Intelligent Turnstile System and Method Based on Image Recognition Technology

Abstract

The present application provides an intelligent turnstile system and method based on image recognition technology, which relates to the field of intelligent recognition. It obtains a face reference image and age information from an ID card recognition image, collects a face detection image by a camera, and uses a data processing algorithm based on deep learning to perform face feature analysis on the face reference image and the face detection image, and encodes the age information. Thus, it intelligently determines whether the identities match based on the face comparison features obtained by calculating the difference between the face reference correction features and the face detection features obtained with the assistance of the age information, and controls the gate opening instruction. In this way, the accuracy of cross-age identity recognition can be significantly improved. At the same time, thanks to the efficient algorithm for data processing, the processing speed of face recognition is also increased, thereby ensuring efficient passage in crowded places such as public transportation.

Description

Technical Field

[0001] This application relates to the field of intelligent recognition, and more specifically, to an intelligent turnstile system and method based on image recognition technology. Background Art

[0002] A turnstile is an automated access control device widely used in various places that need to control the flow of people in and out, such as subway stations, railway stations, airports, office buildings, residential communities, scenic spots, etc. With the rapid development of technology, especially in the fields of artificial intelligence and big data, safe and convenient management of people flow has become crucial, and the modern society has an increasing demand for efficient and intelligent access control turnstile systems.

[0003] However, traditional face recognition technology has limitations in many aspects. For example, it does not fully consider the impact of age changes on facial features. Specifically, as people age, facial features change significantly, such as wrinkles, skin color, and facial contour changes, which makes it difficult for traditional algorithms to accurately recognize faces after time changes. In addition, in terms of real-time performance, traditional algorithms are usually computationally complex and have a slow processing speed. Especially in crowded places, recognition delays may affect the passage efficiency. This limitation is particularly obvious in scenarios such as public transportation where quick identity verification is required, which may lead to queues and congestion, affecting the user experience.

[0004] Therefore, an optimized intelligent turnstile solution is needed. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide an intelligent turnstile system and method based on image recognition technology. It obtains a face reference image and age information from an ID card recognition image, collects a face detection image by a camera, and uses a data processing algorithm based on deep learning to perform face feature analysis on the face reference image and the face detection image, and encodes the age information. Thus, it intelligently determines whether the identities match based on the face comparison features obtained by calculating the difference between the face reference correction features and the face detection features with the assistance of age information, and controls the gate opening instruction. In this way, the accuracy of cross-age identity recognition can be significantly improved. At the same time, thanks to the efficient algorithm for data processing, the processing speed of face recognition is also increased, thereby ensuring efficient passage in crowded places such as public transportation.

[0006] According to one aspect of this application, an intelligent turnstile system based on image recognition technology is provided, which includes:

[0007] A face reference image and age information acquisition module, configured to obtain a face reference image and age information from an ID card recognition image;

[0008] A face reference feature extraction module, configured to input the face reference image into a face feature extractor to obtain a face reference feature map;

[0009] An age information one-hot encoding module, configured to perform one-hot encoding on the age information to obtain an age information one-hot encoding vector;

[0010] An age information-face reference feature joint encoding module, configured to input the age information one-hot encoding vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map;

[0011] A face detection image acquisition module, configured to acquire a face detection image collected by a camera;

[0012] A face detection feature extraction module, configured to input the face detection image into the face feature extractor to obtain a face detection feature map;

[0013] A face feature difference calculation module, configured to calculate a difference feature map between the face detection feature map and the age-assisted face reference corrected feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature;

[0014] An authentication result generation module, configured to obtain an authentication result based on the face comparison high-dimensional feature, and determine whether to generate a gate opening instruction based on the authentication result.

[0015] In the above intelligent gate system based on image recognition technology, the face reference feature extraction module is configured to: input the face reference image into a face feature extractor based on a dilated convolutional neural network model to obtain the face reference feature map.

