Method and system for identifying identity credibility of admission personnel of large-scale activity

Abstract

The invention discloses a credible identification method and system for identities of admission personnel of a large-scale activity, and particularly relates to the technical field of identity identification, and the method comprises the steps: obtaining a first image and a second image of the mth target personnel, and carrying out the histogram statistics of the second image, and obtaining a second gray histogram; analyzing the second grey level histogram by using an automatic threshold algorithm, determining the window width and the window level of the second grey level histogram, and adjusting each grey level value in the second grey level histogram to obtain an adjusted thermal imaging image; according to the method, the thermal imaging image is adjusted through histogram statistics and an automatic threshold algorithm, the quality of the thermal imaging image can be optimized, blood flow distribution information can be enhanced, and then the recognition stability and accuracy, especially the performance under different illumination and environment conditions, are improved. According to the method, the advantages of visible light and thermal imaging are combined, so that the error-tolerant rate of the system to the identified object is higher, and interference factors such as illumination, shielding and makeup can be effectively resisted.

Description

Technical Field

[0001] The present invention relates to the field of identity recognition technology, and more specifically, to a reliable identity recognition method and system for large-scale event attendants. Background Art

[0002] Authentication methods based on biometrics (such as face recognition, fingerprint recognition, iris recognition, etc.), smart ticketing and mobile payment have gradually become effective means to solve the identity verification and identification of people entering large-scale events. Biometric identification has the advantages of being unique and difficult to forge. It can complete identity verification in a short time and greatly improve the efficiency of admission.

[0003] However, existing technologies still have some problems, such as the decrease in biometric accuracy in complex environments (makeup, angle problems, lighting changes, etc.), imperfect privacy protection measures, and limited real-time processing capabilities and scalability of existing systems. Therefore, how to improve the system's anti-interference ability, ensure data privacy, and enhance the system's concurrent processing capabilities has become the key to current technology research.

[0004] In the existing methods, for example, the Chinese patent application with publication number CN108496173A discloses an electronic device and a facial recognition method thereof, when the electronic device needs to perform facial recognition, it is determined whether the current ambient light intensity collected by the ambient light sensor is greater than or equal to a preset threshold; when the current ambient light intensity is greater than or equal to the preset threshold, the fill light device is controlled to perform front fill light; when the current ambient light intensity is less than the preset threshold, the image sensor is controlled to collect the facial image of the object in front of the electronic device; the collected facial image is matched with the preset facial image template to determine whether the identity of the object in front is legal. For another example, the Chinese patent application with publication number CN110851844A discloses an identity recognition method, by establishing an electronic database of ID card information, converting the text part of the ID card information using encryption technology, collecting the portrait on the ID card information, collecting the facial image of the person to be identified, using a voice recognition device to collect the text part of the ID card information spoken by the person to be identified, comparing the target image with the source image to obtain the image matching rate, obtaining the commonality matching rate, and comprehensively judging the image matching rate and the commonality matching rate. Although the above method can realize the identity recognition of the entrants, research and application of the above method and the prior art have found that the above method and the prior art have at least the following defects:

[0005] Due to the use of makeup techniques and human skin masks, the accuracy of the above-mentioned identification methods is still affected, resulting in reduced recognition results, causing entry delays, identity verification failures and other problems, thus affecting the smooth progress of the event.

[0006] To this end, the present invention provides a reliable identification method and system for large-scale event attendees. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for reliable identification of the identities of people entering a large-scale event, so as to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a reliable identification method for a person entering a large-scale event, comprising:

[0010] Step 1: Obtain the first image and the second image of the mth target person, perform histogram statistics on the second image, and obtain a second grayscale histogram;

[0011] Step 2: Analyze the second grayscale histogram using an automatic thresholding algorithm, determine the window width and window position of the second grayscale histogram, and adjust each grayscale value in the second grayscale histogram to obtain an adjusted thermal imaging image;

[0012] Step 3: Fusion analysis is performed on the first image and the thermal imaging image to obtain a multimodal feature vector;

[0013] Step 4: Based on the comparison between the multimodal feature vector and the corresponding preset feature vector, the confidence score and the corresponding identity label of the mth target person are obtained;

[0014] Step 5: Authenticate the mth target person based on the identity tag and confidence score, obtain the authentication result, and set m=m+1, return to step 1, and end the loop when m=M, where M is the number of target persons involved in the verification.

[0015] Furthermore, the first image is a visible light image of the target person's face collected by a high-definition camera; the second image is a thermal imaging image of the target person's face collected by a thermal imaging camera;

[0016] The method of performing histogram statistics on the second image to obtain a second grayscale histogram includes:

[0017] Step a1. Convert the second image into a grayscale image, the conversion formula is:

[0018] ;

[0019] In the formula, is the gray value of the pixel in the second image, , and are the pixel values ​​of the red, green and blue channels respectively;

[0020] Step a2. Initialize the histogram array , the length of the histogram array is equal to the range of gray levels;

[0021] Step a3. Traverse each pixel value in the grayscale image , update the histogram according to its pixel value;

[0022] Among them, the method of updating the histogram according to the pixel value includes:

[0023] According to the pixel value Calculate the number of gray levels n;

[0024] Update the histogram array: , that is, the frequency of the corresponding gray value is increased by 1.

