Contactless Temperature Measurement Method and System Based on Facial Recognition

Abstract

The present invention discloses a contactless temperature measurement method and system based on face recognition. The method includes interactive structure construction, data collection, data preprocessing, live pre-recognition, personnel identity recognition, contactless distance control temperature measurement, and self-service personnel management. The present invention relates to the field of face recognition, specifically a contactless temperature measurement method and system based on face recognition. The present invention optimizes the data quality of face images by using image channel conversion enhancement processing and target segmentation processing; uses a method based on a convolutional neural network of a retina model for live pre-recognition, improving the overall accuracy of contactless temperature measurement; uses a method based on a convolutional neural network combined with a local binary pattern algorithm for personnel identity recognition. Combining the live face and identity information judged by live pre-recognition, it can provide safer and more effective contactless temperature measurement and personnel management in the detention center scenario.

Description

Technical Field

[0001] The present invention relates to the field of face recognition, and specifically refers to a contactless temperature measurement method and system based on face recognition. Background Art

[0002] The contactless temperature measurement method based on face recognition is a method that uses face recognition technology to measure the surface temperature of the human body in real time without directly contacting the skin surface of the person being measured. This method collects the facial image of the person being measured, combines devices such as a thermal imaging sensor or an infrared camera, identifies the facial area and measures the temperature value of this area, thereby judging the body temperature status of the person being measured. This technology has certain application prospects in the context of a detention center.

[0003] However, in the existing contactless temperature measurement methods based on face recognition, there are technical problems such as high energy consumption and time consumption of face recognition due to image quality problems, traditional RGB images causing distortion of face images, and unsegmented face images reducing the accuracy of face recognition; in the existing contactless temperature measurement methods based on face recognition, there is a technical problem of lacking a method to prevent the measured temperature from coming from a photo or other organisms in the context of a detention center; in the existing contactless temperature measurement methods based on face recognition, there is a technical problem of lacking a method to discriminate the identity of a person before managing the body temperature of detainees. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the prior art, the present invention provides a contactless temperature measurement method and system based on face recognition. In the existing contactless temperature measurement method based on face recognition, there are technical problems that the energy consumption and time consumption of face recognition are relatively high due to image quality problems, traditional RGB images will cause distortion of face images, and unsegmented face images will reduce the accuracy of face recognition. This solution creatively adopts image channel transformation enhancement processing and target segmentation processing to optimize the data quality of face images, providing a good data basis for subsequent live pre-recognition and personnel identity recognition. In the existing contactless temperature measurement method based on face recognition, there is a technical problem that there is a lack of a method to prevent the measured temperature from coming from a photo or other organisms in the detention center scenario. This solution creatively adopts a method based on a convolutional neural network of the retina model for live pre-recognition, intelligently detecting and judging whether the face image is a live face, improving the overall accuracy of contactless temperature measurement. In the existing contactless temperature measurement method based on face recognition, there is a technical problem that there is a lack of a method to discriminate the identity of personnel before managing the body temperature of detainees. This solution creatively adopts a method based on a convolutional neural network combined with the local binary pattern algorithm for personnel identity recognition. Combining the live face judged by live pre-recognition and the identity information can provide a more secure and effective contactless temperature measurement and personnel management in the detention center scenario.

[0005] The technical solution adopted by the present invention is as follows: The contactless temperature measurement method based on face recognition provided by the present invention includes the following steps:

[0006] Step S1: Construct an interaction structure;

[0007] Step S2: Data acquisition;

[0008] Step S3: Data preprocessing;

[0009] Step S4: Live pre-recognition;

[0010] Step S5: Personnel identity recognition;

[0011] Step S6: Contactless distance-controlled temperature measurement;

[0012] Step S7: Self-service personnel management.

[0013] Further, in step S1, the construction of the interaction structure is used to construct the interaction structure between the terminal and the server for contactless temperature measurement;

[0014] The terminal specifically refers to an intelligent interaction terminal, and the intelligent interaction terminal is used to register the face update notification of the server in the monitoring room when starting and establish a connection with the server;

[0015] The server specifically refers to an information interaction server, which is used for inputting face images and verifying access information. The face images specifically refer to the face images of detainees in the detention cell.

[0016] After establishing a connection with the information interaction server, the intelligent interaction terminal can perform self-service personnel management, which includes live pre-identification, personnel identity identification, and contactless time and distance temperature measurement.

[0017] Further, in step S2, the data collection is used to collect the training data required for live pre-identification and personnel identity identification. Specifically, it refers to obtaining the original face data from the portrait database through collection. The original face data includes a live face recognition data set and a personnel identity identification data set.

[0018] Further, in step S3, the data preprocessing is used to enhance the data quality of the face image data. Specifically, it refers to performing image enhancement, normalization, and smoothing operations on the original face data to obtain optimized face image data.

[0019] The optimized face image data includes enhanced live pre-identification data and face image segmentation data.

[0020] Specifically, the enhanced live pre-identification data is obtained by performing image channel conversion and enhancement processing on the live face recognition data set in the original face data.

[0021] The face image segmentation data is obtained by performing target segmentation processing on the personnel identity identification data set.

[0022] The image channel conversion and enhancement processing is used to convert the RGB image into an HSV space image. The RGB image specifically refers to a weighted image of three colors: red, green, and blue. The HSV space image specifically refers to a space image of hue, saturation, and brightness.

[0023] The conversion of the RGB image into an HSV space image is specifically achieved by using a light compensation method to generate the HSV space image. The light compensation method specifically refers to calculating the hue component H, saturation component S, and brightness component V of the generated HSV space image through light compensation calculations, including the following steps:

[0024] Step S31: Calculate the saturation component S, and the calculation formula is:

[0025] ;

[0026] In the formula, S is the saturation component, MAX() is the maximum value function, MIN() is the minimum value function, R is the red channel component, G is the green channel component, and B is the blue channel component.

