Identity authentication method and device

Abstract

The invention discloses an identity authentication method and device. The method comprises the following steps: performing feature extraction and classification processing on a collected face image of a driver; identifying whether the classified image has a face area, and if not, comparing the illumination intensity with a first illumination threshold and a second illumination threshold; when the illumination intensity is greater than a first illumination threshold, performing brightness enhancement on the face image, and when the illumination intensity is less than a second illumination threshold, triggering an infrared light supplement lamp to supplement light; performing identity detection on each frame in the continuous frame number of the recognized face image, and recording a frame number after passing the identity detection; the recorded frame number k is counted, and when the k value is larger than a preset threshold value, verification is passed. According to the method, fraud means such as photos, videos, printed photos and screen copying are effectively resisted, and effective face living body detection is realized in the field of driver identity verification.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of manned forklift assisted driving, and particularly relates to an identity authentication method and device. BACKGROUND

[0002] With the development of intelligent technology and the market demand for forklifts, the forklift industry requires that the driver's identity be authenticated before the driver operates.

[0003] In the field of forklift operation, fraud means such as videos, printed photos, and screen flipping are commonly used, and there is no special "liveness detection" means for driver identity verification. In order to ensure operation safety, current enterprises generally rely on foreign supporting manufacturer products or foreign chip-based design products, which are expensive in cost and price, resulting in low selection of forklift industry. Based on low-cost domestic chip platform, there is a lack of lightweight model and algorithm deployment, and the market lacks corresponding forklift driver face liveness anti-fraud detection products. SUMMARY

[0004] To solve the problems in the prior art, the present application provides an identity authentication method and device.

[0005] In a first aspect, the present application provides an identity authentication method, comprising:

[0006] performing feature extraction and classification processing on the collected driver face image;

[0007] identifying whether the processed image has a face region, and if not, comparing the light intensity with the first and second illumination threshold values; when the light intensity is greater than the first illumination threshold value, enhancing the brightness of the face image, and when the light intensity is less than the second illumination threshold value, triggering the infrared fill light to fill light;

[0008] performing identity detection on each frame of the recognized face image, and after the detection, recording a frame number k;

[0009] statistically recording the frame number k, and when the value of k is greater than a preset threshold value, passing the verification.

[0010] Optionally, the collected driver face image is an IR image and an RGB image obtained by using an RGB and IR dual-camera.

[0011] Optionally, the classification processing comprises two classifications of feature vector prediction value and feature vector label value on the feature-extracted face image.

[0012] The inter-class distance is increased and the intra-class distance is reduced.

[0013] Optionally, the increase of the inter-class distance and the reduction of the intra-class distance are calculated by the following formula:

[0014]

[0015] wherein, L is the distance between the feature vector prediction value and the feature vector label value, λ is the weight parameter, N is the sample quantity, C is the class quantity, y c (n) is the label of the nth sample, is the actual value of the nth sample, y i is the sample classification label of the ith class, c yi is the midline point of the sample classification label of the ith class, T is the transpose operation on the current matrix, xi is the feature vector of the N samples.

[0016] Optionally, the brightness enhancement of the face image comprises:

[0017] The size relationship of the face image in the pixel mapping process is kept unchanged, and the contrast is increased;

[0018] The value range of the pixel mapping function is ensured to be between 0 and 255;

[0019] The face image is nonlinearly stretched, and the pixel value of the face image is redistributed;

[0020] The visual effect of the face image is improved by adjusting the brightness distribution of the face image.

[0021] Optionally, the expression of the pixel mapping function is as follows:

[0022]

[0023] wherein, n is the total sum of pixels in the face image, S k is the pixel format of k-level gray scale, and L is the total number of gray scales of the face image.

[0024] Optionally, the redistribution of the pixel value of the face image comprises:

[0025] The pixel probability distribution of the original gray scale of the face image is calculated;

[0026] The cumulative distribution function of the face image is obtained according to the pixel probability distribution;

[0027] The face image after transformation is obtained according to the mapping function.