[0025] Furthermore, the method of analyzing the second grayscale histogram using an automatic thresholding algorithm includes:

[0026] Step b1: Initialize the threshold of the second grayscale histogram to G, and initialize the probabilities of the two classes of the second grayscale histogram; the two classes are foreground class and background class ;

[0027] Step b2: Calculate the average gray value of the foreground class , the average gray value of the background class and the global average gray value ;

[0028] Step b3: Calculate the inter-class variance based on the average gray value and the global average gray value;

[0029] Step b4: Update the thresholds of both classes from the minimum gray level to the maximum gray level and , the threshold of the maximum inter-class variance of the two classes is used as the first threshold and the second threshold ;

[0030] Step b5: Set the first threshold and the second threshold Positioned as the window width W[ , ], and the window width and window position are obtained; the window position L is expressed as .

[0031] Furthermore, the calculation method for initializing the probabilities of the two classes in the second grayscale histogram includes:

[0032] The frequency of each gray level in the second gray histogram is , the total number of gray levels is N, i represents the gray value, and the value range is [0, H-1];

[0033] Foreground The probability calculation formula is: ;

[0034] Background The probability calculation formula is: ;

[0035] Where: Foreground The probability of , which indicates the proportion of foreground pixels to the total pixels; For background class The probability of background pixels accounts for the proportion of total pixels. Z is the current threshold. is the total grayscale threshold.

[0036] Furthermore, the calculation formula of the between-class variance is:

[0037] ;

[0038] Where: represents the between-class variance, Represents the foreground class The probability of Represents the background class probability.

[0039] Furthermore, the method for adjusting each grayscale value in the second grayscale histogram includes:

[0040] Step c1. Determine the cropping range based on the window width W and window level L. , ];

[0041] Step c2. Loop through each gray value i in all pixels;

[0042] Step c3. Adjust each grayscale value according to the clipping formula to obtain a new grayscale value;

[0043] The clipping formula is as follows:

[0044] ;

[0045] In the formula, i is the original gray value, is the gray value after adjustment, Indicates the lower limit of the window width; Indicates the upper limit of the window width.

[0046] Furthermore, the first image is fused and analyzed with the thermal imaging image to obtain a multimodal feature vector including:

[0047] Step d1. Preprocess the first image and input it into a pre-built first machine learning model to predict the feature vector for outputting the target dimension. ; The preprocessing method is to extract an image of the face area from the first image by using a face detection algorithm, and the first machine learning model is FaceNet or ArcFace;

[0048] Step d2. Input the thermal imaging image into the pre-built second machine learning model to predict the feature vector for outputting the target dimension ; The feature vector The pattern of blood flow distribution or the details of temperature change; the second machine learning model is a convolutional neural network (CNN);

[0049] Step d3. Normalize each feature vector so that it is within [0,1];

[0050] Step d4. The normalized feature vectors are concatenated using a feature fusion formula to obtain a multimodal feature vector;

[0051] The feature fusion formula is: ;

[0052] In the formula, represents the multimodal feature vector, represents the weight parameter, Represents the normalized feature vector , Represents the normalized feature vector , .

[0053] Furthermore, the method of normalizing each feature vector includes:

[0054] Step s1. Get the gray value of the corresponding image and mark the maximum gray value of the corresponding image , mark the minimum gray value of the corresponding image ;

[0055] Step s2. Normalize each gray value of the corresponding image, and the calculation formula is:

[0056] ;

[0057] In the formula, Represents the normalized grayscale value of the corresponding image. Represents the grayscale value of the corresponding image;

[0058] Eigenvector The corresponding image is the first image after preprocessing, and the feature vector The corresponding image is a thermal imaging image.

[0059] Furthermore, based on the comparison between the multimodal feature vector and the corresponding preset feature vector, the method for obtaining the confidence score and the corresponding identity label of the mth target person is as follows:

[0060] Mark the multimodal feature vector of the mth target person as the first feature vector , mark the corresponding preset feature vector as the second feature vector ;

[0061] Calculate the first eigenvector With the second eigenvector The similarity between them is calculated as:

[0062] ;

[0063] In the formula, For confidence rating, is the first eigenvector No. feature vectors, is the total number of eigenvectors, is the second eigenvector No. feature vectors;

[0064] Each preset feature vector has one and only one corresponding identity label associated with it.

[0065] Furthermore, the identity authentication of the M target persons is performed based on the identity tags and the confidence scores, and the method for obtaining the identity authentication results includes:

[0066] If the confidence score is higher than the preset score threshold and the identity tag is the same as the preset identity tag, a pass instruction is generated;

[0067] If the confidence score is higher than the preset score threshold and the identity tag is different from the preset identity tag, a denial of admission instruction is generated;

[0068] If the confidence score is lower than the preset score threshold and the identity tag is the same as the preset identity tag, a re-verification instruction is generated;

[0069] If the confidence score is lower than a preset score threshold and the identity tag is different from the preset identity tag, a re-verification instruction is generated.

