gray processing, and the face image after gray processing is subjected to multi-scale down-sampling to obtain a face image at each down-sampling scale for texture feature extraction to obtain a texture feature vector.

[0046] The geometric feature vector and the texture feature vector are spliced to obtain a face feature vector.

[0047] Specifically, the step of obtaining the texture feature vector includes:

[0048] Based on the key feature points selected from the feature points, key feature point detection is performed on the face image at each down-sampling scale.

[0049] According to the down-sampling scale, a neighborhood window size is set, and the gradient direction and gradient amplitude of each pixel point relative to the key feature point are calculated in the neighborhood window with the detected key feature point as the center.

[0050] The neighborhood window is regionally divided to obtain a neighborhood sub-region, and a gradient direction histogram is counted in each neighborhood sub-region. The gradient direction histograms of all neighborhood sub-regions are connected to form a texture feature of the key feature point.

[0051] For each down-sampling scale, a corresponding sampling scale texture feature vector is constructed according to the type of the key feature point, and the texture feature vectors of the various sampling scales are spliced in the order of the down-sampling scale from small to large to obtain the texture feature vector.

[0052] Specifically, the specific steps of identity authentication include:

[0053] Let the set of legal user face data be , wherein is the face data of the i-th legal user, which contains the geometric feature vector of the i-th legal user, the texture feature vector at each down-sampling scale and the face feature vector.

[0054] ​​A fast screening ratio is configured to obtain a geometric feature vector of the collector, and the geometric feature vector of the collector is compared with geometric feature vectors of legal users in the face database set to perform fast screening on the face data set of the legal users, and similarity degrees of the geometric feature vector of the collector and the geometric feature vectors of the legal users in the face database set are calculated;

[0055] The similarity degrees of the geometric feature vector of the collector and the geometric feature vectors of the legal users in the face database set are sorted, and a preliminary legal user set is obtained according to the fast screening ratio;

[0056] A step-by-step screening ratio of each down-sampling scale is configured to obtain a texture feature vector of the collector at each down-sampling scale, and the preliminary legal user set is step-by-step screened, and similarity degrees of the texture feature vector of the collector at each down-sampling scale and the texture feature vectors of the legal users in the fast screening preliminary legal user set are calculated in sequence;

[0057] The texture feature vector similarity degrees of each down-sampling scale are sorted in sequence, and a step-by-step screening legal user set is obtained according to the step-by-step screening ratio of each down-sampling scale to obtain an identity authentication set.

[0058] Specifically, the specific steps of the identity authentication further include:

[0059] A similarity threshold is configured to obtain a face feature vector of the collector, and the collector is finally identity authenticated, and the face feature vector of each legal user in the identity authentication set and the face feature vector of the collector are compared to determine importance scores of each vector element in the face feature vector according to a contribution degree of each vector element in the face feature vector in identity recognition, and importance sorting is performed;

[0060] An importance threshold is configured After the importance sorting is completed, according to the importance threshold, the first vector elements in each face feature vector are selected to construct a key face feature vector;

[0061] According to the importance scores of each vector element in the key face feature vector, similarity degrees of the key face feature vector of each legal user in the identity authentication set and the key face feature vector of the collector are calculated;

[0062] A legal user with the largest similarity degree with the key face feature vector of the collector in the identity authentication set is screened and marked as a target legal user, and if the similarity degree of the target legal user and the key face feature vector of the collector is greater than the similarity threshold, the identity authentication is successful, otherwise the identity authentication fails;

[0063] An identity authentication threshold is configured, and the number of times of identity authentication of the collector is recorded, and when the number of times of identity authentication is greater than the identity authentication threshold, an identity authentication warning is performed, otherwise no operation is performed.

[0064] The face multi-feature fusion charging card issuing machine identification authentication method comprises the following steps:

[0065] When the collector is detected to enter the collection area, the ambient light intensity is acquired through the light sensor of the charging card issuing machine, the camera parameters of the charging card issuing machine are automatically adjusted after adaptive light source light compensation based on the ambient light intensity, and the position and angle of the collector are automatically detected to collect the face image;

[0066] The collected face image is subjected to feature point detection, the face features of the face image are extracted based on the feature points, including geometric features and texture features, the face features are subjected to feature fusion to generate a face feature vector;

[0067] The collector is subjected to identity authentication based on a legal user face data set, a preliminary legal user set is obtained through rapid screening of the geometric feature vector of the collector, an identity authentication set is screened out according to the step-by-step screening ratio of each down-sampling scale and the texture feature vector similarity order, then a key face feature vector is constructed, and a target legal user is screened out to judge whether the collector identity authentication is successful;

[0068] When the collector identity authentication is passed, a driving instruction is sent to the card case of the charging card issuing machine to control the card case to eject the card, and the card identity is dynamically bound according to the user identity information of the collector.

[0069] The face multi-feature fusion charging card issuing machine identification authentication method has the following beneficial effects:

[0070] 1. The collector entering the collection area is accurately recognized through an infrared sensor, a distance sensor and the like, adaptive light source light compensation is automatically performed according to the ambient light intensity, the position and angle of the collector are automatically detected, and the camera parameters of the charging card issuing machine are automatically adjusted, so that suitable face images can be collected under various environments and collector postures, the intelligence and accuracy of collection are improved, and collection failure caused by light and position angle problems is reduced.

[0071] 2. The geometric features and texture features of the face image are extracted, the face feature vector is generated through feature fusion, and the multi-stage identity authentication is performed by the identity authentication module based on the face feature vector, including rapid screening of the geometric feature vector, step-by-step screening of the texture feature vector, construction of the key face feature vector and similarity calculation, etc., so that various feature information of the face is fully utilized, the accuracy and reliability of the identity authentication are greatly improved compared with single feature authentication.

[0072] 3. The multi-stage screening method is adopted to rapidly narrow the range and then gradually and finely screen, so that the authentication accuracy is ensured, the authentication efficiency is improved, unnecessary calculation amount and time consumption are reduced, the collector identity can be quickly and accurately judged, and the user experience and overall performance of the charging card issuing machine are improved. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 Structure diagram of the face multi-feature fusion charging card issuing machine recognition identity authentication system of the present application;

[0074] Figure 2 Flow chart of the specific steps of collecting face images of the present application;

[0075] Figure 3 Schematic diagram of the specific steps of adjusting the parameters of the light supplementing equipment of the present application;

[0076] Figure 4 Flow chart of the specific steps of adjusting the parameters of the camera of the present application;

[0077] Figure 5 Flow chart of the specific steps of generating the face feature vector of the present application;

[0078] Figure 6 Flow chart of the specific steps of identity authentication of the present application;

[0079] Figure 7 Flow chart of the face multi-feature fusion charging card issuing machine recognition identity authentication method of the present application. DETAILED DESCRIPTION