[0027] Step S32: Calculate the luminance component V, and the calculation formula is:

[0028] V = MAX(R, G, B);

[0029] In the formula, V is the luminance component, MAX() is the maximum value function, R is the red channel component, G is the green channel component, and B is the blue channel component;

[0030] Step S33: Calculate the hue component H, and the calculation formula is:

[0031] ;

[0032] In the formula, H is the hue component, S is the saturation component, V is the luminance component, MAX() is the maximum value function, R is the red channel component, G is the green channel component, B is the blue channel component, and & is the logical AND operator;

[0033] Step S34: Image channel conversion and enhancement, specifically, convert the image format in the live face recognition dataset from RGB image to HSV space image by calculating the saturation component S, calculating the luminance component V, and calculating the hue component H, to obtain enhanced live pre-recognition data;

[0034] The image segmentation process for the personnel identity recognition dataset is specifically to perform image segmentation using the discrete cosine transform method to obtain face image segmentation data, including the following steps:

[0035] Step S35: Discrete cosine transform, and the calculation formula is:

[0036] ;

[0037] In the formula, is the output matrix after discrete cosine transform, i is the row index of the output matrix after discrete cosine transform, j is the column index of the output matrix after discrete cosine transform, is the i-th classification of coefficient C, is the j-th component of coefficient C, n is the total number of image pixels in the personnel identity recognition dataset, x is the horizontal direction index of the image pixels in the personnel identity recognition dataset, y is the vertical direction index of the image pixels in the personnel identity recognition dataset, h() is the filter function, is the frequency domain representation of the image in the personnel identity recognition dataset.

[0038] Further, in step S4, the live pre-identification is used to determine whether the face image is a live face through live pre-identification detection before temperature measurement, so as to prevent the measured temperature from coming from a photo or other organisms. Specifically, a method based on a convolutional neural network of a retina model is adopted to perform live pre-identification on the enhanced live pre-identification data. The convolutional neural network based on the retina model includes a basic convolutional subnet, a feature pyramid subnet, a localization subnet, and a classification subnet;

[0039] The basic convolutional subnet is used to extract features from the input image. The basic convolutional subnet has a basic convolutional network structure, including an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer;

[0040] The feature pyramid subnet is used to connect and fuse image features. Specifically, it performs multi-scale feature fusion on the features extracted by the basic convolutional subnet;

[0041] The localization subnet is used to extract the coordinate data of the live face object from the feature pyramid subnet. Specifically, it is trained through a smooth loss function to obtain the localization subnet;

[0042] The classification subnet is used to extract the classification data of the live face object from the feature pyramid subnet. Specifically, it is trained through a focal loss function to obtain the classification subnet;

[0043] The steps of performing live pre-identification on the enhanced live pre-identification data by using the method based on the convolutional neural network of the retina model include:

[0044] Step S41: Construct a basic convolutional subnet for extracting features from the input image. Specifically, construct an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer;

[0045] Step S42: Construct a feature pyramid subnet for connecting and fusing image features. Specifically, on the basis of the basic convolutional subnet, construct the feature pyramid subnet by designing a cross-scale structure. The cross-scale structure generates multi-scale low-resolution feature maps through downsampling and generates multi-scale high-resolution feature maps through upsampling. By constructing the feature pyramid subnet, multi-scale feature fusion is performed to obtain the live pre-identification fusion features;

[0046] Step S43: Construct multiple loss functions for model training, which specifically includes the following steps:

[0047] Step S431: Construct a smooth loss function, and the calculation formula is:

[0048] ;

[0049] In the formula, L Smoothis the smooth loss function, is the smooth threshold calculation parameter, where, is the threshold for switching from L1 loss to L2 loss, is the model error value, where, , y is the true value, F(x) is the model prediction value, || is the modulo operation;

[0050] Step S432: Construct the focal loss function, and the calculation formula is:

[0051] ;

[0052] In the formula, L Focal is the focal loss function, is the prediction probability, is the focal coefficient, and the focal coefficient is used to solve the class imbalance problem;

[0053] Step S433: Construct the model loss function for model training, and the calculation formula is:

[0054] ;

[0055] In the formula, L is the model loss function, is the balance coefficient, L Focal is the focal loss function, L Smooth is the smooth loss function;

[0056] Step S44: Construct a localization subnet for extracting the coordinate data of the live face object from the feature pyramid subnet, specifically training through the smooth loss function to obtain the localization subnet;

[0057] Step S45: Construct a classification subnet for extracting the classification data of the live face object from the feature pyramid subnet, specifically training through the focal loss function to obtain the classification subnet;

[0058] Step S46: Model training, specifically training the model through the construction of the basic convolutional subnet, the construction of the feature pyramid subnet, the construction of multiple loss functions, the construction of the localization subnet, and the construction of the classification subnet to obtain the live pre-recognition model Model PI ;

[0059] Step S47: Live pre-recognition, used to detect and determine whether a face image is a live face, specifically using the pre-recognition model Model PI to perform live pre-recognition to obtain pre-recognition filtered face data.

[0060] Further, in step S5, the personnel identification is used to determine whether the face image is that of an inmate, preventing other personnel from substituting for temperature measurement. Specifically, it uses a method of a convolutional neural network combined with the local binary pattern algorithm to perform personnel identification on the face image segmentation data. The convolutional neural network combined with the local binary pattern algorithm includes a local binary pattern operator and a convolutional network model;

[0061] The local binary pattern operator is used to extract the local texture features of the face to obtain the local texture features of the face;

[0062] The convolutional network model is used to construct a deep convolutional network model based on the local texture features of the face, and reduce the model complexity through shared weights and pooling downsampling techniques to perform feature classification of the face image;

[0063] The steps of using the method of a convolutional neural network combined with the local binary pattern algorithm to perform personnel identification on the face image segmentation data include:

[0064] Step S51: Use the local binary pattern algorithm to calculate the relationship between the pixel center point and the neighborhood. The calculation formula is:

[0065] ;

[0066] In the formula, LBP is the calculation result of the local binary pattern algorithm, d() is the grayscale difference calculation function, R i is the neighborhood pixel value, R j is the center point pixel value, P is the total number of pixels of the neighborhood pixel points, i is the neighborhood pixel point index, g i is the grayscale value of the neighborhood pixel point, g c is the center point pixel grayscale value, S() is the grayscale comparison boolean value calculation function, c() is the grayscale comparison calculation function, and x is the independent variable of the grayscale comparison boolean value calculation function S() used to calculate the grayscale comparison boolean value;

[0067] Step S52: Extract the local binary pattern histogram and generate local texture features, including the following steps:

[0068] Step S521: Perform secondary segmentation on the face image segmentation data to obtain secondary segmentation face image data;