[0070] In a second aspect, the present invention provides a system for reliable identification of the identity of a large-scale event attendee, which is used to implement the above-mentioned method for reliable identification of the identity of a large-scale event attendee, comprising:

[0071] An image acquisition module, used to acquire a first image and a second image of the mth target person, perform histogram statistics on the second image, and obtain a second grayscale histogram;

[0072] A second image analysis module, used to analyze the second grayscale histogram using an automatic thresholding algorithm, determine a window width and window position of the second grayscale histogram, and adjust each grayscale value in the second grayscale histogram to obtain an adjusted thermal imaging image;

[0073] A fusion analysis module, used for fusing and analyzing the first image with the thermal imaging image to obtain a multimodal feature vector;

[0074] A comparison module is used to compare the multimodal feature vector with the corresponding preset feature vector to obtain the confidence score and corresponding identity label of the mth target person;

[0075] The verification module authenticates the mth target person based on the identity tag and confidence score, obtains the authentication result, sets m=m+1, returns to the image acquisition module, and ends the loop when m=M, where M is the number of target persons involved in the verification.

[0076] Technical effects and advantages of the present invention:

[0077] 1. The present invention significantly improves the accuracy and robustness of identity recognition by multimodal fusion of the first image (visible light image) and the second image (thermal imaging image). The first image provides basic facial features, such as contour, eye distance, nose bridge length, etc., while the second image reveals the blood flow distribution of the face through thermal imaging technology, which adds additional physiological features to identity verification. By adjusting the thermal imaging image through histogram statistics and automatic threshold algorithm, the quality of the thermal imaging image can be optimized, the blood flow distribution information can be enhanced, and the stability and accuracy of recognition can be improved, especially under different lighting and environmental conditions. This technology combines the advantages of visible light and thermal imaging, making the system more tolerant to the object being identified and effectively resisting interference factors such as lighting, occlusion and makeup.

[0078] 2. The present invention achieves accurate personnel identification by establishing an identity authentication process based on multimodal feature vectors. Multimodal feature vectors not only integrate facial features and physiological status, but also quantify the reliability of each identification through a confidence scoring mechanism. By comparing with the preset feature vectors, the system can quickly determine the identity of the target person and further verify the uncertain identity according to the preset threshold, thereby avoiding misidentification and ensuring that the identity of each participant in large-scale activities is strictly verified, thereby improving the overall security and efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1This is a flow chart of a reliable identification method for large-scale event attendees in Example 1;

[0080] Figure 2 is a flow chart of a method for obtaining a second grayscale histogram in Example 1;

[0081] Figure 3 This is a flow chart of a method for adjusting each grayscale value in the second grayscale histogram according to Embodiment 1;

[0082] Figure 4 This is a schematic diagram of a trusted identity recognition system for large-scale event entrants according to Example 2. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0085] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and a similar second unit may be referred to as a first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0086] Example 1

[0087] See also Figure 1 As shown, this embodiment discloses a method for reliable identification of the identity of a person entering a large-scale event, the method comprising:

[0088] Step 1: Obtain the first image and the second image of the mth target person, perform histogram statistics on the second image, and obtain a second grayscale histogram; m=1, 2, ..., M; M is the number of target persons participating in the verification.

[0089] It should be understood that: the first image is a visible light image of the target person's face collected by a high-definition camera; the visible light image reflects the basic facial features of the target person, including facial contour, eye distance, nose bridge length, mouth width, etc. The second image is a thermal imaging image of the target person's face collected by a thermal imaging camera, reflecting the thermal distribution characteristics of the target person's face, including facial blood flow distribution information generated based on blood flow.

[0090] It should be noted that the first image and the second image are collected simultaneously.

[0091] It should be understood that performing histogram statistics on the second image refers to performing statistical analysis on the pixel values ​​in the image to generate a grayscale histogram or a color histogram. The histogram is used to represent the frequency distribution of different grayscale levels (or color values) in the image, thereby revealing the brightness distribution, contrast and other visual characteristics of the image.

[0092] In actual operation, the first image and the second image need to be preprocessed to improve the quality of the data and the accuracy of subsequent analysis. The preprocessing operation includes the following steps:

[0093] Image brightness correction: adjust the overall brightness of the image to reduce exposure differences;

[0094] Image bad pixel and bad line correction: repair abnormal pixels or lines caused by sensor defects;

[0095] Image background subtraction: eliminate the influence of background noise or constant offset;

[0096] Image denoising: Reduce random noise and improve image quality. Specific methods include:

[0097] Median filtering: Reduce impulse noise by calculating the local median value;

[0098] Gaussian filtering: Use Gaussian function to smooth the image and reduce high-frequency noise;

[0099] Mean filtering: Reduce small noise in the image by calculating the local mean;

[0100] Wavelet denoising: Reconstruct the signal after decomposition in the wavelet domain and suppress the noise component.

[0101] The above-mentioned preprocessing method is prior art, and the present invention will not elaborate on it in detail.

[0102] See also Figure 2 As shown, in implementation, the method of performing histogram statistics on the second image to obtain the second grayscale histogram includes:

[0103] Step a1. Convert the second image into a grayscale image, the conversion formula is:

[0104] ;

[0105] In the formula, is the gray value of the pixel in the second image, , and They are the pixel values ​​of the red, green and blue channels respectively.