[0080] Embodiment 1

[0081] Referring to Figure 1 , the present embodiment introduces a face multi-feature fusion charging card issuing machine recognition identity authentication system, which comprises a face collecting module, a feature fusion module, an identity authentication module and a card issuing control module;

[0082] The face collecting module is used for acquiring the ambient light intensity through the light sensor of the charging card issuing machine when detecting that the collector enters the collecting area, automatically adjusting the camera parameters of the charging card issuing machine after adaptive light source light supplementing based on the ambient light intensity, and collecting the face image through automatic detection of the position and angle of the collector;

[0083] In this embodiment, when the collector enters the collection area, the precise ambient light intensity data is obtained by the light sensor built-in the charging card machine. If the light intensity is lower than the set threshold, the light supplement device is automatically turned on, and the light supplement brightness and angle are adjusted by the intelligent algorithm to ensure that the light uniformly covers the face and avoid shadow; if the light is too strong, the light supplement brightness is appropriately reduced or the camera exposure parameter is adjusted to ensure that the brightness and contrast of the collected image are appropriate. In addition, the camera parameters of the charging card machine are automatically adjusted, including adjusting the focal length according to the collection distance to ensure that the face occupies a proper proportion in the image, generally the face height accounts for one-third to one-half of the image height; at the same time, according to the environmental light and the shooting demand, the aperture is adjusted to control the depth of field, and the appropriate shutter speed is set to avoid image blur. It also has the functions of automatic detection of position and angle. The body contour information of the collector is captured by the camera to monitor and judge the position and angle of the collector in real time. Once the position or angle deviation is detected, the system will guide the collector to adjust to the best position through voice prompt or on-screen guide mark. Finally, a clear and complete face image is collected.

[0084] Please refer to Figure 2 , preferably, the specific steps of collecting the face image include:

[0085] The infrared sensor of the charging card machine continuously monitors the collection area to obtain the characteristics of the object entering the collection area, including the object shape, object size, object movement speed and object movement trajectory, and the object characteristics are recognized by the machine learning algorithm; whether the object entering the area is a human body is accurately identified, thereby providing a basis for subsequent judgment whether to start the face collection process, avoiding false triggering of the collection process, and improving the pertinence of collection.

[0086] When a human body is recognized, a monitoring time window is configured to continuously monitor the human body, and the number of times of recognizing the human body in the monitoring time window is counted , and the human body recognition success rate is calculated, that is:

[0087]

[0088] Among them, is the human body recognition success rate in the monitoring time window, is the total number of detections in the monitoring time window;

[0089] The human body recognition threshold is configured , when the human body recognition success rate is greater than the human body recognition threshold When a human body enters the collection area, the face collection process is initiated; otherwise, no operation is performed, and area monitoring continues. This avoids erroneous initiation of the collection process due to the brief or accidental appearance of a human body, thus improving the reliability of the face collection process initiation judgment.

[0090] Configure data collection trigger threshold The distance between the subject and the camera of the charging card dispenser is obtained through a distance sensor, and the rate of change of the distance is obtained, i.e.:

[0091]

[0092] in, It is a moment The rate of change of the sampling distance, It is a moment The sampling distance, It is the time interval for obtaining the acquisition distance between the subject and the charging card issuing machine camera. When the acquisition distance change rate is less than the acquisition trigger threshold, it triggers the adjustment of the supplementary lighting equipment parameters and the camera parameters. It can adjust the equipment parameters in a timely manner according to the movement status of the subject, ensuring that the parameters are optimized at the appropriate time to prepare for obtaining high-quality facial images.

[0093] Please see Figure 3 The steps for adjusting the parameters of the supplementary lighting equipment include: configuring the light intensity threshold, including the upper limit threshold and the lower limit threshold of the light intensity, and obtaining the ambient light intensity through the light sensor of the charging card dispenser;

[0094] When the ambient light intensity is below the lower limit threshold, the supplementary lighting device is activated. The initial supplementary lighting brightness is adjusted according to the ratio of the current ambient light intensity to the lower limit threshold; the weaker the light intensity, the stronger the supplementary lighting brightness. Furthermore, the supplementary lighting brightness is optimized based on the position of the face in the image; the farther the face is from the image center, the stronger the supplementary lighting brightness, to ensure uniform overall illumination.

[0095]

[0096] in, This refers to the enhanced brightness of the supplementary lighting after the supplementary lighting device has been used to increase the brightness of the supplementary lighting. This is the initial fill light brightness. It is the ambient light intensity obtained by the light sensor of the charging card dispenser. It is the lower limit threshold of light intensity. This is an environmental adjustment factor, with a value range of (0.1, 1). and It is the offset of the face center from the image center in the horizontal and vertical directions. is the length of the image diagonal; not only the difference in light intensity is considered, but also the position of the face in the image, to adjust the fill light brightness more accurately.

[0097] When the ambient light intensity is greater than or equal to the upper limit threshold of light intensity, that is, the light is too strong, the fill light brightness is reduced, and on the basis of the initial fill light intensity, it is reduced in proportion to the degree that the light intensity exceeds the upper limit threshold of light intensity, until it is reduced to the minimum fill light brightness, that is:

[0098]

[0099] wherein, is the reduced fill light brightness of the fill light device when the fill light brightness is reduced, is the upper limit threshold of light intensity, is the minimum fill light brightness of the fill light device, is the adjustment coefficient of fill light intensity when the light is too strong, and the value range is (0.1, 1);

[0100] After adjusting the fill light brightness, on the basis of the initial fill light angle, the fill light angle is adjusted according to the offset direction of the face center relative to the image center, so that the fill light can cover the face more accurately and avoid producing shadows, that is:

[0101]

[0102] wherein, is the adjusted fill light angle, is the initial fill light angle, and is the offset amount of the face center and the image center in the horizontal and vertical directions, is the angle adjustment coefficient, and the value range is (0.1, 1);

[0103] Please refer to Figure 4 , the steps of adjusting the camera parameters include: obtaining the ambient light intensity after adjusting the fill light device parameters, to adjust the camera parameters, including the sensitivity, focal length, aperture and shutter speed;

[0104] If the ambient light intensity obtained by the light sensor of the charging card machine is still greater than the upper limit threshold of light intensity under the reduced fill light brightness of the fill light device, the camera exposure parameters are adjusted, and the camera sensitivity is reduced in proportion to the degree that the ambient light intensity exceeds the upper limit threshold, that is:

[0105]

[0106] wherein, is the adjusted sensitivity, is the initial sensitivity, is the ambient light intensity after the light supplement device parameter adjustment, is the sensitivity adjustment coefficient, and the value range is (0.1, 1);

[0107] The focal length of the camera is adjusted according to the distance between the collector and the camera. On the basis of the focal length, the focal length is increased or decreased according to a certain rule according to the change of the distance. The farther the distance, the greater the focal length adjustment amount, so as to ensure that the face in the image can maintain a proper proportion, and the face height is about one third to one half of the image height, that is,

[0108]

[0109] wherein, is the adjusted focal length of the camera, is the initial focal length of the camera, is the focal length adjustment coefficient, and the value range is (0.1, 1), is the distance between the collector and the camera;

[0110] The aperture is adjusted according to the ambient light intensity. On the basis of the initial aperture, the aperture is adjusted according to the difference between the current ambient light intensity and the reference light intensity. When the ambient light intensity is higher than the reference light intensity, the aperture of the camera is increased. Conversely, the aperture of the camera is decreased, so as to control the depth of field and ensure that the clarity of the face and the background in the image meets the requirements, that is,

[0111]

[0112] wherein, is the adjusted aperture, is the initial aperture, is the reference light intensity corresponding to the initial aperture, is the ambient light intensity after the light supplement device parameter adjustment, is the aperture adjustment coefficient, and the value range is (0.1, 1), so as to dynamically adjust the aperture size according to the difference between the ambient light intensity and the reference light intensity, and control the depth of field.