[0016] In the above intelligent turnstile system based on image recognition technology, the age information-face reference feature joint encoding module includes: an age information feature linear transformation shape reshaping unit, configured to perform linear transformation and feature shape reshaping on the age information one-hot encoding vector to obtain an age information one-hot encoding modulation matrix; a face reference feature linear transformation unit, configured to perform a first linear transformation and a second linear transformation on the face reference feature map to obtain a first linearly transformed face reference feature map and a second linearly transformed face reference feature map; a face reference feature linear transformation reshaping unit, configured to perform feature shape reshaping on the first linearly transformed face reference feature map and the second linearly transformed face reference feature map to obtain a first linearly transformed face reference feature matrix and a second linearly transformed face reference feature matrix; an age information one-hot encoding attention calculation unit, configured to use the first linearly transformed face reference feature matrix and the second linearly transformed face reference feature matrix as the key matrix and the value matrix, and use the age information one-hot encoding modulation matrix as the query matrix, and input the query matrix, the key matrix, and the value matrix into the Transformer structure to obtain an age information one-hot encoding attention query face reference matrix; an age information feature encoding reshaping unit, configured to perform shape reshaping and point convolution encoding on the age information one-hot encoding attention query face reference matrix to obtain an age information one-hot encoding attention query face reference feature map; an age-assisted face feature correction unit, configured to perform element-wise multiplication on the age information one-hot encoding attention query face reference feature map and the face reference feature map in position to obtain the age-assisted face reference correction feature map.

[0017] In the above intelligent turnstile system based on image recognition technology, the age information feature linear transformation shape reshaping unit is configured to: multiply the age information one-hot encoding vector and the age information modulation matrix in matrix to obtain an age information modulation vector, and then add the age information modulation vector and the age information bias vector in position to obtain an age information one-hot encoding modulation bias vector; input the age information one-hot encoding modulation bias vector into the Sigmoid function for activation processing to obtain an age information one-hot encoding modulation activation vector; perform shape reshaping on the age information one-hot encoding modulation activation vector to obtain the age information one-hot encoding modulation matrix.

[0018] In the above intelligent turnstile system based on image recognition technology, the face reference feature linear transformation unit is used to: perform element-wise multiplication of each feature matrix along the channel dimension in the face reference feature map with the first face reference modulation matrix and then perform element-wise addition with the first face reference bias matrix to obtain a first face reference bias feature map; input the first face reference bias feature map into the Sigmoid function for activation processing to obtain the first linearly transformed face reference feature map; perform element-wise multiplication of each feature matrix along the channel dimension in the face reference feature map with the second face reference modulation matrix and then perform element-wise addition with the second face reference bias matrix to obtain a second face reference bias feature map; input the second face reference bias feature map into the Sigmoid function for activation processing to obtain the second linearly transformed face reference feature map.

[0019] In the above intelligent turnstile system based on image recognition technology, the age information one-hot encoded attention calculation unit is used to: calculate the matrix product between the transpose matrix of the age information one-hot encoded modulation matrix and the first linearly transformed face reference feature matrix and then input it into the softmax function to obtain an age information-face reference feature correlation matrix; use the age information-face reference feature correlation matrix as a weight matrix and calculate the matrix product between the weight matrix and the second linearly transformed face reference feature matrix to obtain the age information one-hot encoded attention query face reference matrix.

[0020] In the above intelligent turnstile system based on image recognition technology, the face detection feature extraction module is used to: input the face detection image into the face feature extractor based on the dilated convolutional neural network model to obtain the face detection feature map.

[0021] In the above intelligent turnstile system based on image recognition technology, the authentication result generation module includes: a face feature parsing unit, which is used to input the face comparison high-dimensional feature map into an identity authenticator based on a classifier to obtain the authentication result, and the authentication result is used to indicate whether the identities match; a gate opening instruction generation unit, which is used to generate a gate opening instruction in response to the authentication result being a match.