[0106] Step a2. Initialize the histogram array , the length of the histogram array is equal to the range of gray levels;

[0107] It should be noted that the length of the array is usually 256, which is used to store the grayscale value frequency from 0 to 255.

[0108] Step a3. Traverse each pixel value in the grayscale image , update the histogram according to the pixel values.

[0109] Specifically, the method for updating the histogram according to the pixel value includes:

[0110] According to the pixel value Calculate its gray level number n (gray value);

[0111] Update the histogram array: , that is, the frequency of the corresponding gray value is increased by 1.

[0112] It should be noted that, in the grayscale histogram, the horizontal axis represents the grayscale value, and the vertical axis represents the number of pixels where the grayscale value appears.

[0113] For ease of understanding, the above steps are described in detail:

[0114] For example, it is assumed that the second image is an 8-bit grayscale image with a size of 4×4 and the pixel values ​​are as follows:

[0115]

[0116] Histogram statistical process:

[0117] a1. Initialize array F[0…255]=0.

[0118] a2. Traverse each pixel value and update the array:

[0119] 12: Appears 3 times, update F

[12] =3.

[0120] 80: Appears 5 times, update F

[80] =5.

[0121] 120: Appears 3 times, update F

[120] =3.

[0122] 255: Appears 5 times, update F

[255] =5.

[0123] The other grayscale values ​​remain at 0.

[0124] a3. Result histogram:

[0125] F=[0,0,…,3,…,5,…,3,…,5].

[0126] The non-zero values ​​are:

[0127] F

[12] =3, F

[80] =5, F

[120] =3, F

[255] =5.

[0128] This step uses a high-definition camera to obtain a visible light image (first image) of the target person, and uses a thermal imaging camera to obtain a thermal imaging image (second image), and then performs histogram statistics on the second image (thermal imaging image) to obtain a second grayscale histogram. This step is very important because the histogram provides the distribution of grayscale values ​​in the image. Subsequent image processing and feature extraction need to be based on this information to provide support for subsequent image processing and analysis.

[0129] Step 2: Analyze the second grayscale histogram using an automatic thresholding algorithm, determine the window width and window position of the second grayscale histogram, adjust each grayscale value in the second grayscale histogram, and obtain an adjusted thermal imaging image.

[0130] It should be understood that the automatic threshold algorithm is a common method for image segmentation in image processing. It can determine a suitable threshold based on the histogram curve of the image to divide the image into different regions. This process is usually used to binarize images to separate the target object from the background in the image.

[0131] It should be noted that the automatic thresholding algorithm is specifically the Otsu's thresholding algorithm. The window width is the range when analyzing the histogram, for example, only the grayscale value between 100 and 200 is analyzed, while the grayscale values ​​below 100 or above 200 are ignored. The window position determines the center of the window width. For example, if the window position is set to 150 and the window width is 100, the grayscale range of interest is 100 to 200. Adjusting the window position can focus on analyzing the grayscale value distribution of different brightness areas in the image, helping to separate the foreground and background in a targeted manner.

[0132] In implementation, the method of analyzing the second grayscale histogram using an automatic thresholding algorithm includes:

[0133] Step b1: Initialize the threshold of the second grayscale histogram to G, and initialize the probabilities of the two classes of the second grayscale histogram; the two classes are foreground class and background class ;

[0134] The method for calculating the probabilities of the two classes in the initialized second grayscale histogram includes:

[0135] The frequency of each gray level in the second gray histogram is , the total number of gray levels is N, i represents the gray value;

[0136] Foreground The probability calculation formula is: ;

[0137] Background The probability calculation formula is: ;

[0138] Where: Foreground The probability of , which indicates the proportion of foreground pixels to the total pixels; For background class The probability of background pixels is expressed as the proportion of background pixels to the total pixels.

[0139] Step b2: Calculate the average gray value of the foreground class , the average gray value of the background class and the global average gray value ;

[0140] The calculation formula for the average gray value of the foreground class is: ;

[0141] The calculation formula for the average gray value of the background class is: ;

[0142] The calculation formula of the global average gray value is: ;

[0143] In the formula, represents the average gray value of the foreground class, Represents the average gray value of the background class, Represents the global average grayscale value; i represents the grayscale value, ranging from [0, H-1], Z is the current threshold, and the grayscale image is divided into foreground and background classes. is the total grayscale threshold;

[0144] Step b3: Calculate the inter-class variance based on the average gray value and the global average gray value;

[0145] The calculation formula of between-class variance is:

[0146] ;

[0147] Where: represents the between-class variance, Represents the foreground class The probability of Represents the background class The probability of Represents the global average gray value.

[0148] Step b4: Update the thresholds of both classes from the minimum gray level to the maximum gray level and , the threshold of the maximum inter-class variance of the two classes is used as the first threshold and the second threshold ;

[0149] Suppose a grayscale image is segmented:

[0150] Grayscale range: 0-255;

[0151] Update threshold: From 0 to 255, iterates over possible and ;

[0152] For each pair and Candidate values:

[0153] Divide the pixels into and ;

[0154] Calculate the between-class variance;

[0155] Record the threshold that maximizes the inter-class variance and .