[0113] The shutter speed of the camera is adjusted according to the ambient light intensity and the collection distance. On the basis of the initial shutter speed, when the ambient light intensity is greater than the reference light intensity, the shutter speed is reduced by a certain proportion. The farther the collection distance, the shutter speed is reduced by a certain proportion, so as to avoid image blur caused by too strong light or movement of the collector, that is,

[0114]

[0115] wherein, is the adjusted shutter speed, is the initial shutter speed, is the reference light intensity corresponding to the initial shutter speed, is the ambient light intensity after the parameter adjustment of the light supplement device, is the distance between the collector and the camera, is the shutter speed adjustment coefficient, which comprehensively considers the influence of light intensity and collection distance on shutter speed to avoid image blur.

[0116] After the parameter adjustment of the light supplement device and the camera, the body contour information of the collector is captured by the camera, and the position and angle of the collector are monitored and judged in real time by using image processing algorithm, and the key point coordinates of the human body contour are extracted, including the head key point, the shoulder key point and the waist key point.

[0117] According to the key point coordinates of the human body contour, the horizontal and vertical offset of the head relative to the image center are calculated, and the inclination angle of the head of the collector is calculated , that is,

[0118]

[0119]

[0120]

[0121] wherein, is the horizontal offset of the head relative to the image center, is the vertical offset of the head relative to the image center, is the head key point coordinate, is the image center coordinate;

[0122] The position deviation range and the angle deviation range are configured, when the horizontal offset of the head relative to the image center, the vertical offset of the head relative to the image center are within the position deviation range, and the inclination angle of the head is within the angle deviation range, the face image is collected, otherwise, according to the horizontal offset of the head relative to the image center, the vertical offset of the head relative to the image center and the inclination angle of the head, the voice prompt is given to guide the collector to adjust the position and angle, and the voice prompt includes "please move a little to the left" "please lift your head a little";

[0123] The collected face image is preliminarily evaluated to obtain the face image definition, the face image integrity and the face image brightness uniformity, wherein the face image definition is obtained by calculating the variance of the image Laplace operator, that is,

[0124]

[0125] wherein, is the face image definition, It is the Laplace operator. It is a variance function;

[0126] The completeness of a face image is obtained by calculating the completeness and missing proportion of key points in the facial contour, that is:

[0127]

[0128] in, It is a facial image Completeness It detected a face image. Number of key points It is the total number of key points in the facial contour;

[0129] The brightness uniformity of a face image is obtained by calculating the standard deviation of brightness in different regions of the face image. The face image is divided into multiple regions, and the brightness of each region is compared with the average brightness of the face image. The smaller the standard deviation of brightness in the face image regions, the more uniform the brightness.

[0130]

[0131] in, It is a facial image Brightness uniformity It is the number of regions that a face image is divided into. It is the first Brightness of each area It is the average brightness of the face image;

[0132] Configure clarity thresholds, integrity thresholds, and brightness thresholds. When the clarity of a face image is less than the clarity threshold, the integrity of a face image is less than the integrity threshold, or the brightness uniformity of a face image is less than the brightness threshold, image acquisition is retried; otherwise, no operation is performed. The acquired face images meet certain quality standards, improving the reliability and usability of face image data, which is beneficial for subsequent applications such as identity authentication.

[0133] Configure a re-acquisition threshold and monitor the number of times image acquisition is retried. When the number of re-acquisitions exceeds the threshold, the acquisition process is paused, an acquisition warning is issued, and the user is prompted to check their own status and environmental conditions. This avoids resource waste caused by continuously acquiring low-quality images and reminds users of factors that may affect acquisition quality, helping them to make timely adjustments and ensuring smooth acquisition.

[0134] The feature fusion module is used to detect feature points in the acquired face images, extract facial features from the face images based on the feature points, including geometric features and texture features, and perform feature fusion on the facial features to generate a facial feature vector.

[0135] In this embodiment, from the collected face image, the feature fusion module uses multiple advanced feature extraction algorithms to extract different types of facial features. The key geometric features of the face are measured and recorded, such as the distance between the two eyes, the length and width of the nose, the outline and position of the mouth, etc. These features constitute the basic shape framework of the face and are one of the important bases for recognition. Using texture analysis technology, the texture information of the face skin surface is extracted, such as wrinkles, pores, etc. These texture features have individual uniqueness, even identical twins have differences, providing more detailed information for identity recognition. The extracted geometric features and texture features are fused to generate a comprehensive facial feature vector, which contains multi-dimensional feature information of the face, greatly improving the accuracy and reliability of identity recognition.

[0136] Please refer to Figure 5 , preferably, the specific steps of generating a facial feature vector include:

[0137] Obtain a face image, detect feature points of the face image through a deep learning network, the feature points including eye feature points, nose feature points, mouth feature points, and face contour feature points; the eye feature points including a canthus point, a pupil center point, a pupil edge point, and an eyelid edge point; the nose feature points including a tip point, a wing edge point, and a bridge point; the mouth feature points including a corner point, a lip peak point, and a lip valley point; the face contour feature points including a chin top point, a cheek edge point, and a forehead edge point; providing accurate position identification for geometric feature extraction and texture feature analysis, and being the basis for further analysis of face features, which helps to accurately depict the shape and structure of the face.

[0138] Extract the geometric features of the face image based on the feature points of the face image, including distance features, angle features, and proportion features; the distance features including an interocular distance, an inner canthus distance, an outer canthus distance, an eye height, a nose length, a wing width, a mouth width, a lip thickness, a face length, and a face width; the angle features including an eye tilt angle, a bridge tilt angle, and a mandible angle; the proportion features including an eye-nose proportion, a mouth-nose proportion, a three- court proportion, and a five-eye proportion; constructing a basic framework of the face shape, and providing an important shape basis for recognition.

[0139] Select the diagonal length of the face image as a scale factor, and normalize the distance features in the geometric features based on the scale factor to eliminate the influence of the size of the face image on the distance features, i.e.