[0069] Step S522: Extract local binary pattern features from the sub-paths of the secondary segmentation face image data and generate a local binary pattern histogram through statistics;

[0070] Step S523: Connect in sequence according to the local binary pattern histogram to generate local texture features to obtain the local texture features K for personnel identification;

[0071] Step S53: Construct a convolutional network model. Specifically, construct a convolutional network model based on the local texture features K for personnel identification, including the following steps:

[0072] Step S531: Local feature clustering. Specifically, use the faces in the local texture features K for personnel identification as the initial clustering centers, calculate the distances between the remaining local texture features of the faces and the initial clustering centers, and assign each calculated texture feature to the corresponding class. By iteratively adjusting the clustering centers of the classes, local feature clustering data is obtained;

[0073] Step S532: Construct the convolutional layer of the personnel identification model. Specifically, use the local feature clustering data as the input of the convolutional layer to construct the convolutional layer of the personnel identification model. The calculation formula is:

[0074] ;

[0075] In the formula, is the output feature map of the convolutional layer, max() is the maximum value calculation function, is the feature map index, is the convolution kernel, is the feature map index of the previous layer, is the convolution operator, is the local feature clustering data used as the input of the convolutional layer, LBP is the calculation result of the local binary algorithm, is the basis vector, used as the bias term;

[0076] Step S533: Construct the pooling layer of the personnel identification model. Specifically, use the max-pooling operation to construct the pooling layer of the personnel identification model, and use downsampling to reduce the spatial resolution of the convolutional layer and reduce the number of weights of the training face features;

[0077] Step S534: Construct the fully connected layer of the personnel identification model. Specifically, use the local feature clustering data as the input of the fully connected layer and activate it through the activation function to construct the fully connected layer of the personnel identification model. The calculation formula is:

[0078] ;

[0079] In the formula, is the output feature vector of the fully connected layer, F() is the activation function, W is the weight matrix, is the local feature clustering data used as the input of the fully connected layer, is the basis vector, used as the bias term;

[0080] Step S54: Training the identity recognition model. Specifically, calculate the relationship between the pixel center point and the neighborhood by using the local binary pattern algorithm, extract the local binary pattern histogram to construct a local binary pattern operator, and train the model by constructing a convolutional network model to obtain a personnel identity recognition model Model. ID ;

[0081] Step S55: Personnel identity recognition. Specifically, use the personnel identity recognition model Model ID to perform personnel identity recognition and obtain the identity data of the detainees.

[0082] Furthermore, in step S6, the contactless distance and temperature measurement is used to measure the body temperature under non-contact conditions. Specifically, a thermal imager and a thermal sensor are used for body temperature measurement, and combined with the identity data of the detainees, the body temperature data of the detainees is obtained.

[0083] Furthermore, in step S7, the self-service personnel management is used to perform self-service personnel management based on the pre-identified and filtered face data, the identity data of the detainees, and the body temperature data of the detainees. Specifically, set the body temperature abnormality threshold through the intelligent interaction terminal. The body temperature abnormality threshold includes a normal body temperature range and an abnormal body temperature range. The normal body temperature range is specifically 36 degrees Celsius to 37.2 degrees Celsius, and the abnormal body temperature range specifically refers to 37.2 degrees Celsius to 46.5 degrees Celsius. The intelligent interaction terminal obtains personnel management information through self-service personnel management and completes self-service personnel management by sending the personnel management information to the information interaction server.

[0084] The contactless body temperature measurement system based on face recognition provided by the present invention includes an interaction structure construction module, a data acquisition module, a data preprocessing module, a live pre-identification module, a personnel identity recognition module, a contactless distance and temperature measurement module, and a self-service personnel management module.

[0085] The interaction structure construction module is used to construct the interaction structure between the terminal and the server for contactless body temperature measurement, and perform self-service personnel management by establishing a connection. The self-service personnel management includes live pre-identification, personnel identity recognition, and contactless distance and temperature measurement. After establishing the connection, the interaction structure construction module sends the connection information to the data acquisition module;

[0086] The data acquisition module is used to collect the training data required for the live pre-identification and personnel identity recognition, collect the original face data, and send the original face data to the data preprocessing module;

[0087] The data preprocessing module is used to enhance the data quality of face image data. Through data preprocessing, optimized face image data is obtained. The optimized face image data includes enhanced liveness pre-recognition data and face image segmentation data. The data preprocessing module sends the enhanced liveness pre-recognition data to the liveness pre-recognition module and sends the face image segmentation data to the personnel identity recognition module;

[0088] The liveness pre-recognition module is used to, before temperature measurement, determine whether a face image is a live face through liveness pre-recognition detection, preventing the measured temperature from coming from a photo or other organisms. Through liveness pre-recognition, pre-recognition filtered face data is obtained and sent to the self-service personnel management module;

[0089] The personnel identity recognition module is used to determine whether a face image is that of an inmate, preventing other personnel from substituting for temperature measurement. Through personnel identity recognition, inmate identity data is obtained and sent to the contactless distance-controlled temperature measurement module and the self-service personnel management module;

[0090] The contactless distance-controlled temperature measurement module is used to measure body temperature under contactless conditions. Through contactless distance-controlled temperature measurement, inmate body temperature data is obtained.

[0091] The self-service personnel management module is used to perform self-service personnel management based on the pre-recognition filtered face data, the inmate identity data, and the inmate body temperature data. Through self-service personnel management, personnel management information is obtained, and by sending the personnel management information to the information interaction server, self-service personnel management is completed.