[0028] Optionally, the adjustment of the brightness distribution of the face image comprises:

[0029] The product operation is performed on each pixel value of the original face image, and the operation expression is as follows:

[0030] s = Cr γ

[0031] wherein C and γ are positive constants, r is a pixel value of a corresponding position of the face image, and s is a pixel value of the face image after the operation;

[0032] normalizing the operation result to a range of [0, 255];

[0033] correcting the face image after the operation to a picture with a preset range threshold gray scale.

[0034] In a second aspect, an identity authentication device is provided, comprising:

[0035] an image processing module configured to perform feature extraction and classification processing on the collected face image of the driver;

[0036] an image brightness enhancement module configured to identify whether the image after the classification processing has a face region, and if not, compare the illumination intensity with a first illumination threshold and a second illumination threshold; when the illumination intensity is greater than the first illumination threshold, perform brightness enhancement on the face image, and when the illumination intensity is less than the second illumination threshold, trigger the infrared fill light to fill light;

[0037] an identity detection module configured to perform identity detection on each frame in the continuous frame number of the recognized face image, and after the identity detection, record a frame number;

[0038] a statistical verification module configured to statistically record the frame number k, and when the k value is greater than a preset threshold, pass the verification.

[0039] The technical solutions provided by some embodiments of the present application have at least the following beneficial effects:

[0040] 1. The identity authentication anti-fraud demand of the driver in the forklift industry is solved;

[0041] 2. The product cost is greatly reduced by adapting to the embedded domestic chip, the forklift intelligent auxiliary safety driving product is popularized, and the driving behavior of the forklift driver is standardized;

[0042] 3. The safety and efficiency of forklift operation are improved.

[0043] Other features and advantages of the present application will be described in the following description, and some will become apparent from the description, or will be understood from the practice of the present application. The purposes and other advantages of the present application are realized and obtained by the structures specifically pointed out in the specification, claims, and drawings.

[0044] In order to make the above purposes, features and advantages of the present application more apparent and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are referred to for detailed description.

[0045] Advantages of the additional aspects of the application will become apparent in light of the following description. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments of the present application or the prior art description. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0047] Figure 1 The identity authentication method flow chart shown in the embodiments of the present application;

[0048] Figure 2 The binocular camera imaging principle diagram shown in the embodiments of the present application;

[0049] Figure 3 The FaceBagNet three modal algorithm model structure diagram shown in the embodiments of the present application;

[0050] Figure 4 The optimized deep learning model diagram of the dual modal small area picture shown in the embodiments of the present application;

[0051] Figure 5 The IR camera picture under strong light shown in the embodiments of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical scheme and advantages of the present application more clear, the following will further describe the embodiments of the present application in combination with the drawings.

[0053] The following description refers to the accompanying drawings. Unless otherwise noted, like numbers in different drawings refer to like or similar elements. The following examples of embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of methods and apparatuses consistent with some aspects of the present application as detailed in the appended claims.

[0054] In the description of the present application, it should be understood that the terms "first", "second" and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood in specific cases. In addition, in the description of the present application, "multiple" refers to two or more than two. "And / or", the association relationship of the associated object, indicates that there can be three relationships, for example, A and / or B, which can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. The character " / " generally represents that the front and rear associated objects are a "or" relationship.

[0055] The existing intelligent auxiliary driving device technology is mainly used in the automobile industry, and the driver assistance system is used to automatically understand and represent the behavior of the driver by using image processing technology and artificial intelligence method. A deep learning method with a large model is used, including image classification and target detection method, which requires powerful GPU computing platform for calculation and relies on foreign hardware processing platform.

[0056] The forklift industry driver identity authentication system mainly verifies the operation permission of the driver through biological information such as fingerprint and face feature, or a medium uniquely bound with personal identity information such as magnetic card. When the collector is invalid, removed or the driver information is incorrect, the vehicle cannot be started, thereby effectively preventing unauthorized personnel from operating the forklift. However, there is currently a lack of driver identity verification technology and device with anti-fraud detection function to ensure the legal identity of the driver, standardize his operation behavior, and prevent safety risks caused by improper behavior or fraudulent activities.