[0140]

[0141] wherein, is the normalized distance feature of the i-th feature point. Personal face image distance feature, is the first Personal face image distance feature, The value range of , is the total number of face image distance features, is the face image diagonal length; the influence of the size difference of the face image caused by the shooting distance, image scaling and other factors on the distance feature is eliminated, so that the distance features in different images are comparable, and the stability and reliability of the geometric features in the subsequent fusion and recognition process are improved.

[0142] The normalized distance feature, angle feature and proportion feature are sorted in the order of the face image from top to bottom and from left to right, respectively, to construct a geometric distance feature vector, a geometric angle feature vector and a geometric proportion feature vector;

[0143] The geometric features are fused, the geometric distance feature vector, the geometric angle feature vector and the geometric proportion feature vector are sequentially spliced to construct a geometric feature vector; the geometric features are presented in a structured form, which is convenient for subsequent fusion with other features and utilization by the recognition algorithm, while the information of the geometric relationship of each part of the face is retained.

[0144] The face image is subjected to gray processing, and the face image after gray processing is subjected to multi-scale down-sampling to obtain a face image at each down-sampling scale; different scale images can capture different levels of detail information, small scale images contain more global features, and large scale images retain more local details, providing a rich data basis for comprehensive extraction of texture features.

[0145] Based on the feature points, key feature points are selected, including the eye corner point, the nose tip point, the nose wing edge point, the mouth corner point, the lip peak point and the lip valley point, the chin top point, the cheek edge point and the forehead edge point, and the key feature points of the face image at each down-sampling scale are detected;

[0146] According to the down-sampling scale, the size of the neighborhood window is set, and the gradient direction and gradient amplitude of each pixel point relative to the key feature point are calculated in the neighborhood window with the detected key feature point as the center; the detail features of the face texture, such as wrinkles and pores, are highlighted, which provides key information for the extraction of texture features.

[0147] The neighborhood window is divided into regions to obtain a neighborhood sub-region, the gradient direction histogram is counted in each neighborhood sub-region, and the gradient direction histograms of all neighborhood sub-regions are connected to form the texture feature of the key feature point;

[0148] For each downsampling scale, based on the type of key feature points, a corresponding sampling scale texture feature vector is constructed. The texture feature vectors of each sampling scale are concatenated in ascending order of downsampling scale to obtain the final texture feature vector.

[0149] The geometric feature vector and the texture feature vector are concatenated to obtain the facial feature vector. This facial feature vector integrates the geometric and texture features of the face, contains rich facial information, and provides important feature basis for subsequent applications such as face recognition and identity authentication.

[0150] The identity authentication module is used to authenticate the identity of the subject based on the set of legitimate user face data. It quickly filters the geometric feature vectors of the subject to obtain a preliminary set of legitimate users. Then, it sorts the identity authentication set according to the progressive filtering ratio of each downsampling scale and the similarity of texture feature vectors. Finally, it constructs key facial feature vectors to filter out target legitimate users and determine whether the identity authentication of the subject is successful.

[0151] In this embodiment, the identity authentication module compares the geometric feature vector, texture feature vector at each downsampling scale, and facial feature vector generated by the feature fusion module with a pre-stored set of legitimate user face data belonging to the registered user. The feature vectors stored in the database are generated during user registration through a rigorous collection and feature extraction process, and each vector is associated with the identity information of a specific user. Through a multi-level rapid authentication process, the similarity between the collected user and known users in the set of legitimate user face data is obtained to determine whether the identity authentication is successful.

[0152] Please see Figure 6 Preferably, the specific steps for identity authentication include:

[0153] Let the set of legitimate user face data be . ,in, It is the first The facial data of a legitimate user, including the first... The geometric feature vectors of each legitimate user, the texture feature vectors at each downsampling scale, and the facial feature vectors are: ,in, It is the first Geometric feature vectors of a legitimate user It is the first A legitimate user at the downsampling scale The texture feature vector below, It is the number of downsampling scales. It is the first Facial feature vector of a legitimate user It refers to the number of legitimate users; it clarifies the data foundation for identity authentication, and stores the facial data of legitimate users in a structured manner to provide a complete information source for subsequent authentication processes, including multi-dimensional information such as geometric features, texture features, and facial features, to ensure that the authentication process has comprehensive data support.

[0154] Configure quick filter ratio The process involves obtaining the geometric feature vector of the subject, quickly filtering the set of legitimate user face data, and calculating the similarity between the subject's geometric feature vector and the geometric feature vector of each legitimate user in the face database. For example, for the first legitimate user face data in the set of legitimate user face data... The formula for calculating the similarity of the geometric feature vectors of a group of legitimate users is:

[0155]

[0156] in, The subject of collection and the first Geometric feature vector similarity among legitimate users It is the first Geometric feature vectors of a legitimate user It is the geometric feature vector of the subject being collected. It is the first The geometric feature vector of the _th legitimate user One element, It is the first of the geometric feature vectors of the subject being collected. One element, It is the number of geometric feature vector elements; by quickly calculating the similarity between the geometric feature vectors of the subject and the legitimate users, and based on the fast screening ratio, a small preliminary set of legitimate users is quickly selected from the huge legitimate user face database, which greatly reduces the scope of subsequent processing, improves the initial efficiency of authentication, and reduces unnecessary computation.

[0157] The similarity between the geometric feature vectors of the data collector and the legitimate users in the face database is ranked, and then a rapid screening ratio is applied. To obtain a preliminary set of legitimate users, namely:

[0158]

[0159] in, It is the first in the face database collection Facial data of a legitimate user, The range of values ​​is , It involves selecting a preliminary set of legitimate users. The subject of collection and the first Geometric feature vector similarity of a legitimate user rankings of, is a rounding up function;

[0160] a step-by-step screening ratio of each down-sampling scale is configured, that is, wherein, is a step-by-step screening ratio of the down-sampling scale, is a step-by-step screening ratio of the down-sampling scale, is the number of down-sampling scales, the texture feature vector of each down-sampling scale of the collected person is obtained, the preliminary legitimate user set is step-by-step screened, and the similarity of the texture feature vectors of the down-sampling scales of the collected person and each legitimate user in the fast screening preliminary legitimate user set is calculated in turn. For example, for the texture feature vector similarity of the first legitimate user in the preliminary legitimate user set at the down-sampling scale , the calculation formula is:

[0161]

[0162] wherein, is the similarity of the texture feature vectors of the collected person and the first legitimate user in the preliminary legitimate user set at the down-sampling scale , is the texture feature vector of the collected person at the down-sampling scale , is the first element of the texture feature vector of the first legitimate user in the preliminary legitimate user set at the down-sampling scale , is the first element of the texture feature vector of the collected person at the down-sampling scale , is the value range of , is the number of elements of the texture feature vector at the down-sampling scale , , is the number of elements of the texture feature vector at the down-sampling scale ; the texture feature vectors of different down-sampling scales are used to calculate the similarity in turn, and multiple rounds of screening are performed according to the step-by-step screening ratio, so as to gradually and finely narrow down the range of legitimate users, so that the identity authentication set obtained finally is more targeted and accurate, further eliminates users with low similarity to the collected person, and improves the reliability of the authentication result.