[0092] The beneficial effects achieved by the present invention using the above solution are as follows:

[0093] (1) Aiming at the technical problems in the existing contactless temperature measurement method based on face recognition, where the energy consumption and time consumption of face recognition are relatively high due to image quality problems, traditional RGB images can cause face image distortion, and unsegmented face images can reduce the accuracy of face recognition. This solution creatively uses image channel transformation enhancement processing and target segmentation processing to optimize the data quality of face images, providing a good data basis for subsequent liveness pre-recognition and personnel identity recognition;

[0094] (2) Aiming at the technical problems in the existing contactless temperature measurement method based on face recognition, where there is a lack of a method to prevent the measured temperature from coming from a photo or other organisms in the detention center scenario. This solution creatively uses the method of convolutional neural network based on the retina model for liveness pre-recognition, intelligently detecting and determining whether a face image is a live face, improving the overall accuracy of contactless temperature measurement;

[0095] (3) In the existing contactless temperature measurement method based on face recognition, there is a technical problem that there is a lack of a method for discriminating the identity of personnel before managing the body temperature of detainees. This solution creatively uses a method of combining a convolutional neural network with a local binary pattern algorithm for personnel identity recognition. By combining the living face and identity information judged by the pre-recognition of liveness, it can provide a more secure and effective contactless temperature measurement and personnel management in the detention center scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 It is a schematic flow chart of the contactless temperature measurement method based on face recognition provided by the present invention;

[0097] Figure 2 It is a schematic diagram of the contactless temperature measurement system based on face recognition provided by the present invention;

[0098] Figure 3 It is a data flow diagram of the contactless temperature measurement method based on face recognition provided by the present invention;

[0099] Figure 4 It is a schematic flow chart of step S3;

[0100] Figure 5 It is a schematic flow chart of step S4;

[0101] Figure 6 It is a schematic flow chart of step S5.

[0102] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0103] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0104] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention.

[0105] Example 1, refer to Figure 1 , the contactless temperature measurement method based on face recognition provided by the present invention, the method includes the following steps:

[0106] Step S1: Interaction structure construction;

[0107] Step S2: Data collection;

[0108] Step S3: Data preprocessing;

[0109] Step S4: Live pre-recognition;

[0110] Step S5: Personnel identity recognition;

[0111] Step S6: Contactless distance control and temperature measurement;

[0112] Step S7: Self-service personnel management.

[0113] Example 2, refer to Figure 1 , Figure 2 and Figure 3 , in step S1, the interaction structure construction is used to construct the interaction structure between the terminal and the server for contactless temperature measurement;

[0114] The terminal specifically refers to an intelligent interaction terminal, and the intelligent interaction terminal is used to register the face update notification of the server in the monitoring room when starting up and establish a connection with the server;

[0115] The server specifically refers to an information interaction server, and the information interaction server is used for entering face pictures and access information verification, and the face pictures specifically refer to the face pictures of the persons in custody in the detention center;

[0116] After the intelligent interaction terminal and the information interaction server establish a connection, self-service personnel management can be carried out, and the self-service personnel management includes live pre-recognition, personnel identity recognition and contactless distance control and temperature measurement.

[0117] Example 3, refer to Figure 1 , Figure 2 and Figure 3 , based on the above example, in step S2, the data collection is used to collect the training data required for the live pre-recognition and personnel identity recognition, specifically referring to obtaining the original face data from the portrait database, and the original face data includes the live face recognition data set and the personnel identity recognition data set.

[0118] Example 4, refer to Figure 1 , Figure 2 and Figure 4, this embodiment is based on the above embodiment. In step S3, the data preprocessing is used to enhance the data quality of the face image data, specifically referring to performing image enhancement, normalization, and smoothing operations on the original face data to obtain optimized face image data;

[0119] The optimized face image data includes enhanced liveness pre-recognition data and face image segmentation data;

[0120] Specifically, the enhanced liveness pre-recognition data is obtained by performing image channel conversion enhancement processing on the liveness face recognition data set in the original face data;

[0121] The face image segmentation data is obtained by performing target segmentation processing on the personnel identity recognition data set;

[0122] The image channel conversion enhancement processing is used to convert the RGB image into an HSV space image. The RGB image specifically refers to a weighted image of three colors: red, green, and blue, and the HSV space image specifically refers to a hue, saturation, and brightness space image;

[0123] The conversion of the RGB image into the HSV space image is specifically to generate the HSV space image using a light compensation method. The light compensation method specifically refers to calculating the hue component H, saturation component S, and brightness component V of the generated HSV space image through light compensation calculations, including the following steps:

[0124] Step S31: Calculate the saturation component S, and the calculation formula is:

[0125] ;

[0126] In the formula, S is the saturation component, MAX() is the maximum value function, MIN() is the minimum value function, R is the red channel component, G is the green channel component, and B is the blue channel component;

[0127] Step S32: Calculate the brightness component V, and the calculation formula is:

[0128] V = MAX(R, G, B);

[0129] In the formula, V is the brightness component, MAX() is the maximum value function, R is the red channel component, G is the green channel component, and B is the blue channel component;

[0130] Step S33: Calculate the hue component H, and the calculation formula is:

[0131] ;

[0132] Wherein, H is the hue component, S is the saturation component, V is the brightness component, MAX() is the maximum value function, R is the red channel component, G is the green channel component, B is the blue channel component, and & is the logical AND operator;

[0133] Step S34: Image channel conversion and enhancement, specifically, by calculating the saturation component S, calculating the brightness component V, and calculating the hue component H, convert the image format in the live face recognition dataset from RGB image to HSV space image to obtain enhanced live pre-recognition data;

[0134] The image segmentation process for the personnel identity recognition dataset is specifically to perform image segmentation using the discrete cosine transform method to obtain face image segmentation data, including the following steps:

[0135] Step S35: Discrete cosine transform, and the calculation formula is:

[0136] ;

[0137] Wherein, is the output matrix after discrete cosine transform, i is the row index of the output matrix after discrete cosine transform, j is the column index of the output matrix after discrete cosine transform, is the i-th classification of the coefficient C, is the j-th component of the coefficient C, n is the total number of image pixels in the personnel identity recognition dataset, x is the horizontal direction index of the image pixels in the personnel identity recognition dataset, y is the vertical direction index of the image pixels in the personnel identity recognition dataset, and h() is the filter function, is the frequency domain representation of the image in the personnel identity recognition dataset.

[0138] By performing the above operations, in the existing contactless temperature measurement method based on face recognition, there are technical problems that the energy consumption and time consumption of face recognition are relatively high due to image quality problems, traditional RGB images will cause distortion of face images, and unsegmented face images will reduce the accuracy of face recognition. This solution creatively uses image channel conversion and enhancement processing and target segmentation processing to optimize the data quality of face images, providing a good data basis for subsequent live pre-recognition and personnel identity recognition.