[0057] The present application is directed to a forklift industry driver face identity authentication system, and designs an identity verification method and device for anti-fraud face living body detection, based on data collection in the forklift operation scene, by capturing RGB and infrared binocular camera data, constructing a dual-mode living body detection model, and deploying it on a domestic embedded ARM low-cost hardware platform, to realize a forklift driver anti-fraud identity authentication device. The device of the present application is widely used in forklift safety management of enterprises of various scales, and can improve the safety and efficiency of forklift operation by installing the device in large manufacturing enterprises or small production workshops.

[0058] Specifically, based on the RGB and infrared binocular alignment images captured by the forklift driver face recognition binocular camera, it is detected whether there is a face region image in the image; according to the detected face region image, a dual-mode intermediate fusion method is designed, the RGB and infrared data are first converted into high-dimensional features, and then fused with the high-dimensional features in the intermediate layer of the model, a flexible feature extraction branch optimization method and a lightweight living body detection model are designed, and the living body anti-fraud detection function can be realized for screen photos, videos and printed photos in the image.

[0059] The application installs an RGB or infrared camera in the middle position of the left front column of the forklift, selects a domestic embedded low-cost chip-on-controller, the controller is automatically started after starting, realizes driver live body fraud detection, when a face picture of illegal fraud appears, the fraud frequency is counted, voice reminding is performed, and the fraud face picture is locked (after locking, the face is not directly detected, and needs to be unlocked according to the device configuration time parameter), and short video storage of 10 seconds before and after the fraud scene is stored.

[0060] Embodiment one

[0061] The application provides an identity verification method, as shown in the figure, comprising the following steps: Figure 1

[0062] Step S1: feature extraction and classification processing are performed on the collected face image of the driver;

[0063] Specifically, RGB and IR face data of the forklift driver in the working scene are collected, a dual-mode fraud detection model and a single-mode detection model are designed, feature extraction is performed on the image, and then the image is mapped to a nonlinear high-dimensional space for classification processing; the specific process is as follows:

[0064] Step S1.1: the forklift is powered on, the face recognition system is started and continuously reads the driver monitoring binocular camera data, and aligned RGB and IR images are obtained;

[0065] Specifically, the forklift driver monitoring camera in the embodiment is a RGB and IR binocular low-cost ordinary camera. Since the installation physical positions of the optical centers of the RGB and IR cameras are different, the RGB and IR original image data are offset. Considering that the binocular live body detection model needs to ensure that the captured RGB and IR data are aligned, a camera alignment algorithm is designed to make the obtained RGB and IR images have no deviation, and the specific process is as follows:

[0066] The coordinates of any certain P in the three-dimensional space under the camera coordinate system are P(X, Y, Z), and the physical coordinates of the corresponding point P1 in the camera imaging plane are P1(x, y), f x , f y is the focal length;

[0067] In the camera coordinate system, the three-dimensional space coordinates of point P are (Xc, Yc, Zc), wherein,

[0068] X c = (u-c x ) x d / f x …………………………(1)

[0069] Y c = (v-c y ) x d / f​y …………………………(2)

[0070] Z c = d ……………………………………(3)

[0071] Wherein, u and v are respectively P point in camera image pixel horizontal coordinate and vertical coordinate, d is P point in camera coordinate system measured depth value; f x And f y It is the pixel focal length of horizontal and vertical direction of depth camera respectively, c x And c y It is the coordinate of depth camera optical center in pixel coordinate system, d / f x It is the unit length of a pixel point in x axis, and d x ; d / f y It is expressed as a pixel point in y axis unit length, and d y ;

[0072] According to matrix method:

[0073]

[0074] Wherein,

[0075] It is monocular camera imaging matrix, also the camera itself inherent coefficient, the application directly uses the camera manufacturer given, also can be calibrated by Zhang Zhengyou calibration method and obtains.

[0076] According to the above reasoning mode, for binocular camera imaging, since the installation position of two cameras is different, there is error in camera coordinate system.But it is observed that the object relative world coordinate system (absolute position) does not change.Based on the invariance of world coordinate system position, design binocular camera calibration matrix.