[0022] According to another aspect of the present application, there is provided an intelligent turnstile method based on image recognition technology, which includes:

[0023] Obtain a face reference image and age information from the ID card recognition image;

[0024] Input the face reference image into a face feature extractor to obtain a face reference feature map;

[0025] One-hot encode the age information to obtain an age information one-hot encoded vector;

[0026] Input the age information one-hot encoded vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map;

[0027] Obtain a face detection image collected by a camera;

[0028] Input the face detection image into the face feature extractor to obtain a face detection feature map;

[0029] Calculate a differential feature map between the face detection feature map and the age-assisted face reference corrected feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature;

[0030] Based on the face comparison high-dimensional feature, obtain an authentication result, and based on the authentication result, determine whether to generate a gate opening instruction.

[0031] In the above intelligent gate method based on image recognition technology, inputting the face reference image into the face feature extractor to obtain a face reference feature map includes: inputting the face reference image into a face feature extractor based on a dilated convolutional neural network model to obtain the face reference feature map.

[0032] Compared with the prior art, the intelligent gate system and method provided by the present application obtain a face reference image and age information from an ID card recognition image, collect a face detection image by a camera, and use a data processing algorithm based on deep learning to perform face feature analysis on the face reference image and the face detection image, encode the age information, and thus intelligently determine whether the identities match based on the face comparison feature obtained by calculating the difference between the face reference corrected feature and the face detection feature obtained with the assistance of age information, and control the gate opening instruction. In this way, the accuracy of cross-age identity recognition can be significantly improved. At the same time, thanks to the efficient algorithm for data processing, the processing speed of face recognition is also increased, thereby ensuring efficient passage in crowded places such as public transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0034] Figure 1 It is a system block diagram of an intelligent turnstile system based on image recognition technology according to an embodiment of the present application.

[0035] Figure 2 It is a schematic diagram of data flow of an intelligent turnstile system based on image recognition technology according to an embodiment of the present application.

[0036] Figure 3 It is a block diagram of an age information - face reference feature joint encoding module in an intelligent turnstile system based on image recognition technology according to an embodiment of the present application.

[0037] Figure 4 It is a block diagram of an authentication result generation module in an intelligent turnstile system based on image recognition technology according to an embodiment of the present application.

[0038] Figure 5 It is a flowchart of an intelligent turnstile method based on image recognition technology according to an embodiment of the present application. Detailed implementation manners

[0039] Next, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described herein.

[0040] As an automated access control device, turnstiles are widely used in various places that need to control the flow of people in and out, such as subway stations, railway stations, airports, office buildings, residential communities, and scenic spots. With the rapid progress of technology, especially in the fields of artificial intelligence and big data, safe and convenient dynamic management of people flow has become increasingly crucial, and the demand for efficient and intelligent access control turnstile systems in modern society is increasing day by day.

[0041] However, traditional face recognition technology has limitations in many aspects. For example, it does not fully consider the impact of age changes on facial features. Specifically, as age increases, facial features will change significantly, such as the increase in wrinkles, skin color changes, and changes in facial contours, which makes it difficult for traditional algorithms to accurately recognize faces after a long time. In addition, from the perspective of real-time performance, traditional algorithms usually have complex calculations and slow processing speeds. Especially in crowded places, recognition delays are very likely to have an adverse impact on the passing efficiency. Such limitations are particularly prominent in scenarios such as public transportation that require rapid identity verification, which may cause queuing and congestion, thus bringing a bad experience to users.

[0042] Therefore, to address the above technical problems, the present application proposes an intelligent turnstile system based on image recognition technology. It obtains a face reference image and age information from an ID card recognition image, collects a face detection image through a camera, and uses image recognition technology and information processing algorithms based on deep learning to perform face feature analysis on the face reference image and the face detection image, and encodes the age information. Based on this, a face comparison feature obtained by calculating the difference between the face reference correction feature and the face detection feature with the assistance of the age information is used to intelligently determine whether the identities match, and to control the turnstile opening instruction. In this way, by utilizing the age information on the ID card, the system can predict and adapt to time-varying facial features, such as wrinkles and facial contour changes, thereby improving the recognition accuracy across different ages. At the same time, an efficient deep learning model and optimized algorithms are adopted to precisely compare the face reference features and the detection features, improving the processing speed and accuracy of face recognition, and thus ensuring efficient passage in crowded places such as public transportation.