[0139] Example Five, refer to Figure 1 、 Figure 2 、 Figure 3 and Figure 5, this embodiment is based on the above embodiment. In step S4, the live pre-identification is used to detect and determine whether the face image is a live face through live pre-identification before temperature measurement, so as to prevent the measured temperature from coming from a photo or other organisms. Specifically, a method based on a convolutional neural network of the retina model is adopted to perform live pre-identification on the enhanced live pre-identification data. The convolutional neural network based on the retina model includes a basic convolutional subnet, a feature pyramid subnet, a localization subnet, and a classification subnet;

[0140] The basic convolutional subnet is used to extract features from the input image. The basic convolutional subnet has a basic convolutional network structure, including an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer;

[0141] The feature pyramid subnet is used to connect and fuse image features. Specifically, it performs multi-scale feature fusion on the features extracted by the basic convolutional subnet;

[0142] The localization subnet is used to extract the coordinate data of the live face object from the feature pyramid subnet. Specifically, it is trained through a smooth loss function to obtain the localization subnet;

[0143] The classification subnet is used to extract the classification data of the live face object from the feature pyramid subnet. Specifically, it is trained through a focal loss function to obtain the classification subnet;

[0144] The steps of performing live pre-identification on the enhanced live pre-identification data by using the method based on the convolutional neural network of the retina model include:

[0145] Step S41: Construct a basic convolutional subnet for extracting features from the input image. Specifically, construct an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer;

[0146] Step S42: Construct a feature pyramid subnet for connecting and fusing image features. Specifically, on the basis of the basic convolutional subnet, construct the feature pyramid subnet by designing a cross-scale structure. The cross-scale structure generates multi-scale low-resolution feature maps through downsampling and generates multi-scale high-resolution feature maps through upsampling. By constructing the feature pyramid subnet, multi-scale feature fusion is performed to obtain the live pre-identification fusion features;

[0147] Step S43: Construct multiple loss functions for model training, specifically including the following steps:

[0148] Step S431: Construct a smooth loss function, and the calculation formula is:

[0149] ;

[0150] In the formula, LSmooth is the smooth loss function, is the smooth threshold calculation parameter, where, is the threshold for switching from L1 loss to L2 loss, is the model error value, where, , y is the true value, F(x) is the model prediction value, and || is the modulo operation;

[0151] Step S432: Construct the focal loss function, and the calculation formula is:

[0152] ;

[0153] In the formula, L Focal is the focal loss function, is the predicted probability, is the focal coefficient, and the focal coefficient is used to solve the class imbalance problem;

[0154] Step S433: Construct the model loss function for model training, and the calculation formula is:

[0155] ;

[0156] In the formula, L is the model loss function, is the balance coefficient, L Focal is the focal loss function, L Smooth is the smooth loss function;

[0157] Step S44: Construct a localization subnet for extracting the coordinate data of the live face object from the feature pyramid subnet, specifically, training through the smooth loss function to obtain the localization subnet;

[0158] Step S45: Construct a classification subnet for extracting the classification data of the live face object from the feature pyramid subnet, specifically, training through the focal loss function to obtain the classification subnet;

[0159] Step S46: Model training, specifically, training the model through the construction of the basic convolutional subnet, the construction of the feature pyramid subnet, the construction of multiple loss functions, the construction of the localization subnet, and the construction of the classification subnet to obtain the live pre-recognition model Model PI ;

[0160] Step S47: Live pre-recognition, used to detect and judge whether a face image is a live face, specifically, using the pre-recognition model Model PI to perform live pre-recognition to obtain pre-recognition filtered face data;

[0161] By performing the above operations, in the existing contactless temperature measurement method based on face recognition, there is a technical problem that there is a lack of a method to prevent the measured temperature from coming from a photo or other organisms in the detention center scenario. This solution creatively uses a method based on a convolutional neural network of a retina model for live pre-identification, intelligently detecting and judging whether a face image is a live face, and improving the overall accuracy of contactless temperature measurement.

[0162] Example 6, refer to Figure 1 、 Figure 2 、 Figure 3 and Figure 6 Based on the above embodiments, in step S5, the personnel identity recognition is used to determine whether a face image is an inmate, preventing other personnel from substituting for temperature measurement. Specifically, a method based on a convolutional neural network combined with a local binary pattern algorithm is used to perform personnel identity recognition on the face image segmentation data. The convolutional neural network combined with the local binary pattern algorithm includes a local binary pattern operator and a convolutional network model;

[0163] The local binary pattern operator is used to extract the local texture features of the face to obtain the local texture features of the face;

[0164] The convolutional network model is used to construct a deep convolutional network model based on the local texture features of the face, and reduce the model complexity through shared weights and pooling downsampling techniques for feature classification of the face image;

[0165] The step of using the method based on a convolutional neural network combined with a local binary pattern algorithm to perform personnel identity recognition on the face image segmentation data includes:

[0166] Step S51: Use the local binary pattern algorithm to calculate the relationship between the pixel center point and the neighborhood. The calculation formula is:

[0167] ;

[0168] In the formula, LBP is the calculation result of the local binary pattern algorithm, d() is the gray difference calculation function, R i is the neighborhood pixel value, R j is the center point pixel value, P is the total number of pixels of the neighborhood pixel points, i is the neighborhood pixel point index, g i is the gray value of the neighborhood pixel point, g c is the center point pixel gray value, S() is the gray comparison boolean value calculation function, c() is the gray comparison calculation function, and x is the independent variable of the gray comparison boolean value calculation function S() for calculating the gray comparison boolean value;

[0169] Step S52: Extract the local binary pattern histogram and generate local texture features, including the following steps:

[0170] Step S521: Perform secondary segmentation on the face image segmentation data to obtain secondary segmented face image data;

[0171] Step S522: Extract local binary features from the sub-paths of the secondary segmented face image data, and generate a local binary histogram through statistics;

[0172] Step S523: Connect the local binary histograms in sequence to generate local texture features, and obtain the local texture feature K for personnel identity recognition;

[0173] Step S53: Construct a convolutional network model. Specifically, construct a convolutional network model based on the local texture feature K for personnel identity recognition, including the following steps:

[0174] Step S531: Local feature clustering. Specifically, use the faces in the local texture feature K for personnel identity recognition as the initial clustering centers, calculate the distances between the remaining face local texture features and the initial clustering centers, and assign each calculated texture feature to the corresponding class. Adjust the clustering centers of the classes through iteration to obtain local feature clustering data;

[0175] Step S532: Construct the convolutional layer of the personnel identity recognition model. Specifically, use the local feature clustering data as the input of the convolutional layer to construct the convolutional layer of the personnel identity recognition model. The calculation formula is:

[0176] ;

[0177] In the formula, is the output feature map of the convolutional layer, max() is the maximum value calculation function, is the feature map index, is the convolutional kernel, is the feature map index of the previous layer, is the convolutional operator, is the local feature clustering data used as the input of the convolutional layer, LBP is the calculation result of the local binary algorithm, is the basis vector, used as the bias term;

[0178] Step S533: Construct the pooling layer of the personnel identity recognition model. Specifically, use the maximum pooling operation to construct the pooling layer of the personnel identity recognition model, and use downsampling to reduce the spatial resolution of the convolutional layer and reduce the number of weights of the training face features;

[0179] Step S534: Construct the fully connected layer of the personnel identity recognition model. Specifically, use the local feature clustering data as the input of the fully connected layer and activate it through an activation function to construct the fully connected layer of the personnel identity recognition model. The calculation formula is as follows:

[0180] ;

[0181] In the formula, is the output feature vector of the fully connected layer, F() is the activation function, W is the weight matrix, is the local feature clustering data used as the input of the fully connected layer, is the basis vector, which is used as the bias term;

[0182] Step S54: Train the identity recognition model. Specifically, calculate the relationship between the pixel center point and the neighborhood by using the local binary pattern algorithm and extract the local binary pattern histogram to construct the local binary pattern operator, and perform model training through the constructed convolutional network model to obtain the personnel identity recognition model Model ID ;

[0183] Step S55: Personnel identity recognition. Specifically, use the personnel identity recognition model Model ID to perform personnel identity recognition and obtain the identity data of the detainees;

[0184] By performing the above operations, in the existing contactless temperature measurement method based on face recognition, there is a technical problem that there is a lack of a method for discriminating personnel identity before managing the body temperature of detainees. This solution creatively uses the method of combining a convolutional neural network with the local binary pattern algorithm for personnel identity recognition, and combines the living face and identity information judged by the living body pre-recognition, which can provide a more secure and effective contactless temperature measurement and personnel management in the detention center scenario.

[0185] Example 7, refer to Figure 2 and Figure 3 In this example, based on the above example, in step S6, the contactless distance-controlled temperature measurement is used to measure the body temperature under contactless conditions. Specifically, a thermal imager and a thermal sensor are used for body temperature measurement, and combined with the identity data of the detainees, the body temperature data of the detainees is obtained

[0186] Example 8, refer to Figure 2 and Figure 3, based on the above embodiment, in step S7, the self-service personnel management is used to perform self-service personnel management based on the pre-identified and filtered face data, the identity data of the detainees, and the body temperature data of the detainees. Specifically, the intelligent interaction terminal is used to set the body temperature abnormality threshold, which includes the normal body temperature range and the abnormal body temperature range. The normal body temperature range is specifically 36 degrees Celsius to 37.2 degrees Celsius, and the abnormal body temperature range specifically refers to 37.2 degrees Celsius to 46.5 degrees Celsius. The intelligent interaction terminal obtains the personnel management information through self-service personnel management and completes the self-service personnel management by sending the personnel management information to the information interaction server.

[0187] Embodiment Nine, refer to Figure 2 and Figure 3 , based on the above embodiment, the contactless body temperature measurement system based on face recognition provided by the present invention includes an interactive structure construction module, a data acquisition module, a data preprocessing module, a living body pre-identification module, a personnel identity recognition module, a contactless distance control and temperature measurement module, and a self-service personnel management module.

[0188] The interactive structure construction module is used to construct the interactive structure between the terminal and the server for contactless body temperature measurement, and perform self-service personnel management by establishing a connection. The self-service personnel management includes living body pre-identification, personnel identity recognition, and contactless distance control and temperature measurement. After establishing the connection, the interactive structure construction module sends the connection information to the data acquisition module;

[0189] The data acquisition module is used to collect the training data required for the living body pre-identification and personnel identity recognition, obtain the original face data through acquisition, and send the original face data to the data preprocessing module;

[0190] The data preprocessing module is used to enhance the data quality of the face image data. Through data preprocessing, optimized face image data is obtained. The optimized face image data includes enhanced living body pre-identification data and face image segmentation data. The data preprocessing module sends the enhanced living body pre-identification data to the living body pre-identification module and sends the face image segmentation data to the personnel identity recognition module;

[0191] The living body pre-identification module is used to detect and judge whether the face image is a living face through living body pre-identification before temperature measurement, prevent the measured temperature from coming from a photo or other organisms. Through living body pre-identification, pre-identified and filtered face data is obtained, and the pre-identified and filtered face data is sent to the self-service personnel management module;

[0192] The personnel identification module is used to determine whether a face image belongs to an inmate, prevent other personnel from substituting for temperature measurement, obtain inmate identity data through personnel identification, and send the inmate identity data to the contactless distance-controlled temperature measurement module and the self-service personnel management module.

[0193] The contactless distance-controlled temperature measurement module is used to measure body temperature under contactless conditions and obtain inmate body temperature data through contactless distance-controlled temperature measurement.

[0194] The self-service personnel management module is used to perform self-service personnel management based on the pre-identified and filtered face data, the inmate identity data, and the inmate body temperature data, obtain personnel management information through self-service personnel management, and complete self-service personnel management by sending the personnel management information to the information interaction server.

[0195] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0196] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