[0077] As Figure 2 Shown, let left RGB camera be L, right IR camera be R, left and right interval be l, set imaging plane center point pixel as (c Lx ,c Ly ), d Lx It is the unit length of a pixel on the imaging plane in x axis, d Ly It is the unit length of a pixel on the imaging plane in y axis.World coordinate system under any point P (X, Y, Z) in left camera pixel coordinate (u L ,v L ), derive the position of the point under RGB camera:

[0078]

[0079] Similarly, set the right IR camera, focal length f R , imaging plane center pixel coordinates (c Rx , c Ry ), d Rx is the unit length of a pixel on the imaging plane in the x-axis, d Ry is the unit length of a pixel on the imaging plane in the y-axis. The world coordinate system under any point P (X, Y, Z) in the right camera pixel coordinates (u R , v R ), derive the point in the IR camera position:

[0080]

[0081] According to formula (6) and (7) calculation, deduce the transformation relationship between the camera RGB coordinate system relative to the camera IR coordinate system:

[0082]

[0083] The conversion to matrix form is as follows:

[0084]

[0085] Indicated as: Form,

[0086] Wherein, is the transformation matrix, is the offset matrix.

[0087] According to the transformation matrix of A and B, the IR camera imaging point position is obtained by traversing the RGB camera imaging point position, so as to complete the binocular camera alignment algorithm work.

[0088] Step S1.2: Detect whether there is a face in the image through the face recognition algorithm, and determine the face image area under the premise of having a face; Then, based on the open source dataset CASIA-SURF and FaceBagNet algorithm, the features of the collected driver's face image are extracted and classified;

[0089] Specifically, according to the detected RGB and IR aligned driver face frame position, based on the open source FaceBagNet model training algorithm, the forward calculation and back propagation of the convolutional neural network and the loss function are optimized and designed.

[0090] Based on the open source FaceBagNet algorithm, experiments are done on the CASIA-SURF dataset, which involves three different face modalities: RGB, Depth, IR. When the three modalities are fused for training, the model size is as high as 178MB, such as Figure 3as shown.

[0091] Considering that the device runs on low-cost domestic chips, it is necessary to design a set of end-to-end lightweight, accurate, and real-time algorithm models. At the same time, considering the cost problem of the forklift driver monitoring camera, the RGB and IR binocular camera mode is adopted, and the dual-modal data input needs to be designed.

[0092] In the forward calculation, the network structure (convolutional layer and pooling layer) extracts the features, the fully connected layer calculates the feature vector, and then the corresponding prediction value is obtained, and the loss function is used to measure the distance between the prediction value and the actual label value. Considering that the scope of the attack may involve the whole face region, a plurality of small regions are randomly or uniformly cut out from the face picture and input into the CNN for feature extraction.

[0093] In this embodiment, the uniform cutting of a plurality of small regions is adopted, and each small region is uniformly selected from the RGB and IR images. As shown in Figure 4 For the input cropped small region RGB and IR data, the data input is 48*48*3, and the convolution transformation is obtained through the first layer convolution to obtain 48*48*32 features. Then, three groups of convolution layers are used, and the bottleneck operation is performed to obtain a 6*6*128 feature map. Finally, it is expanded into a 2048-dimensional feature vector, and a fully connected layer is established, and a softmax function is used for binary classification.

[0094] In the living body detection, since the real face and the high-simulation face are very similar, it is necessary to increase the inter-class distance and reduce the intra-class distance, and the loss function needs to be optimized and designed. The method of this embodiment designs two parts for the loss function: the softmax cross-entropy loss function and the square distance loss function.

[0095] (1) The softmax cross-entropy loss function is expressed as follows:

[0096]

[0097] Where N represents the number of samples, C represents the number of categories, y c (n) represents the label of the nth sample, which takes the value of 0 or 1, represents the actual value of the nth sample, where T represents the transpose operation on the current matrix W, and the second summation in formula (10) can be written in the form of vector inner product:

[0098]

[0099] Where,

[0100] (2) The square distance loss function is expressed as follows:

[0101] In classification training, let the classification label of the sample of category i be y i , this category has a midline point c yi , c yi It is also the feature vector x of N samples i The center point of . Use distance to measure the feature vector x i with c yi The relationship between them is expressed as square distance d:

[0102]

[0103] The present invention corrects the loss function Lcenter to distance the positive and negative categories and eliminates the unnecessary distance gap. The design is as follows:

[0104]

[0105] This application is submitted by L softmax With L center Perform fusion to obtain the improved loss function:

[0106]

[0107] Where λ is a weight parameter for adjusting the softmax cross entropy loss function and the center loss function in order to adjust the optimal network model. In this embodiment, λ is set to 0.2.