[0043] Figure 1 FIG. is a system block diagram of an intelligent turnstile system based on image recognition technology according to an embodiment of the present application. Figure 2 FIG. is a schematic diagram of data flow of an intelligent turnstile system based on image recognition technology according to an embodiment of the present application. As Figure 1 and Figure 2 shown, in the intelligent turnstile system 100 based on image recognition technology, it includes: a face reference image age information acquisition module 110 for obtaining a face reference image and age information from an ID card recognition image; a face reference feature extraction module 120 for inputting the face reference image into a face feature extractor to obtain a face reference feature map; an age information one-hot encoding module 130 for performing one-hot encoding on the age information to obtain an age information one-hot encoding vector; an age information-face reference feature joint encoding module 140 for inputting the age information one-hot encoding vector and the face reference feature map into a heterogeneous Transformer cross-domain joint encoder based on modality assistance to obtain an age-assisted face reference correction feature map; a face detection image acquisition module 150 for acquiring a face detection image collected by a camera; a face detection feature extraction module 160 for inputting the face detection image into the face feature extractor to obtain a face detection feature map; a face feature difference calculation module 170 for calculating a difference feature map between the face detection feature map and the age-assisted face reference correction feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature; an identity verification result generation module 180 for obtaining an identity verification result based on the face comparison high-dimensional feature, and determining whether to generate a turnstile opening instruction based on the identity verification result.

[0044] In the embodiment of the present application, the face reference image age information acquisition module 110 is used to obtain a face reference image and age information from the ID card recognition image. It should be understood that the face reference image information in the ID card recognition image is the core data for the turnstile to recognize an individual's identity. It provides a standard facial feature image for comparison with the face image captured in real time, so as to perform identity verification. The age information in the ID card image is used to adjust the face reference image to adapt to the influence of the change of facial features with age. As people age, their facial features change (such as wrinkles, skin color changes, etc.), which may lead to a decrease in recognition accuracy if only relying on static reference images. By combining age information with face information in the present application, an age-assisted corrected feature map can be generated, thereby improving the accuracy of the turnstile in identifying individuals of different age groups. In particular, in a specific implementable manner of the embodiment of the present application, face detection algorithms in libraries such as OpenCV or Dlib can be used to identify and locate the face in the ID card image and extract the corresponding face reference image data; and the age information on the ID card can be extracted using optical character recognition technology.

[0045] In the embodiment of the present application, the face reference feature extraction module 120 is used to input the face reference image into a face feature extractor to obtain a face reference feature map. Specifically, in the embodiment of the present application, the face reference feature extraction module is used to: input the face reference image into a face feature extractor based on a dilated convolutional neural network model to obtain the face reference feature map. It should be understood that the face reference image contains key face feature information at different scales. Dilated convolution has good advantages in image processing and can extract face features at different scales to enhance the recognition ability of faces with different resolutions and sizes. Therefore, in the technical solution of the present application, the face reference image is input into a face feature extractor based on a dilated convolutional neural network model to capture and mine rich face feature detail information to obtain a face reference feature map. The face feature extractor based on the dilated convolutional neural network model here is the dilated convolutional neural network model.

[0046] In the embodiment of the present application, the one-hot encoding module 130 for age information is configured to perform one-hot encoding on the age information to obtain a one-hot encoded vector of age information. Correspondingly, considering that age is usually a continuous variable, if it is directly input as a numerical value, the model may misinterpret it as having an ordinal relationship. Therefore, in order to convert age information into the input numerical data required by machine learning algorithms to better understand and analyze age information, in the technical solution of the present application, one-hot encoding is performed on the age information to obtain a one-hot encoded vector of age information. One-hot encoding is a coding method for converting categorical variables into a numerical format. Its main idea is to represent each category (such as the category of year, month, and day on the ID card) as a binary vector, where only one position is 1 and the rest are 0. By performing one-hot encoding on age information, the model can better understand and process these features, thereby improving the accuracy and efficiency of subsequent identity judgment.