[0197] The above describes the present invention and its implementation manners. Such a description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. Contactless temperature measurement method based on face recognition, Characterized in that: This method includes the following steps: Step S1: Construction of an interaction structure; Step S2: Data collection to obtain original face data; Step S3: Data preprocessing to obtain optimized face image data, and the optimized face image data includes enhanced live pre-recognition data and face image segmentation data; Step S4: Live pre-recognition, which is used to detect and judge whether the face image is a live face before temperature measurement to prevent the measured temperature from coming from a photo or other organisms. Specifically, a method based on a convolutional neural network of a retina model is used to perform live pre-recognition on the enhanced live pre-recognition data. The convolutional neural network based on the retina model includes a basic convolutional subnet, a feature pyramid subnet, a localization subnet, and a classification subnet; The basic convolutional subnet is used to extract features from the input image. The basic convolutional subnet has a basic convolutional network structure, including an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; The feature pyramid subnet is used to connect and fuse image features. Specifically, it performs multi-scale feature fusion on the features extracted by the basic convolutional subnet; The localization subnet is used to extract the coordinate data of the live face object from the feature pyramid subnet. Specifically, it is trained through a smooth loss function to obtain the localization subnet; The classification subnet is used to extract the classification data of the live face object from the feature pyramid subnet. Specifically, it is trained through a focal loss function to obtain the classification subnet; Step S5: Personnel identity recognition, which is used to judge whether the face image is an inmate to prevent other personnel from substituting for temperature measurement. Specifically, a method based on a convolutional neural network combined with a local binary pattern algorithm is used to perform personnel identity recognition on the face image segmentation data. The convolutional neural network combined with the local binary pattern algorithm includes a local binary pattern operator and a convolutional network model; The local binary pattern operator is used to extract the local texture features of the face to obtain face local texture features; The convolutional network model is used to construct a deep convolutional network model based on the face local texture features, and reduce the model complexity through shared weights and pooling downsampling techniques for feature classification of face images; The step of performing personnel identity recognition on the face image segmentation data by using the method of a convolutional neural network combined with a local binary pattern algorithm includes: Step S51: Use the local binary pattern algorithm to calculate the relationship between the pixel center point and the neighborhood. The calculation formula is: Wherein, LBP is the calculation result of the local binarization algorithm, d() is the grayscale difference calculation function, R i is the neighborhood pixel value, R j is the central point pixel value, P is the total number of pixels of the neighborhood pixel points, i is the neighborhood pixel point index, g i is the grayscale value of the neighborhood pixel point, g c is the grayscale value of the central point pixel, S() is the grayscale contrast boolean value calculation function, c() is the grayscale contrast calculation function, x is the independent variable of the grayscale contrast boolean value calculation function S(), and is used to calculate the grayscale contrast boolean value; Step S52: Extract the local binary pattern histogram and generate local texture features, including the following steps: Step S521: Perform secondary segmentation on the face image segmentation data to obtain secondary segmented face image data; Step S522: Extract local binary pattern features from the subgraphs of the secondary segmented face image data, and generate a local binary pattern histogram through statistics; Step S523: Connect and generate local texture features in sequence according to the local binary pattern histogram to obtain the local texture feature K for personnel identity recognition; Step S53: Construct a convolutional network model. Specifically, construct a convolutional network model based on the local texture features K of personnel identity recognition, including the following steps: Step S531: Local feature clustering. Specifically, use the faces in the local texture features K of personnel identity recognition as the initial clustering centers, calculate the distances between the remaining local texture features of the faces and the initial clustering centers, and assign each calculated texture feature to the corresponding class. By iteratively adjusting the clustering centers of the classes, obtain the local feature clustering data; Step S532: Construct the convolutional layer of the personnel identity recognition model. Specifically, use the local feature clustering data as the input of the convolutional layer to construct the convolutional layer of the personnel identity recognition model. The calculation formula is: Where Y i′ is the output feature map of the convolutional layer, max() is the maximum value calculation function, i′ is the feature map index, K i′j′ is the convolution kernel, j′ is the feature map index of the previous layer, is the convolution operator, X′ i′ is the local feature clustering data as the input of the convolutional layer, LBP is the calculation result of the local binarization algorithm, and B′ is the basis vector used as the bias term; Step S533: Construct the pooling layer of the personnel identity recognition model. Specifically, use the maximum pooling operation to construct the pooling layer of the personnel identity recognition model, and use downsampling to reduce the spatial resolution of the convolutional layer and reduce the number of weights of the training face features; Step S534: Construct the fully connected layer of the personnel identity recognition model. Specifically, use the local feature clustering data as the input of the fully connected layer and activate it through an activation function to construct the fully connected layer of the personnel identity recognition model. The calculation formula is: where Y″ is the output feature vector of the fully connected layer, F() is the activation function, W is the weight matrix, and X″ i′ is the local feature clustering data serving as the input of the fully connected layer, and B′ is the basis vector used as the bias term; Step S54: Training of the identity recognition model. Specifically, calculate the relationship between the pixel center point and the neighborhood by using the local binary pattern algorithm, extract the local binary pattern histogram to construct the local binary pattern operator, and train the model by constructing a convolutional network model to obtain the personnel identity recognition model Model ID ; Step S55: Personnel identity recognition, specifically using the described personnel identity recognition model Model ID to perform personnel identity recognition and obtain the identity data of the detainees; Step S6: Contactless distance and temperature measurement, which is used to measure body temperature under contactless conditions. Specifically, use a thermal imager and a thermal sensor to measure body temperature, and combine the identity data of the detainees to obtain the body temperature data of the detainees; Step S7: Self-service personnel management.

2. The contactless body temperature measurement method based on face recognition according to claim 1, characterized in that: In step S3, the data preprocessing is used to enhance the data quality of the face image data. Specifically, it refers to performing image enhancement, normalization, and smoothing operations on the original face data to obtain optimized face image data; The optimized face image data includes enhanced live pre-recognition data and face image segmentation data; The enhanced live pre-recognition data is obtained by performing image channel conversion and enhancement processing on the live face recognition data set in the original face data; The face image segmentation data is obtained by performing target segmentation processing on the personnel identity recognition data set; The image channel conversion and enhancement processing is used to convert the RGB image into an HSV space image. The RGB image specifically refers to a weighted image of three colors: red, green, and blue. The HSV space image specifically refers to a hue, saturation, and brightness space image; The conversion of the RGB image into the HSV space image is specifically to generate the HSV space image using a light compensation method. The light compensation method specifically refers to calculating the hue component H, saturation component S, and brightness component V of the generated HSV space image through light compensation calculations, including the following steps: Step S31: Calculate the saturation component S. The calculation formula is: In the formula, S is the saturation component, MAX() is the maximum value function, MIN() is the minimum value function, R is the red channel component, G is the green channel component, and B is the blue channel component; Step S32: Calculate the brightness component V. The calculation formula is: V = MAX(R, G, B); Wherein, V is the luminance component, MAX() is the maximum value function, R is the red channel component, G is the green channel component, and B is the blue channel component; Step S33: Calculate the hue component H, and the calculation formula is: Wherein, H is the hue component, S is the saturation component, V is the luminance component, MAX() is the maximum value function, R is the red channel component, G is the green channel component, B is the blue channel component, and & is the logical AND operator; Step S34: Image channel conversion and enhancement, specifically by calculating the saturation component S, calculating the luminance component V, and calculating the hue component H, converting the image format in the live face recognition data set from an RGB image to an HSV space image, and obtaining enhanced live pre-recognition data; The image segmentation process for the personnel identity recognition data set is specifically to perform image segmentation processing using the discrete cosine transform method to obtain face image segmentation data, including the following steps: Step S35: Discrete cosine transform.