[0108] In back propagation, by setting the learning rate and selecting the optimizer to perform derivative operations, the weight W and the bias value b update amount are calculated, and then the update iteration is performed to obtain the optimal model. This belongs to the existing technology in this field and is not described in this invention.

[0109] Step S2: Identify whether the image after classification processing contains a face area. If not, compare the light intensity with the first illumination threshold and the second illumination threshold; when the light intensity is greater than the first illumination threshold, enhance the brightness of the face image; when the light intensity is less than the second illumination threshold, trigger the infrared fill light to fill in the light;

[0110] In the specific implementation process, when working in a strong light environment, the brightness of the image obtained by the IR camera is extremely low, such as Figure 5 As shown in the figure, the human eye cannot recognize the face at all. The algorithm cannot recognize the face frame directly.

[0111] To improve the system's adaptability to working environments with strong light, this application designs an IR image brightness enhancement algorithm to improve facial area recognition performance. Specifically, the algorithm includes the following steps:

[0112] Step S2.1: Keep the size relationship of the face image unchanged during pixel mapping and increase the contrast;

[0113] Specifically, the histogram equalization is used to ensure that the original size relationship remains unchanged during the image pixel mapping process, that is, the brighter area is still brighter and the darker area is still darker, only the contrast is increased, and the light and dark cannot be reversed; and the value range of the pixel mapping function is between 0 and 255. The cumulative distribution function is a single growth function, and the value range is 0 to 1.

[0114] In the histogram equalization process, the mapping function is:

[0115]

[0116] Where n is the total number of pixels in the image, nk is the pixel format of the current gray level, and L is the total number of possible gray levels of the image.

[0117] Step S2.2: Ensure that the value range of the pixel mapping function is between 0 and 255;

[0118] Specifically, by calculating the histogram cumulative distribution function, comparing the probability distribution function and the cumulative distribution function, the two-dimensional graph of the former is uneven, and the latter is monotonically increasing.

[0119] Step S2.3: Nonlinear stretching of the face image and re-distribution of the face image pixel value;

[0120] The gray scale of the original image is uniformly distributed in the entire gray space from a certain gray interval in the comparison set, realizing the nonlinear stretching of the image and re-distribution of the image pixel value. It includes:

[0121] Calculate the pixel probability distribution of the original gray image; obtain the image cumulative distribution function according to the pixel probability distribution; obtain the transformed image according to the mapping function.

[0122] Step S2.4: Improve the visual effect of the face image by adjusting the brightness distribution of the face image.

[0123] Specifically, by performing a gamma function operation on each pixel of the original image, and then normalizing it to the range of 0-255. Correct the picture with too high or too low gray scale and enhance the contrast. The transformation formula is to do a multiplication operation on each pixel value of the original image:

[0124] s = Cr γ ,

[0125] Where C and γ are positive constants, r is the pixel value of the corresponding position of the original image, and s is the pixel value of the image after gamma transformation.

[0126] The value of gamma is divided by 1, and the smaller the value, the stronger the expansion effect on the low gray part of the image; the larger the value, the stronger the expansion effect on the high gray part of the image. Through different gamma values, the details of the low gray or high gray part can be enhanced. By adjusting the parameters of the gamma function, the brightness distribution of the image can be changed, thereby improving the contrast and visual effect of the image. Using gamma transformation can enhance the detail information and level of the image, especially when processing low-contrast images, the effect is more significant.

[0127] In the case of strong light direct illumination, after using the above image brightness enhancement algorithm, the face detection algorithm is used to detect whether there is a face in the IR image. When a face frame is detected, the enhanced IR image may lose the liveness detection feature, and directly participating in liveness detection may be detected as a non-living body. Therefore, when the IR camera is backlit, when the liveness detection is unsuccessful or the face authentication fails after detecting the face, the infrared fill light is driven to light up.