[0156] Finally, the selected and It is the optimal threshold, that is, the grayscale image segmentation effect is best.

[0157] It should be noted that the number of gray levels is set by technicians in this field based on experience. Exemplarily, the second grayscale histogram is divided into 16 grayscale levels (such as 0-15, 16-31, ...) to optimize the inter-class variance and quickly separate the blood flow area and the background area in the second grayscale histogram.

[0158] Step b5: Set the first threshold and the second threshold Positioned as the window width W[ , ], and the window width and window position are obtained; the window position L is expressed as .

[0159] It should be noted that the methods for adjusting each pixel value in the thermal imaging image include linear stretching adjustment, window width and window position clipping, and contrast limited adaptive histogram equalization (CLAHE). According to the characteristics of the thermal imaging image, the present embodiment preferably uses window width and window position clipping.

[0160] See also Figure 3 As shown, specifically, the method for adjusting each grayscale value in the second grayscale histogram includes:

[0161] Step c1. Determine the cropping range based on the window width W and window level L. , ];

[0162] Step c2. Loop through each gray value i in all pixels;

[0163] Step c3. Adjust each grayscale value according to the clipping formula to obtain a new grayscale value;

[0164] The clipping formula is as follows:

[0165] ;

[0166] In the formula, i is the original gray value, is the gray value after adjustment, Indicates the lower limit of the window width; Indicates the upper limit of the window width.

[0167] This step uses an automatic threshold algorithm (such as Otsu's algorithm) to automatically select an appropriate threshold from the grayscale histogram. By adjusting the window width and window position, the range and brightness of the grayscale image are controlled, making the features of the thermal imaging image more prominent, thereby better representing the facial blood flow distribution characteristics of the target person. The adjusted image can more clearly show the changes in blood flow, which is helpful for subsequent identity recognition.

[0168] Step 3: Fusion analysis is performed on the first image and the thermal imaging image to obtain a multimodal feature vector;

[0169] It should be noted that multimodal feature vectors refer to feature vectors extracted by fusing images of different modalities (in this case, visible light images and thermal imaging images). These vectors contain the common and complementary information of the two images, and can more effectively perform tasks such as identity authentication and emotion recognition.

[0170] In implementation, the first image is fused and analyzed with the thermal imaging image to obtain a multimodal feature vector, including:

[0171] Step d1. Preprocess the first image and input it into a pre-built first machine learning model to predict the feature vector for outputting the target dimension. ;

[0172] It should be noted that: the preprocessing method is to extract an image of the face area from the first image by using a face detection algorithm.

[0173] The method for constructing the first machine learning model includes:

[0174] Acquire historical image training data, and divide the historical image training data into an image training set and an image test set; the historical image training data includes a preprocessed first image and a feature vector of a corresponding target dimension ; The target dimension is set by technicians in this field based on experience and can be 128 dimensions or 512 dimensions.

[0175] Constructing a first deep learning network, taking the preprocessed first image in the image training set as input data of the first deep learning network, taking the feature vector of the target dimension in the image training set as output data of the first deep learning network, training the first deep learning network, and obtaining an initial image deep learning network;

[0176] The initial image deep learning network is model verified using an image test set, and an initial image deep learning network with an output prediction error less than or equal to a preset error threshold is used as the first machine learning model; the first deep learning network is specifically FaceNet or ArcFace.

[0177] Step d2. Input the thermal imaging image into the pre-built second machine learning model to predict the feature vector for outputting the target dimension ; The feature vector For example, the pattern of blood flow distribution or details of temperature changes.

[0178] It should be noted that the training method of the second machine learning model is the same as that of the first machine learning model, and will not be repeated here. It is worth noting that the second machine learning model is a convolutional neural network (CNN).

[0179] Step d3. Normalize each feature vector so that it is within [0,1];

[0180] The method of normalizing each feature vector includes:

[0181] Step s1. Get the gray value of the corresponding image and mark the maximum gray value of the corresponding image , mark the minimum gray value of the corresponding image ;

[0182] Step s2. Normalize each gray value of the corresponding image, and the calculation formula is:

[0183] ;

[0184] In the formula, Represents the normalized grayscale value of the corresponding image. Represents the grayscale value of the corresponding image.

[0185] It should be noted that the corresponding image (feature vector The corresponding image is the first image after preprocessing, and the feature vector The corresponding image is a thermal imaging image) and its grayscale values ​​are all within [0,1].

[0186] Step d4. Concatenate the two normalized feature vectors using a feature fusion formula to obtain a multimodal feature vector;

[0187] The feature fusion formula is: ;

[0188] In the formula, represents the multimodal feature vector, represents the weight parameter, Represents the normalized feature vector , Represents the normalized feature vector , .

[0189] It should be noted that the weight parameters are set by those skilled in the art based on experience.

[0190] This step fuses the visible light image (first image) with the thermal image (processed and adjusted image) to generate a multimodal feature vector, which contains information from two modalities: the visible light image provides basic facial features (such as contour, eye distance, etc.), and the thermal image provides the thermal distribution (blood flow distribution) of the face. Combining the information from these two modalities helps improve the accuracy of identity recognition.