3. The contactless temperature measurement method based on face recognition according to claim 1, characterized in that: In step S4, the step of performing live pre-recognition on the enhanced live pre-recognition data by using the method of a convolutional neural network based on a retina model includes: Step S41: Construct a basic convolutional subnet for extracting features from the input image, specifically constructing an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; Step S42: Construct a feature pyramid subnet for connecting and fusing image features. Specifically, on the basis of the basic convolutional subnet, by designing a cross-scale structure, construct the feature pyramid subnet. The cross-scale structure generates multi-scale low-resolution feature maps through downsampling and generates multi-scale high-resolution feature maps through upsampling. By constructing the feature pyramid subnet, multi-scale feature fusion is performed to obtain live pre-recognition fusion features; Step S43: Construct multiple loss functions for model training, specifically including the following steps: Step S431: Construct a smooth loss function, and the calculation formula is: where L Smooth is the smoothing loss function, β is the smoothing threshold calculation parameter, where is the threshold for switching from L1 loss to L2 loss, ε is the model error value, where ε = |y - F(x)|, y is the true value, F(x) is the model predicted value, and || is the modulo operation; Step S432: Construct a focal loss function, and the calculation formula is: L Focal = -(1 - p t ) γ log p t ; where L Focal is the focal loss function, p t is the predicted probability, and γ is the focal coefficient, which is used to solve the class imbalance problem; Step S433: Construct a model loss function for model training, and the calculation formula is: L = αL Focal +L Smooth ; where L is the model loss function, α is the balance coefficient, L Focal is the focal loss function, and L Smooth is the smooth loss function; Step S44: Construct a localization subnet for extracting the coordinate data of the live face object from the feature pyramid subnet, specifically training through the smooth loss function to obtain the localization subnet; Step S45: Construct a classification subnet for extracting the classification data of the live face object from the feature pyramid subnet, specifically training through the focal loss function to obtain the classification subnet; Step S46: Model training, specifically, training the model by constructing a basic convolutional subnet, constructing a feature pyramid subnet, constructing multiple loss functions, constructing a localization subnet, and constructing a classification subnet to obtain a live pre-recognition model Model PI ; Step S47: Live pre-recognition, which is used to detect and determine whether a face image is a live face. Specifically, the pre-recognition model Model is used PI to perform live pre-recognition and obtain pre-recognition filtered face data.

4. The contactless temperature measurement method based on face recognition according to claim 3, characterized in that: In step S1, the interactive structure construction is used to construct the interactive structure between the terminal and the server for contactless temperature measurement; The terminal specifically refers to an intelligent interactive terminal, and the intelligent interactive terminal is used to register the server face update notification of the monitoring room at startup and establish a connection with the server; The server specifically refers to an information interaction server, which is used for inputting face pictures and verifying access information. The face pictures specifically refer to the face pictures of detainees in the detention cell. After establishing a connection with the information interaction server, the intelligent interaction terminal conducts self-service personnel management, which includes live pre-identification, personnel identity identification, and contactless distance and temperature measurement.

5. The contactless temperature measurement method based on face recognition according to claim 4, characterized in that: In step S7, the self-service personnel management is used to perform self-service personnel management based on the pre-identified filtered face data, the identity data of detainees, and the body temperature data of detainees. Specifically, a body temperature abnormality threshold is set through the intelligent interaction terminal. The body temperature abnormality threshold includes a normal body temperature range and an abnormal body temperature range. The normal body temperature range is specifically 36 degrees Celsius to 37.2 degrees Celsius, and the abnormal body temperature range specifically refers to 37.2 degrees Celsius to 46.5 degrees Celsius. The intelligent interaction terminal obtains personnel management information through self-service personnel management and completes self-service personnel management by sending the personnel management information to the information interaction server.

6. A contactless temperature measurement system based on face recognition is used to implement the contactless temperature measurement method based on face recognition according to any one of claims 1-5, characterized in that: It includes an interaction structure construction module, a data collection module, a data preprocessing module, a live pre-identification module, a personnel identity identification module, a contactless distance and temperature measurement module, and a self-service personnel management module.

7. The contactless temperature measurement system based on face recognition according to claim 6, characterized in that: The interaction structure construction module is used to construct the interaction structure between the terminal and the server for contactless temperature measurement, and through establishing a connection, perform self-service personnel management, which includes live pre-identification, personnel identity identification, and contactless distance and temperature measurement. After establishing a connection, the interaction structure construction module sends the connection information to the data collection module; The data collection module is used to collect the training data required for the live pre-identification and personnel identity identification, obtain the original face data through collection, and send the original face data to the data preprocessing module; The data preprocessing module is used to enhance the data quality of the face image data. Through data preprocessing, optimized face image data is obtained. The optimized face image data includes enhanced live pre-identification data and face image segmentation data. The data preprocessing module sends the enhanced live pre-identification data to the live pre-identification module and sends the face image segmentation data to the personnel identity identification module; The live pre-identification module is used to detect and judge whether the face image is a live face through live pre-identification before temperature measurement, prevent the measured temperature from coming from a photo or other organisms. Through live pre-identification, pre-identified filtered face data is obtained, and the pre-identified filtered face data is sent to the self-service personnel management module; The personnel identification module is used to determine whether a face image belongs to an inmate, prevent other personnel from substituting for temperature measurement, obtain inmate identity data through personnel identification, and send the inmate identity data to the contactless distance-controlled temperature measurement module and the self-service personnel management module; The contactless distance-controlled temperature measurement module is used to measure body temperature under contactless conditions and obtain inmate body temperature data through contactless distance-controlled temperature measurement; The self-service personnel management module is used to perform self-service personnel management based on the pre-identified and filtered face data, the inmate identity data, and the inmate body temperature data, obtain personnel management information through self-service personnel management, and complete self-service personnel management by sending the personnel management information to the information interaction server.

Images