[0163] the texture feature vector similarity of each down-sampling scale is sorted in turn, and the step-by-step screening legitimate user set is obtained according to the step-by-step screening ratio of each down-sampling scale, so as to obtain the identity authentication set, that is,

[0164]

[0165] wherein, is the identity authentication set, is the down-sampling scale , the number of legal users in the step-by-step screening legal user set is the face data of the th legal user in the step-by-step screening legal user set , the value range of , is the down-sampling scale , the number of legal users in the step-by-step screening legal user set is the face data of the th legal user in the step-by-step screening legal user set , the value range of , , is the similarity ranking of the texture feature vector of the th legal user in the step-by-step screening legal user set and the collector under the down-sampling scale , is the step-by-step screening ratio when the down-sampling scale is , the target legal user range is gradually reduced, is a ceiling function;

[0166] The similarity threshold is configured, the face feature vector of the collector is obtained, the final identity authentication of the collector is performed, and the face feature vector of each legal user in the identity authentication set and the face feature vector of the collector are determined according to the contribution degree of each vector element in the face feature vector in identity recognition to determine the importance score of each vector element, and the importance is sorted. For example, the importance score of each vector element can be obtained by training through a machine learning algorithm including random forest, gradient boosting tree, etc., and then sorted according to the score. By determining the contribution degree of each element in the face feature vector in identity recognition and sorting, the key features are highlighted, the subsequent similarity calculation is more focused on the features that have an important influence on identity recognition, and the accuracy and effectiveness of the authentication are improved.

[0167] The importance threshold is configured After the importance sorting is completed, the first vector elements in each face feature vector are selected according to the importance threshold to construct a key face feature vector; the key elements are selected according to the importance threshold to construct a key face feature vector, which simplifies the data dimension, reduces the calculation complexity, and retains the most representative feature information, which is helpful for more efficient similarity comparison.

[0168] Based on the importance score of each vector element within the key facial feature vector, the similarity between the key facial feature vector of each legitimate user in the authentication set and the key facial feature vector of the subject is calculated. For example, the formula for calculating the similarity of the key facial feature vector of the first legitimate user in the authentication set is as follows:

[0169]

[0170] in, It is the similarity between the collected data and the key facial feature vector of the first legitimate user in the identity verification set. It is the key facial feature vector. Importance score of each vector element The range of values ​​is , It is the key facial feature vector of the first legitimate user in the identity authentication set. It is the key facial feature vector of the subject being collected. It is the first key facial feature vector of the first legitimate user in the identity authentication set. One element, It is the first key facial feature vector of the subject. The system uses a number of elements; it calculates similarity based on importance scores, selects the most similar legitimate users as target legitimate users, and accurately determines the identity authentication result by comparing it with a similarity threshold, thus achieving accurate identification and confirmation of the identity of the user being collected.

[0171] The system selects the legitimate user from the identity verification set whose key facial feature vector is most similar to that of the person being captured, and marks them as the target legitimate user. If the similarity between the target legitimate user and the person being captured is greater than a similarity threshold, the identity verification is successful, and the person being captured is identified as the target legitimate user. Otherwise, the identity verification fails, and an authentication failure warning is issued, reminding the person being captured to re-capture their facial image. If authentication fails, timely warnings and reminders to re-capture facial images help improve the success rate of the next authentication. If authentication is successful, the identity of the person being captured is confirmed, ensuring the normal operation of the system and the legitimate rights and interests of the user.

[0172] Configure an identity authentication threshold and record the number of times the user authenticates. When the number of authentications exceeds the threshold, issue an authentication warning, send a notification to the administrator, temporarily lock the user account, and display stricter security verification prompts to prevent potential malicious attacks or abnormal situations. At the same time, it can also provide the user with additional identity verification methods to ensure the security and reliability of the system. Otherwise, no action is taken.

[0173] The card issuing control module is configured to send a driving instruction to the card cassette of the charging card issuing machine when the identity of the collector is authenticated, to control the card cassette to pop out a card, and to dynamically bind the identity of the card according to the user identity information of the collector.

[0174] In the embodiment, the card issuing control module performs corresponding operations according to the result of the identity authentication module. When the identity authentication is successful, the card issuing control module immediately starts the card issuing process. It first sends an instruction to the card cassette of the charging card issuing machine to drive the mechanical device in the card cassette to push out a charging card. At the same time, the user's identity information, such as name, ID number, and related use permission information, is written into the chip of the charging card. During the card issuing process, a warm prompt is displayed on the display screen to inform the user to take the card, and the use method and matters needing attention of the charging card are briefly explained in the form of animation or text, such as the position of the charging interface, the charging operation steps, the return time requirement, etc. If the identity authentication fails, the card issuing control module will not perform the card issuing operation. At this time, the display screen will display clear prompt information, and it is also responsible for monitoring and maintaining the state of the hardware devices of the charging card issuing machine, to ensure that the card issuing device, card cassette and other components are in normal operation, and to timely discover and handle possible abnormal situations such as paper jam and no card in the card cassette.

[0175] Preferably, the specific steps of dynamically binding the identity of the card include:

[0176] Receiving the identity information of the collector who has passed the authentication, including name, ID number, user account, user level, user permission;

[0177] Reading the unique identifier of the charging card to be issued, including card number, chip serial number, to ensure that each card has an independent identifier for subsequent binding and management;

[0178] Sending an activation signal to the charging card to be issued through the radio frequency identification or near field communication interface of the charging card issuing machine, to make the card enter a readable and writable state and prepare to receive data;

[0179] Using an encryption protocol to establish a secure communication channel between the charging card issuing machine and the charging card to be issued, to ensure the confidentiality and integrity of the data transmission process and prevent information from being stolen or tampered with;

[0180] According to the preset storage structure, a region for storing user identity information is divided in the chip storage area of the charging card to be issued, and initialization and formatting operations are performed to ensure the accuracy and consistency of data writing;

[0181] The obtained user identity information is written into the chip of the charging card byte by byte according to the preset encoding format and storage location; for example, the name is stored in the first 10 bytes, the ID number is stored in the next 20 bytes, etc.

[0182] After the user identity information is written, the written information of the to-be-issued charging card is read and compared with the user identity information to perform data verification and check whether there is data error or loss. For example, checksum, CRC, or other methods are used to perform data integrity verification;

[0183] If the data verification passes, a confirmation instruction is sent to the to-be-issued charging card, and the to-be-issued charging card returns a confirmation signal, indicating that the identity information and the permission information have been successfully bound;

[0184] The relevant information of this binding operation is recorded, including the user identity, the card identifier, the binding time, and the like, so as to facilitate subsequent query and management;

[0185] The state of the issued card in the card cassette of the charging card issuing machine is marked as bound and issued, and the number and position information of the remaining cards in the card cassette are updated, so as to monitor the inventory of the card cassette in real time.