[0191] Step 4: Based on the comparison between the multimodal feature vector and the corresponding preset feature vector, the confidence score and corresponding identity label of the mth target person are obtained.

[0192] It should be noted that each preset feature vector has one and only one corresponding identity tag bound to it. The preset feature vector is obtained by an extraction method consistent with the multimodal feature vector and is stored in the system database in advance. The identity tag (usually refers to a unique identifier associated with a specific person, such as the person's name, ID number, etc.) is bound to each feature vector in the database. The identity tag of each target person is unique, so each preset feature vector can only correspond to one identity tag. This means that during the identity authentication process, the system will match the multimodal feature vector with the preset feature vector in the database, and determine the identity of the target person by calculating the confidence score. The matching result of each feature vector will only be associated with one identity tag.

[0193] It is worth noting that: the identity label refers to whether there is a record of behavior that endangers public safety. The behavior record that endangers public safety includes a criminal record or a credit record, etc., which is specifically marked by technical personnel in this field based on the behavior record. The marking of the behavior record is not the focus of this application and will not be described in detail here.

[0194] In implementation, based on the comparison between the multimodal feature vector and the corresponding preset feature vector, the method for obtaining the confidence score and the corresponding identity label of the mth target person is as follows:

[0195] Mark the multimodal feature vector of the mth target person as the first feature vector , mark the corresponding preset feature vector as the second feature vector ;

[0196] Calculate the first eigenvector With the second eigenvector The similarity between them is calculated as:

[0197] ;

[0198] In the formula, For confidence rating, is the first eigenvector No. eigenvectors, K is the total number of eigenvectors, is the second eigenvector No. feature vectors.

[0199] It should be noted that the above formula is the Euclidean distance formula, which is a distance value (the smaller the distance, the more similar the two are). In order to make the confidence score as large as possible, a negative sign is used.

[0200] This step compares the generated multimodal feature vector with the feature vector of the target person stored in the database. Based on the comparison result, the confidence score, that is, the credibility of the current recognition result, is calculated. If the confidence score is higher than the preset threshold, the match is considered successful and the corresponding identity tag is output.

[0201] Step 5: Based on the identity tag and confidence score, authenticate the mth target person, obtain the authentication result, and set m=m+1, return to step 1, and end the loop when m=M, where M is the number of target persons involved in the authentication;

[0202] It should be noted that the identity tags include VIP, ordinary audience, staff or special identity tags, etc. The special identity tags refer to persons marked by the system as prohibited from entering (such as those with bad records or forged identities).

[0203] In implementation, the mth target person is authenticated based on the identity tag and the confidence score, and the method for obtaining the authentication result includes:

[0204] If the confidence score is higher than the preset score threshold and the identity tag is the same as the preset identity tag, a pass instruction is generated, indicating that the identity authentication of the mth target person is successful.

[0205] Note: When the confidence score is higher than the preset score threshold (for example, 0.9), and the identified identity type (label) matches the preset identity type (label) (that is, the system determines that the person is the correct identity), it can be determined that the person's identity authentication is successful and is allowed to pass.

[0206] If the confidence score is higher than the preset score threshold and the identity tag is different from the preset identity tag, a denial of entry instruction is generated, indicating that the identity authentication of the mth target person is unsuccessful.

[0207] Note: When the confidence score is higher than the preset score threshold, but the identified identity type does not match the preset identity type, it means that although the system has a high degree of confidence in identifying the identity, it has identified the wrong identity. Therefore, identity verification fails, entry is denied, or further verification is required.

[0208] If the confidence score is lower than the preset score threshold and the identity tag is the same as the preset identity tag, a re-verification instruction is generated, indicating that the recognition result is unreliable. Even if the identity tags are consistent, the system considers the recognition result to be unreliable and needs to be re-verified.

[0209] If the confidence score is lower than the preset score threshold and the identity tag is different from the preset identity tag, a re-verification instruction is generated, indicating that the recognition is unclear due to poor image quality, environmental interference or other factors. The recognized identity type is usually marked as "uncertain" and needs to be re-verified. The re-verification method includes manual confirmation or auxiliary verification (for example, through other biometric verification, such as fingerprint or mobile phone verification) to enhance the recognition result.

[0210] This embodiment significantly improves the accuracy and robustness of identity recognition by multimodally fusing the first image (visible light image) and the second image (thermal imaging image). The first image provides basic facial features, such as contour, eye distance, nose bridge length, etc., while the second image reveals the blood flow distribution of the face through thermal imaging technology, which adds additional physiological features to identity verification. By adjusting the thermal imaging image through histogram statistics and automatic threshold algorithm, the quality of the thermal imaging image can be optimized, the blood flow distribution information can be enhanced, and the stability and accuracy of recognition can be improved, especially under different lighting and environmental conditions. This technology combines the advantages of visible light and thermal imaging, making the system more tolerant to the object being identified and effectively resisting interference factors such as lighting, occlusion and makeup.