[0047] In the embodiment of the present application, the age information-face reference feature joint encoding module 140 is configured to input the one-hot encoded vector of age information and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map. Correspondingly, considering the mutual correlation between the one-hot encoded vector of age information and the face reference feature map, for example, some facial features of the face may be related to age. Based on this, in order to be able to more accurately analyze and understand the interaction and influence relationship between age information and face reference features, in the technical solution of the present application, the one-hot encoded vector of age information is used as modality-assisted information to enhance the face feature representation, that is, age information is used as additional context information to enrich the face features. Specifically, the one-hot encoded vector of age information and the face reference feature map are input into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map.

[0048] It is worth mentioning that the heterogeneous Transformer cross-domain joint encoder based on modal assistance is a processing technology for cross-modal data. Through fine-grained feature processing and the self-attention mechanism of the Transformer structure, it realizes the effective fusion and optimization of different modal features. Specifically, first, a linear transformation and feature shape reshaping are performed on the one-hot encoded vector of the age information to generate a one-hot encoded modulation matrix of the age information as auxiliary information for the face features. At the same time, a linear transformation and shape reshaping are performed on the face reference feature map, and the most critical information of the face features is extracted therefrom to obtain the first linearly transformed face reference feature matrix and the second linearly transformed face reference feature matrix. Subsequently, the first linearly transformed face reference feature matrix and the second linearly transformed face reference feature matrix are used as the key matrix and the value matrix respectively, and the one-hot encoded modulation matrix of the age information is used as the query matrix and input into the Transformer structure to achieve cross-modal dynamic interaction and mutual correlation between the age information and the face feature information. Then, shape reshaping and point convolution encoding are performed on the one-hot encoded attention query face reference matrix of the age information obtained after the interaction to further capture and extract the key features in the features, generating a one-hot encoded attention query face reference feature map of the age information. Finally, by calculating the element-wise multiplication between the one-hot encoded attention query face reference feature map of the age information and the face reference feature map, the cross-modal features between the age information and the face features and the information of the original face reference feature map are integrated to generate the final age-assisted face reference corrected feature map.

[0049] Specifically, Figure 3 is a block diagram of the age information-face reference feature joint encoding module in the intelligent turnstile system based on the picture recognition technology according to the embodiment of the present application. As Figure 3As shown, the age information-face reference feature joint encoding module 140 includes: an age information feature linear transformation shape reshaping unit 141, configured to perform linear transformation and feature shape reshaping on the age information one-hot encoding vector to obtain an age information one-hot encoding modulation matrix; a face reference feature linear transformation unit 142, configured to perform a first linear transformation and a second linear transformation on the face reference feature map to obtain a first linearly transformed face reference feature map and a second linearly transformed face reference feature map; a face reference feature linear transformation reshaping unit 143, configured to perform feature shape reshaping on the first linearly transformed face reference feature map and the second linearly transformed face reference feature map to obtain a first linearly transformed face reference feature matrix and a second linearly transformed face reference feature matrix; an age information one-hot encoding attention calculation unit 144, configured to use the first linearly transformed face reference feature matrix and the second linearly transformed face reference feature matrix as the key matrix and the value matrix, and use the age information one-hot encoding modulation matrix as the query matrix, and input the query matrix, the key matrix, and the value matrix into the Transformer structure to obtain an age information one-hot encoding attention query face reference matrix; an age information feature encoding reshaping unit 145, configured to perform shape reshaping and point convolution encoding on the age information one-hot encoding attention query face reference matrix to obtain an age information one-hot encoding attention query face reference feature map; an age-assisted face feature correction unit 146, configured to perform element-wise multiplication of the age information one-hot encoding attention query face reference feature map and the face reference feature map to obtain the age-assisted face reference correction feature map.

[0050] More specifically, in the embodiment of the present application, the age information feature linear transformation shape reshaping unit is configured to: multiply the age information one-hot encoding vector and the age information modulation matrix to obtain an age information modulation vector, and then perform element-wise addition of the age information modulation vector and the age information bias vector to obtain an age information one-hot encoding modulation bias vector; input the age information one-hot encoding modulation bias vector into the Sigmoid function for activation processing to obtain an age information one-hot encoding modulation activation vector; perform shape reshaping on the age information one-hot encoding modulation activation vector to obtain the age information one-hot encoding modulation matrix.