[0186] Embodiment 2

[0187] Please refer to Figure 7 This embodiment introduces a face multi-feature fusion identity authentication method of a charging card issuing machine, including the following steps:

[0188] When it is detected that the collector enters the collection area, the ambient light intensity is acquired through the light sensor of the charging card issuing machine, the camera parameters of the charging card issuing machine are automatically adjusted based on the ambient light intensity after adaptive light source light compensation, and the position and angle of the collector are automatically detected to collect the face image;

[0189] The feature points of the collected face image are detected, the face features of the face image are extracted based on the feature points, including geometric features and texture features, the face features are fused to generate a face feature vector;

[0190] The collector is authenticated based on a legal user face data set, a preliminary legal user set is obtained through rapid screening of the geometric feature vector of the collector, an identity authentication set is screened out according to the step-by-step screening ratio of each down-sampling scale and the texture feature vector similarity order, and then a key face feature vector is constructed to screen out a target legal user, so as to judge whether the identity authentication of the collector is successful;

[0191] When the identity authentication of the collector passes, a driving instruction is sent to the card cassette of the charging card issuing machine to control the card cassette to eject the card, and the card identity is dynamically bound according to the user identity information of the collector.

[0192] Specifically, the step of generating the face feature vector includes:

[0193] Obtaining a face image, performing feature point detection on the face image, the feature points including: eye feature points, nose feature points, mouth feature points and face contour feature points;

[0194] Extracting geometric features of the face image based on the feature points of the face image, including: distance features, angle features and proportion features;

[0195] Selecting a diagonal length of the face image as a scale factor, and performing normalization processing on the distance features in the geometric features based on the scale factor, to eliminate the influence of the size of the face image on the distance features;

[0196] Sorting the normalized distance features, angle features and proportion features in the order of the face image from top to bottom and from left to right, respectively, to construct a geometric distance feature vector, a geometric angle feature vector and a geometric proportion feature vector;

[0197] Performing feature fusion on the geometric features, and sequentially concatenating the geometric distance feature vector, the geometric angle feature vector and the geometric proportion feature vector to construct a geometric feature vector;

[0198] Performing grayscale processing on the face image, and performing multi-scale down-sampling on the face image after the grayscale processing to obtain a face image at each down-sampling scale, for texture feature extraction, to obtain a texture feature vector;

[0199] Concatenating the geometric feature vector and the texture feature vector to obtain a face feature vector.

[0200] Specifically, the steps of obtaining the texture feature vector include:

[0201] Selecting key feature points based on the feature points, and detecting the key feature points on the face image at each down-sampling scale;

[0202] Setting a neighborhood window size according to the down-sampling scale, and calculating the gradient direction and gradient amplitude of each pixel point relative to the detected key feature points in the neighborhood window;

[0203] Dividing the neighborhood window into regions to obtain neighborhood sub-regions, and counting the gradient direction histogram in each neighborhood sub-region, and connecting the gradient direction histograms of all neighborhood sub-regions to form the texture features of the key feature points;

[0204] For each down-sampling scale, constructing a corresponding sampling scale texture feature vector according to the type of the key feature points, and concatenating the texture feature vectors of the various sampling scales in the order of the down-sampling scales from small to large to obtain the texture feature vector.

[0205] Specifically, the specific steps of identity authentication include:

[0206] A legal user face data set is denoted as wherein, is the face data of the i-th legal user, containing the geometric feature vector of the i-th legal user, the texture feature vector at each down-sampling scale, and the face feature vector; A fast screening ratio is configured, the geometric feature vector of the collector is obtained, and the legal user face data set is fast screened to calculate the similarity of the geometric feature vector of the collector and each legal user in the face database set;

[0207] The similarity of the geometric feature vector of the collector and the legal user in the face database set is sorted, and a preliminary legal user set is obtained according to the fast screening ratio;

[0208] A step-by-step screening ratio at each down-sampling scale is configured, the texture feature vector of the collector at each down-sampling scale is obtained, and the preliminary legal user set is step-by-step screened to sequentially calculate the similarity of the texture feature vector of the collector at each down-sampling scale and each legal user in the fast screening preliminary legal user set;

[0209] The texture feature vector similarity at each down-sampling scale is sequentially sorted, and a step-by-step screening legal user set is obtained according to the step-by-step screening ratio at each down-sampling scale, so as to obtain an identity authentication set.

[0210] Specifically, the specific steps of identity authentication further include:

[0211] A similarity threshold is configured, the face feature vector of the collector is obtained, and the collector is finally identity authenticated, the face feature vector of each legal user in the identity authentication set and the face feature vector of the collector are sorted according to the importance score of each vector element in the face feature vector in identity recognition, and the importance is sorted;

[0212] An importance threshold is configured After the importance sorting is completed, the first K vector elements in each face feature vector are selected according to the importance threshold to construct a key face feature vector;

[0213] The similarity of the key face feature vector of each legal user in the identity authentication set and the key face feature vector of the collector is calculated according to the importance score of each vector element in the key face feature vector;

[0214]

[0215] ​​​Screening the legal user with the largest similarity with the key facial feature vector of the collector in the identity authentication set, marking as the target legal user, if the similarity between the target legal user and the key facial feature vector of the collector is greater than the similarity threshold, the identity authentication is successful, otherwise the identity authentication fails;

[0216] Configuring an identity authentication threshold, recording the number of identity authentication of the collector, when the number of identity authentication is greater than the identity authentication threshold, identity authentication warning is performed, otherwise no operation is performed.

[0217] Working principle and effect:

[0218] When the face multi-feature fusion charging card dispenser identity authentication system is working, the face collection module monitors the collection area with the help of an infrared sensor, and starts collection when a human body is recognized and the recognition success rate meets the standard. According to the collection distance change rate of the distance sensor and the ambient light intensity of the light sensor, the light supplement device and the camera parameters are adjusted adaptively. The camera captures the body contour, and determines whether to collect images according to the head position and angle deviation. After image collection, the image quality is evaluated, and if it does not meet the standard, it is re-collected, and if the number of times exceeds, it is warned. The feature fusion module detects feature points of the collected face image, extracts eye, nose, mouth and face contour feature points, extracts geometric features based on these feature points, normalizes the distance features with the diagonal length of the image, constructs geometric distance, angle and proportion feature vectors, and splices them into a geometric feature vector; meanwhile, the image is processed for grayscale, and the texture feature vector of the face image at each down-sampling scale is calculated based on the key feature points. Finally, the geometric and texture feature vectors are spliced to generate a face feature vector, realizing multi-feature fusion and providing comprehensive and accurate feature data for identity authentication. The identity authentication module is based on a legal user face database. First, a preliminary legal user set is quickly screened according to the geometric feature vector, and then the identity authentication set is obtained according to the texture feature vector similarity and the gradual screening ratio. Then the importance of the face feature vector elements is determined and sorted, a key face feature vector is constructed, the similarity is calculated to screen the target legal user, compared with the threshold to determine the authentication result, the authentication times are recorded, and if the threshold is exceeded, a warning is given. This multi-stage and refined authentication method effectively improves the authentication accuracy and security. After the authentication is passed, the card dispenser control module receives the user identity information, reads the to-be-issued card identifier, activates the card and establishes an encrypted communication channel, writes and verifies the identity information, confirms the binding, records the information, updates the card box state, realizes the dynamic binding of the card and the user identity, and ensures the accuracy and safety of the charging card dispenser.