[0211] This embodiment achieves accurate personnel identification by establishing an identity authentication process based on multimodal feature vectors. Multimodal feature vectors not only integrate facial features and physiological status, but also quantify the reliability of each identification through a confidence scoring mechanism. By comparing with the preset feature vectors, the system can quickly determine the identity of the target person and further verify the uncertain identity according to the preset threshold, thereby avoiding misidentification and ensuring that the identity of each participant in large-scale activities is strictly verified, thereby improving the overall security and efficiency of the system.

[0212] Example 2

[0213] See also Figure 4 As shown, this embodiment provides a trusted identification system for large-scale event attendants, including an image acquisition module, a second image analysis module, a fusion analysis module, a comparison module, and a verification module; each module is connected by wire and / or wireless means to achieve data transmission between modules; the system includes:

[0214] An image acquisition module, used to acquire a first image and a second image of the mth target person, perform histogram statistics on the second image, and obtain a second grayscale histogram;

[0215] A second image analysis module, used to analyze the second grayscale histogram using an automatic thresholding algorithm, determine a window width and window position of the second grayscale histogram, and adjust each grayscale value in the second grayscale histogram to obtain an adjusted thermal imaging image;

[0216] A fusion analysis module, used for fusing and analyzing the first image with the thermal imaging image to obtain a multimodal feature vector;

[0217] A comparison module is used to compare the multimodal feature vector with the corresponding preset feature vector to obtain the confidence score and corresponding identity label of the mth target person;

[0218] The verification module authenticates the mth target person based on the identity tag and confidence score, obtains the authentication result, sets m=m+1, returns to the image acquisition module, and ends the loop when m=M, where M is the number of target persons involved in the verification.

[0219] Example 3

[0220] This embodiment provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable codes, and when the computer-readable codes are executed by the one or more processors, the kiln control method for ceramic firing as described above may be executed.

[0221] The system according to the embodiment of the present application can also be implemented with the help of the architecture of the electronic device described below. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output, a hard disk, etc. A storage device in the electronic device, such as a ROM or a hard disk, can store the kiln control method for ceramic firing provided in the present application. Furthermore, the electronic device may also include a user interface. Of course, the above architecture is only exemplary. When implementing different devices, one or more components in the above electronic device may be omitted according to actual needs.

[0222] Example 4

[0223] This embodiment provides a computer-readable storage medium storing instructions. When the instructions are executed on a computer, the computer executes a method for reliable identification of the identities of people entering a large-scale event according to Embodiment 1.

[0224] The formulas involved in the above are all calculated by removing dimensions and taking their numerical values. They are a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The weight factors in the formula and the preset thresholds in the analysis process are set by technicians in this field according to actual conditions or obtained by simulating a large amount of data; the size of the weight factor is to quantify each parameter to obtain a specific value for subsequent comparison. The size of the weight factor depends on the amount of sample data and the corresponding processing coefficient initially set by technicians in this field for each group of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0225] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0226] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A reliable identification method for large-scale event attendees, characterized in that: include: Step 1: Obtain the first image and the second image of the mth target person, perform histogram statistics on the second image, and obtain a second grayscale histogram; Step 2: Analyze the second grayscale histogram using an automatic thresholding algorithm, determine the window width and window position of the second grayscale histogram, and adjust each grayscale value in the second grayscale histogram to obtain an adjusted thermal imaging image; Step 3: Fusion analysis is performed on the first image and the thermal imaging image to obtain a multimodal feature vector; Step 4: Based on the comparison between the multimodal feature vector and the corresponding preset feature vector, the confidence score and the corresponding identity label of the mth target person are obtained; Step 5: Authenticate the mth target person based on the identity tag and confidence score, obtain the authentication result, and set m=m+1, return to step 1, and end the loop when m=M, where M is the number of target persons involved in the verification.

2. A reliable identification method for large-scale event attendees according to claim 1, characterized in that: The first image is a visible light image of the target person's face captured by a high-definition camera; The second image is a thermal imaging image of the target person's face collected by a thermal imaging camera; The method of performing histogram statistics on the second image to obtain a second grayscale histogram includes: Step a1. Convert the second image into a grayscale image, the conversion formula is: ； In the formula, is the gray value of the pixel in the second image, , and are the pixel values ​​of the red, green and blue channels respectively; Step a2. Initialize the histogram array , the length of the histogram array is equal to the range of gray levels; Step a3. Traverse each pixel value in the grayscale image , update the histogram according to the pixel value; Among them, the method of updating the histogram according to the pixel value includes: According to the pixel value Calculate the number of gray levels n; Update the histogram array: , that is, the frequency of the corresponding gray value is increased by 1.

3. A reliable identification method for large-scale event attendees according to claim 2, characterized in that: The method of analyzing the second grayscale histogram using an automatic thresholding algorithm includes: Step b1: Initialize the threshold of the second grayscale histogram to G, and initialize the probabilities of the two classes of the second grayscale histogram; the two classes are foreground class and and background class ; Step b2: Calculate the average gray value of the foreground class , the average gray value of the background class and the global average gray value ; Step b3: Calculate the inter-class variance based on the average gray value and the global average gray value; Step b4: Update the thresholds of both classes from the minimum gray level to the maximum gray level and , the threshold of the maximum inter-class variance of the two classes is used as the first threshold and the second threshold ; Step b5: Set the first threshold and the second threshold Positioned as the window width W[ , ], and the window width and window position are obtained; the window position L is expressed as .