[0051] More specifically, in the embodiments of the present application, the face reference feature linear transformation unit is configured to: perform element-wise multiplication of each feature matrix along the channel dimension in the face reference feature map with the first face reference modulation matrix and then perform element-wise addition with the first face reference bias matrix to obtain a first face reference bias feature map; input the first face reference bias feature map into the Sigmoid function for activation processing to obtain the first linearly transformed face reference feature map; perform element-wise multiplication of each feature matrix along the channel dimension in the face reference feature map with the second face reference modulation matrix and then perform element-wise addition with the second face reference bias matrix to obtain a second face reference bias feature map; input the second face reference bias feature map into the Sigmoid function for activation processing to obtain the second linearly transformed face reference feature map.

[0052] More specifically, in the embodiments of the present application, the age information one-hot encoded attention calculation unit is configured to: calculate the matrix product between the transposed matrix of the age information one-hot encoded modulation matrix and the first linearly transformed face reference feature matrix and then input it into the softmax function to obtain an age information-face reference feature association matrix; use the age information-face reference feature association matrix as a weight matrix and calculate the matrix product between the weight matrix and the second linearly transformed face reference feature matrix to obtain the age information one-hot encoded attention query face reference matrix.

[0053] In the embodiments of the present application, specifically, the age information-face reference feature joint encoding module is configured to: input the age information one-hot encoded vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder and process them according to the following cross-domain joint encoding formula to obtain the age-assisted face reference corrected feature map; where the cross-domain joint encoding formula is: ;

[0054] Where, is the age information one-hot encoded vector, is the age information modulation matrix, is the age information bias vector, is the Sigmoid function, is the age information one-hot encoded modulation activation vector, is the shape reshaping operation, is the age information one-hot encoded modulation matrix, is the face reference feature map, and are the first face reference modulation matrix and the second face reference modulation matrix respectively, and are the first face reference offset matrix and the second face reference offset matrix respectively, and are the first linearly transformed face reference feature map and the second linearly transformed face reference feature map respectively, and are the first linearly transformed face reference feature matrix and the second linearly transformed face reference feature matrix respectively, is the transpose matrix of the age information one-hot encoding modulation matrix, is a function, is the age information-face reference feature correlation matrix, is the matrix product, is the age information one-hot encoding attention query face reference matrix, is the point convolution operation, is the age information one-hot encoding attention query face reference feature map, is the element-wise multiplication, is the age-assisted face reference correction feature map.

[0055] In the embodiment of the present application, the face detection image acquisition module 150 is used to obtain a face detection image collected by a camera. It should be understood that the face detection image collected by the camera is the basic data for identity verification. By capturing the user's facial features in real time, it can be compared with the stored identity information to confirm the user's identity.

[0056] In an embodiment of the present application, the face detection feature extraction module 160 is used to input the face detection image into the face feature extractor to obtain a face detection feature map. Specifically, in an embodiment of the present application, the face detection feature extraction module is used to: input the face detection image into the face feature extractor based on the dilated convolutional neural network model to obtain the face detection feature map. Accordingly, considering that the face detection image reflects the key feature information of face detection, therefore, in the technical solution of the present application, the face detection image is input into the face feature extractor based on the dilated convolutional neural network model to capture and extract richer and more critical face feature information to obtain a face detection feature map. It should be understood that the dilated convolutional neural network model processes the input face detection image by using a special dilated convolution kernel, thereby achieving the extraction of facial features. That is, the atrous convolution expands the receptive field by inserting "holes" in the convolution kernel, which enables the model to capture a wider range of contextual information with fewer parameters. As a result, the atrous convolutional neural network can still provide excellent performance when processing complex scenes, such as subtle changes in user expressions or in unfavorable lighting conditions, and can more accurately recognize and understand the user's facial features.