[0219] Through the close cooperation of each module, from accurate face image collection, multi-feature fusion extraction to rigorous identity authentication and safe card binding, the accuracy, security and intelligence level of the charging card dispenser identity authentication are effectively improved, providing users with convenient and reliable service experience.

[0220] If the technical solutions of the present disclosure involve personal information, the product applying the technical solutions of the present disclosure has been explicitly informed of the personal information processing rules before processing the personal information and has obtained the personal independent consent. If the technical solutions of the present disclosure involve sensitive personal information, the product applying the technical solutions of the present disclosure has obtained the personal independent consent before processing the sensitive personal information and at the same time meets the requirement of "explicit consent". For example, at the personal information collection device such as a camera, a clear and prominent sign is set to inform that the personal information collection range has been entered and the personal information will be collected. If the person voluntarily enters the collection range, it is considered that the person agrees to collect the personal information. Or, on the device for processing personal information, the personal information processing rules are informed by using obvious signs / information, and the personal authorization is obtained by means of pop-up information or asking the person to upload the personal information by himself / herself. The personal information processing rules can include the personal information processor, the purpose of processing personal information, the processing method, and the type of processed personal information, etc.

[0221] The above only describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as the protection scope of the present application.

Claims

1. A face multi-feature fusion charging card issuer identification authentication system, characterized in that, The face collection module, the feature fusion module, the identity authentication module and the card issuing control module are included. The face collection module is used for acquiring the ambient light intensity through the light sensor of the charging card issuing machine when detecting that the collector enters the collection area, automatically adjusting the camera parameters of the charging card issuing machine after adaptive light source compensation based on the ambient light intensity, and collecting the face image through automatic detection of the position and angle of the collector. The feature fusion module is used for detecting feature points of the collected face image, extracting facial features including geometric features and texture features based on the feature points, and performing feature fusion on the facial features to generate a facial feature vector. The texture feature extraction includes: The face image is subjected to gray processing, and the gray-processed face image is subjected to multi-scale down-sampling to obtain a face image of each down-sampling scale; key feature points are detected for the face image of each down-sampling scale based on the selected key feature points; a neighborhood window size is set according to the down-sampling scale, and the gradient direction and gradient amplitude of each pixel point relative to the key feature points are calculated in the neighborhood window with the detected key feature points as the center; the neighborhood window is divided into regions to obtain a neighborhood sub-region, and the gradient direction histogram is counted in each neighborhood sub-region; the gradient direction histograms of all neighborhood sub-regions are connected to form the texture feature of the key feature points; for each down-sampling scale, a corresponding sampling scale texture feature vector is constructed according to the type of the key feature points, and the texture feature vectors of various sampling scales are spliced in the order of the down-sampling scale from small to large to obtain the texture feature vector; The identity authentication module is used for performing identity authentication on the collector based on a legal user face data set, quickly screening the collector through the geometric feature vector to obtain a preliminary legal user set, screening out an identity authentication set according to the step-by-step screening ratio of each down-sampling scale and the texture feature vector similarity order, then constructing a key facial feature vector, and screening out a target legal user to determine whether the identity authentication of the collector is successful. The specific steps of the identity authentication include: The legal user face data set is denoted as, wherein is the face data of the th legal user, and contains the geometric feature vector of the th legal user, the texture feature vector at each down-sampling scale, and the face feature vector; a fast screening ratio is configured, the geometric feature vector of the collector is obtained, the legal user face data set is fast screened, and the similarity of the geometric feature vector of the collector and each legal user in the face database set is calculated; the similarity of the geometric feature vector of the collector and the legal user in the face database set is sorted, the preliminary legal user set is obtained according to the fast screening ratio; a step-by-step screening ratio of each down-sampling scale is configured, the texture feature vector of each down-sampling scale of the collector is obtained, the preliminary legal user set is step-by-step screened, and the similarity of the texture feature vector of each down-sampling scale of the collector and each legal user in the fast screening preliminary legal user set is calculated in turn; the texture feature vector similarity of each down-sampling scale is sorted in turn, the step-by-step screening legal user set is obtained according to the step-by-step screening ratio of each down-sampling scale, and the identity authentication set is obtained; a similarity threshold is configured, the face feature vector of the collector is obtained, the final identity authentication of the collector is performed, the face feature vector of each legal user in the identity authentication set and the face feature vector of the collector are sorted according to the importance score of each vector element in the face feature vector in the identity recognition, and the importance is sorted; an importance threshold is configured, after the importance sorting is completed, the first vector elements in each face feature vector are selected according to the importance threshold, and a key face feature vector is constructed; the similarity of the key face feature vector of each legal user in the identity authentication set and the key face feature vector of the collector is calculated according to the importance score of each vector element in the key face feature vector; the legal user with the greatest similarity to the key face feature vector of the collector in the identity authentication set is screened, and is marked as a target legal user; if the similarity of the target legal user to the key face feature vector of the collector is greater than the similarity threshold, the identity authentication is successful, otherwise, the identity authentication fails; an identity authentication threshold is configured, the identity authentication times of the collector are recorded, when the identity authentication times are greater than the identity authentication threshold, an identity authentication warning is performed, otherwise, no operation is performed; The card issuing control module is configured to send a driving instruction to the card cassette of the charging card issuing machine when the identity authentication of the collector is passed, control the card cassette to pop out a card, and dynamically bind the identity of the card according to the user identity information of the collector. 2.The face multi-feature fusion charging kiosk identification authentication system of claim 1, wherein, The step of collecting the face image comprises: The infrared sensor of the charging card issuing machine continuously monitors the collection area to obtain the features of the object entering the collection area, and performs human body recognition based on the features of the object; When the human body is recognized, a monitoring time window is configured, human body monitoring is performed, the number of times that the human body is recognized in the monitoring time window is counted, and the human body recognition success rate is calculated; A human body recognition threshold is configured, when the human body recognition success rate is greater than the human body recognition threshold, the face collection process is started, otherwise, no operation is performed; A collection trigger threshold is configured, the collection distance between the collector and the camera of the charging card machine is obtained by the distance sensor, the collection distance change rate is obtained, when the collection distance change rate is less than the collection trigger threshold, the light supplement device parameter adjustment and the camera parameter adjustment are triggered; After the light supplement device parameter and the camera parameter adjustment, the body contour information of the collector is captured by the camera, and the key point coordinates of the human body contour are extracted; According to the key point coordinates of the human body contour, the horizontal offset and the vertical offset of the head relative to the image center are calculated, and the tilt angle of the head of the collector is calculated; The position deviation range and the angle deviation range are configured, when the horizontal offset of the head relative to the image center, the vertical offset of the head relative to the image center are within the position deviation range, and the tilt angle of the head is within the angle deviation range, the face image collection is performed, otherwise, according to the horizontal offset of the head relative to the image center, the vertical offset of the head relative to the image center and the tilt angle of the head, the voice prompt is performed to guide the collector to adjust the position and the angle. 3.The face multi-feature fusion charging card issuer identification authentication system of claim 2, wherein, The step of collecting the face image further comprises: The collected face image is preliminarily quality evaluated, the face image definition, the face image integrity and the face image brightness uniformity are obtained; The face image definition is obtained by calculating the variance of the image Laplace operator; The face image integrity is obtained by calculating the integrity and the missing proportion of the face contour key points; The face image brightness uniformity is obtained by calculating the brightness standard deviation of different regions of the face image; The definition threshold, the integrity threshold and the brightness threshold are configured, when the face image definition is less than the definition threshold, or the face image integrity is less than the integrity threshold, or the face image brightness uniformity is less than the brightness threshold, the image collection is retriggered, otherwise, no operation is performed; The re-collection threshold is configured, the number of retriggered image collection is monitored, when the number of retriggered image collection is greater than the re-collection threshold, the collection process is suspended, and the collection warning is performed. 4.The face multi-feature fusion charging card issuer identification authentication system of claim 3, wherein, The step of adjusting the light supplement device parameter comprises: The illumination intensity threshold is configured, including the upper limit threshold of the illumination intensity and the lower limit threshold of the illumination intensity, the ambient illumination intensity is obtained by the light sensor of the charging card machine; When the ambient illumination intensity is less than the lower limit threshold of the illumination intensity, the light supplement brightness and the light supplement intensity are enhanced, the initial light supplement brightness is adjusted according to the proportion of the current ambient illumination intensity and the lower limit threshold of the illumination intensity, and the light supplement brightness is optimized in combination with the position of the face in the image; When the ambient illumination intensity is greater than or equal to the upper limit threshold of the illumination intensity, the light supplement brightness is reduced, on the basis of the initial light supplement intensity, the light supplement brightness is reduced in proportion to the degree that the illumination intensity exceeds the upper limit threshold of the illumination intensity, until it is reduced to the minimum light supplement brightness; After adjusting the light supplement brightness, the light supplement angle is adjusted according to the offset direction of the face center relative to the image center on the basis of the initial light supplement angle. 5.The face multi-feature fusion charging card issuer identification authentication system of claim 4, wherein, The step of adjusting the camera parameter comprises: The environment light intensity after the parameter adjustment of the light supplement device is acquired, and if the environment light intensity acquired by the light sensor of the charging card machine is still greater than the upper limit threshold of the light intensity under the light supplement brightness after the light supplement device is reduced, the camera exposure parameter is adjusted, and the camera sensitivity is reduced in proportion according to the degree that the environment light intensity exceeds the upper limit threshold; The camera focal length is adjusted according to the distance between the collector and the camera; The camera aperture is adjusted based on the initial aperture according to the difference between the current environment light intensity and the reference light intensity, and the camera aperture is increased when the environment light intensity is higher than the reference light intensity, and vice versa. The camera shutter speed is adjusted based on the environment light intensity and the collection distance, and the shutter speed is reduced in proportion when the environment light intensity is greater than the reference light intensity based on the initial shutter speed, and the collection distance is inversely proportional to the shutter speed. 6.The face multi-feature fusion charging kiosk identification authentication system of claim 1, wherein, The face feature vector generation step includes: Acquiring a face image, detecting feature points of the face image, the feature points including eye feature points, nose feature points, mouth feature points and face contour feature points; Extracting geometric features of the face image based on the feature points of the face image, including distance features, angle features and proportion features; Selecting the diagonal length of the face image as a scale factor, and normalizing the distance features in the geometric features based on the scale factor to eliminate the influence of the size of the face image on the distance features; Sorting the normalized distance features, angle features and proportion features in the order of from top to bottom and from left to right of the face image to construct a geometric distance feature vector, a geometric angle feature vector and a geometric proportion feature vector; Concatenating the geometric distance feature vector, the geometric angle feature vector and the geometric proportion feature vector in sequence to construct a geometric feature vector; Performing grayscale processing on the face image, and performing multi-scale down-sampling on the face image after grayscale processing to obtain a face image at each down-sampling scale for texture feature extraction to obtain a texture feature vector; Concatenating the geometric feature vector and the texture feature vector to obtain a face feature vector.