4. A reliable identification method for large-scale event attendees according to claim 3, characterized in that: The calculation method of initializing the probabilities of the two classes in the second grayscale histogram includes: The frequency of each gray level in the second gray histogram is , the total number of gray levels is N, i represents the gray value, and the value range is [0, H-1]; Foreground The probability calculation formula is: ; Background The probability calculation formula is: ; Where: Foreground The probability of , which indicates the proportion of foreground pixels to the total pixels; For background class The probability of background pixels accounts for the proportion of total pixels. Z is the current threshold. is the total grayscale threshold.

5. A reliable identification method for large-scale event attendees according to claim 3, characterized in that: The calculation formula for between-class variance is: ； Where: represents the between-class variance, Represents the foreground class The probability of Represents the background class probability.

6. A reliable identification method for large-scale event attendees according to claim 5, characterized in that: The method for adjusting each grayscale value in the second grayscale histogram includes: Step c1. Determine the cropping range based on the window width W and window level L. , ]; Step c2. Loop through each gray value i in all pixels; Step c3. Adjust each grayscale value according to the clipping formula to obtain a new grayscale value; The clipping formula is as follows: ； In the formula, i is the original gray value, is the gray value after adjustment, Indicates the lower limit of the window width; Indicates the upper limit of the window width.

7. A reliable identification method for large-scale event attendees according to claim 6, characterized in that: The first image is fused and analyzed with the thermal imaging image to obtain a multimodal feature vector including: Step d1. Preprocess the first image and input it into a pre-built first machine learning model to predict the feature vector for outputting the target dimension. ; The preprocessing method is to extract an image of a face area from the first image by using a face detection algorithm, and the first machine learning model is FaceNet or ArcFace; Step d2. Input the thermal imaging image into the pre-built second machine learning model to predict the feature vector for outputting the target dimension ; The feature vector The pattern of blood flow distribution or the details of temperature change; the second machine learning model is a convolutional neural network; Step d3. Normalize each feature vector so that it is within [0,1]; Step d4. The normalized feature vectors are concatenated using a feature fusion formula to obtain a multimodal feature vector; The feature fusion formula is: ; In the formula, represents the multimodal feature vector, represents the weight parameter, Represents the normalized feature vector , Represents the normalized feature vector , .

8. A reliable identification method for large-scale event attendees according to claim 7, characterized in that: Methods for normalizing each feature vector include: Step s1. Get the gray value of the corresponding image and mark the maximum gray value of the corresponding image , mark the minimum gray value of the corresponding image ; Step s2. Normalize each gray value of the corresponding image, and the calculation formula is: ； In the formula, Represents the normalized grayscale value of the corresponding image. Represents the grayscale value of the corresponding image; Eigenvector The corresponding image is the first image after preprocessing, and the feature vector The corresponding image is a thermal imaging image.

9. A reliable identification method for large-scale event attendees according to claim 8, characterized in that: Based on the comparison between the multimodal feature vector and the corresponding preset feature vector, the method for obtaining the confidence score and the corresponding identity label of the mth target person is as follows: Mark the multimodal feature vector of the mth target person as the first feature vector , mark the corresponding preset feature vector as the second feature vector ; Calculate the first eigenvector With the second eigenvector The similarity between them is calculated as: ； In the formula, Score the confidence level, is the first eigenvector No. eigenvectors, K is the total number of eigenvectors, is the second eigenvector No. feature vectors; Each preset feature vector has one and only one corresponding identity label associated with it.

10. A reliable identification method for large-scale event attendees according to claim 9, characterized in that: The method of authenticating the M target persons based on the identity tags and confidence scores and obtaining the authentication results includes: If the confidence score is higher than the preset score threshold and the identity tag is the same as the preset identity tag, a pass instruction is generated; If the confidence score is higher than the preset score threshold and the identity tag is different from the preset identity tag, a denial of admission instruction is generated; If the confidence score is lower than the preset score threshold and the identity tag is the same as the preset identity tag, a re-verification instruction is generated; If the confidence score is lower than a preset score threshold and the identity tag is different from the preset identity tag, a re-verification instruction is generated.

11. A reliable identification system for large-scale event attendees, used to implement a reliable identification method for large-scale event attendees as claimed in any one of claims 1 to 10, characterized in that: include: An image acquisition module, used to acquire a first image and a second image of the mth target person, perform histogram statistics on the second image, and obtain a second grayscale histogram; A second image analysis module, used to analyze the second grayscale histogram using an automatic thresholding algorithm, determine a window width and window position of the second grayscale histogram, and adjust each grayscale value in the second grayscale histogram to obtain an adjusted thermal imaging image; A fusion analysis module, used for fusing and analyzing the first image with the thermal imaging image to obtain a multimodal feature vector; A comparison module is used to compare the multimodal feature vector with the corresponding preset feature vector to obtain the confidence score and corresponding identity label of the mth target person; The verification module authenticates the mth target person based on the identity tag and confidence score, obtains the authentication result, sets m=m+1, returns to the image acquisition module, and ends the loop when m=M, where M is the number of target persons involved in the verification.

Images