[0128] Through testing, under backlit conditions, after the infrared fill light is turned on, the IR image can return to normal. Therefore, in a strong light working environment, the image enhancement algorithm is started by detecting the IR brightness, and the infrared fill light is turned on when the face frame is recognized by the face frame recognition algorithm and the liveness detection is unsuccessful or the face authentication is recognized, thereby solving the problem of liveness detection and identity authentication of the forklift driver under strong light direct illumination.

[0129] In addition, when the light intensity is too weak, the infrared fill light is triggered by the light sensing device to obtain a normal state IR image, and the RGB and IR images are obtained for binocular liveness detection.

[0130] In the case of extremely dark light (night without light operation), when the RGB image cannot clearly see the face image, the IR image is directly used for single-mode liveness detection, and the single-mode liveness detection accuracy is lower than that of the double-mode. It is also trained by using a deep learning method, and the model input is 48*48, which will not be described here.

[0131] Through the special processing of the algorithms and devices under strong light and weak light, the application realizes anti-fraud detection in various complex lighting environments.

[0132] Step S3: identity detection is performed on each frame of the continuous frame number of the recognized face image, and after passing, a frame number is recorded;

[0133] Specifically, considering the accuracy and accuracy of the algorithm itself, anti-fraud detection and identity authentication detection need to be performed on each frame of the continuous frame number K, and when both detections pass, it is considered that the detection is successful once;

[0134] Step S4: The recorded frame number k is counted, and when the k value is greater than a preset threshold, the verification is passed.

[0135] The number of passed frames k in a continuous number of frames K is counted, and when the value of k exceeds half of K, it is considered that the driver identity authentication is successful.

[0136] The application designs a multi-frame fusion processing method. The specific algorithm is as follows:

[0137] 1) Initialization: set the detection frame number as K (10 in the application), the minimum continuous detection success frame number as dcontin (4 in the application), the detection success number variable in the K-frame detection process as dsucess, and define the current detection frame number F as 0.

[0138] 2) Algorithm steps;

[0139] After the first detection success, F = 1, dcontin = 1, and dsucess = 1; every time an image is detected, F is accumulated by 1;

[0140] If no detection success, dcontin = 0; if the last frame is detected successfully, the current frame is also detected successfully (i.e. continuous detection success), dcontin is accumulated by 1, and dsucess is accumulated by 1.

[0141] If dcontin >= 4, it is judged that the current device identity authentication is successful, all multi-frame fusion parameters are reset, and return;

[0142] else if dsucess > K / 2, it is judged that the current device identity authentication is successful, all multi-frame fusion parameters are reset, and return;

[0143] else when the accumulated detection image frame rate exceeds K times, it is determined that the detection fails.

[0144] The application method is based on a domestic chip platform, provides a self-controllable forklift driver face living body anti-fraud detection technology, greatly reduces the product cost, popularizes the forklift intelligent auxiliary safety driving product, and standardizes the forklift driver driving behavior.

[0145] Embodiment two

[0146] The application provides an identity verification device, which comprises:

[0147] An image processing module is configured to perform feature extraction and classification processing on the collected driver face image.

[0148] An image brightness enhancement module is configured to identify whether the image after classification processing has a face region, and if not, compare the illumination intensity with the first and second illumination threshold values; when the illumination intensity is greater than the first illumination threshold value, perform brightness enhancement on the face image, and when the illumination intensity is less than the second illumination threshold value, trigger the infrared fill light to fill light.

[0149] An identity detection module is configured to perform identity detection on each frame in a continuous frame number of the identified face image, and then record a frame number;

[0150] A statistical verification module is configured to count the recorded frame number k, and pass verification when the k value is greater than a preset threshold.

[0151] The image processing module, the image brightness enhancement module, the identity detection module and the statistical verification module are sequentially and orderly connected.

[0152] In a feasible implementation device, in the process of living body detection, when malicious attack behavior is found, i.e., multiple living body detection is false (3 times in the present application), full-screen video recording and tracking can also be adopted, 10s short video before and after the occurrence of multiple living body detection failures is recorded, and the behavior of multiple malicious attacks is captured and reported; meanwhile, voice reminding is performed.