7. The face multi-feature fusion charged card issuing machine identification authentication system according to claim 6, characterized in that, The texture feature vector acquisition step includes: Selecting key feature points based on the feature points, and detecting key feature points of the face image at each down-sampling scale; Setting a neighborhood window size according to the down-sampling scale, taking the detected key feature points as the center, and calculating the gradient direction and gradient amplitude of each pixel point relative to the key feature points in the neighborhood window; Dividing the neighborhood window into regions to obtain a neighborhood sub-region, and counting the gradient direction histogram in each neighborhood sub-region, and connecting the gradient direction histograms of all neighborhood sub-regions to form the texture features of the key feature points; For each down-sampling scale, a corresponding sampling scale texture feature vector is constructed according to the type of the key feature points, and the texture feature vectors of various sampling scales are concatenated in the order of the down-sampling scales from small to large to obtain a texture feature vector.

8. The face multi-feature fusion charging card dispenser identity authentication method based on the face multi-feature fusion charging card dispenser identity authentication system of any one of claims 1-7, wherein The method includes the following steps: When the collector is detected to enter the collection area, the ambient light intensity is obtained by the light sensor of the charging card machine, the camera parameters of the charging card machine are automatically adjusted after adaptive light source light compensation based on the ambient light intensity, and the face image is collected by automatically detecting the position and angle of the collector; The feature points of the collected face image are detected, the face features of the face image are extracted based on the feature points, including geometric features and texture features, the face features are fused to generate a face feature vector; The identity of the collector is authenticated based on a legal user face data set, the initial legal user set is obtained by quickly screening the geometric feature vector of the collector, the identity authentication set is screened according to the step-by-step screening ratio of each down-sampling scale and the texture feature vector similarity degree, then the key face feature vector is constructed, the target legal user is screened out, and whether the identity authentication of the collector is successful is judged; When the identity authentication of the collector is passed, a driving instruction is sent to the card box of the charging card machine to control the card box to eject the card, and the card identity is dynamically bound according to the user identity information of the collector.

Images