[0153] The present application is based on a domestic embedded low-cost ARM chip hardware platform, and provides a face authentication device, which is mainly applied to a forklift operation scene and is applied to forklift safety management of enterprises of various scales, and performs language warning and video recording and reporting on malicious attack behavior.

[0154] Finally, it should be noted that: the above-described embodiments are merely specific implementations of the present application, which are used to illustrate the technical solutions of the present application, rather than limit the same, the protection scope of the present application is not limited thereto, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art in the technical range disclosed by the present application can still modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and all should be covered in the protection scope of the present application.

Claims

1. An identity authentication method, characterized in that: include: Perform feature extraction and classification processing on the collected driver's face images; Identify whether there is a face area in the image after classification processing. If not, compare the light intensity with the first illumination threshold and the second illumination threshold; when the light intensity is greater than the first illumination threshold, enhance the brightness of the face image; when the light intensity is less than the second illumination threshold, trigger the infrared fill light to fill in the light; Perform identity detection on each of the consecutive frames of the recognized face image, and record a frame number after passing the identity detection. The number of recorded frames k is counted. When the value of k is greater than the preset threshold, the verification is passed.

2. The identity authentication method according to claim 1, characterized in that: The collected driver's face image is an IR image and an RGB image using RGB and IR binocular cameras.

3. The identity authentication method according to claim 1, characterized in that: The classification process includes: performing binary classification of feature vector prediction value and feature vector label value on the face image after feature extraction; Increase the distance between classes and reduce the distance within classes.

4. The identity authentication method according to claim 3, characterized in that: To increase the distance between classes and reduce the distance within classes, the following calculation formula is used: Among them, L is the distance between the feature vector prediction value and the feature vector label value, λ is the weight parameter, N is the number of samples, C is the number of categories, and y c (n) is the label of the nth sample, is the actual value of the nth sample, y i is the sample classification label of category i, c yi is the midline point of the sample classification label of the i-th category, T is the transpose operation of the current matrix, and xi is the eigenvector of N samples.

5. The identity authentication method according to claim 2, characterized in that: The step of enhancing the brightness of the face image comprises: The size relationship of the face image is kept unchanged during pixel mapping, and the contrast is increased; Ensure that the value range of the pixel mapping function is between 0 and 255; Nonlinearly stretch the face image and redistribute the pixel values ​​of the face image; Improve the visual effect of facial images by adjusting the brightness distribution of facial images.

6. The identity authentication method according to claim 5, characterized in that: The expression of the pixel mapping function is as follows: Where n is the sum of pixels in the face image, S k is the pixel format of k gray levels, and L is the total number of gray levels of the face image.

7. The identity authentication method according to claim 6, characterized in that: The reallocation of facial image pixel values ​​includes: Calculate the pixel probability distribution of the original grayscale of the face image; Obtaining the cumulative distribution function of the face image based on the pixel probability distribution; Obtain the transformed face image according to the mapping function.

8. The identity authentication method according to claim 5, characterized in that: The adjusting the brightness distribution of the facial image includes: Perform a product operation on each pixel value of the original face image. The operation expression is as follows: s=Cr γ Where C and γ are positive constants, r is the pixel value of the corresponding position of the face image, and s is the pixel value of the face image after operation; Normalize the operation result to the range of [0, 255]; The face image after the correction operation is an image with a grayscale of a preset range threshold.

9. An identity authentication device, characterized in that: include: An image processing module is used to extract features and classify the collected driver's facial images; An image brightness enhancement module is used to identify whether there is a face area in the image after classification processing. If not, it compares the light intensity with the first illumination threshold and the second illumination threshold. When the light intensity is greater than the first illumination threshold, the brightness of the face image is enhanced. When the light intensity is less than the second illumination threshold, the infrared fill light is triggered to fill in the light. The identity detection module is used to perform identity detection on each frame of the continuous frames of the recognized face image, and record a frame number after passing the identity detection; The statistical verification module is used to count the number of recorded frames k. When the k value is greater than the preset threshold, the verification is passed.

Images