[0057] In an embodiment of the present application, the facial feature differential calculation module 170 is used to calculate the differential feature map between the face detection feature map and the age-assisted face reference correction feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature. It should be understood that in order to more carefully and accurately display the difference between the face detection feature map and the age-assisted face reference correction feature map, so as to show the subtle changes in the face, in the technical solution of the present application, calculating the differential feature map between the face detection feature map and the age-assisted face reference correction feature map to obtain a face comparison high-dimensional feature map helps to identify whether the identity matches. That is, by focusing on the differential features between the two, the system can more sensitively identify the unique and subtle differences between different individuals, thereby improving the accuracy of recognition.

[0058] In the embodiment of the present application, the identity verification result generation module 180 is used to obtain the identity verification result based on the face comparison high-dimensional feature, and determine whether to generate a gate opening instruction based on the identity verification result. Specifically, Figure 4 FIG. 1 is a block diagram of an identity authentication result generation module in an intelligent gate system based on image recognition technology according to an embodiment of the present application. Figure 4As shown, the authentication result generation module 180 includes: a face feature parsing unit 181, which is used to input the face comparison high-dimensional feature map into an authenticator based on a classifier to obtain the authentication result, and the authentication result is used to indicate whether the identities match; a gate opening instruction generation unit 182, which is used to generate a gate opening instruction in response to the authentication result being an identity match. That is, classification processing is performed using the face comparison high-dimensional feature obtained by performing differential calculation between the face detection feature map and the age-assisted face reference correction feature map, so as to intelligently determine whether the identities match and control the gate opening instruction. In this way, by using the age information on the ID card, the system can predict and adapt to the time-varying facial features of a person, such as wrinkles and changes in facial contours, thereby improving the recognition accuracy across different ages. At the same time, an efficient deep learning model and optimized algorithms are used to accurately compare the face reference features and detection features, improving the processing speed and accuracy of face recognition, and thus ensuring efficient passage in crowded places such as public transportation.

[0059] Specifically, in a preferred example, when the face reference feature map and the face detection feature map respectively represent the face image semantic features of a face reference image and a face detection image, after the modality-assisted heterogeneous Transformer cross-domain joint encoder performs channel constraint on the face reference feature map based on age information encoding features, the face comparison high-dimensional feature map, which is the differential feature map between the face detection feature map and the age-assisted face reference correction feature map, has a distribution regression pattern offset caused by the semantic association alignment offset in the channel dimension. That is, it causes an overflow of the authentication result obtained by inputting the face comparison high-dimensional feature map into the authenticator based on the classifier, affecting its accuracy.

[0060] Based on this, inputting the face comparison high-dimensional feature map into the classifier-based authenticator to obtain an authentication result specifically includes: determining a matching probability value corresponding to identity matching obtained by inputting the face comparison high-dimensional feature map into the classifier-based authenticator, and passing the face comparison high-dimensional feature map through a probabilistic activation function to obtain a probabilistic face comparison high-dimensional feature map; subtracting the probabilistic face comparison high-dimensional feature map from the unit feature map point by point, and using the result feature map of the point subtraction as the base to calculate a power function with the matching probability value as the exponent; multiplying the power function feature map and the probabilistic face comparison high-dimensional feature map point by point to obtain a face comparison high-dimensional intermediate feature map; using the absolute value of each eigenvalue of the probabilistic face comparison high-dimensional feature map as the base, calculating a power function with the reciprocal of the matching probability value as the exponent, and summing all the eigenvalues to obtain a face comparison high-dimensional bias eigenvalue; multiplying the face comparison high-dimensional bias eigenvalue by a weight as a hyperparameter, and adding it to the face comparison high-dimensional intermediate feature map point by point to obtain an optimized face comparison high-dimensional feature map; inputting the optimized face comparison high-dimensional feature map into the classifier-based authenticator to obtain an authentication result.

[0061] Among them, let the probabilistic face comparison high-dimensional feature map be denoted as Then the optimized face comparison high-dimensional feature map is expressed as: ; where is the probabilistic face comparison high-dimensional feature map, is a unit feature map with all eigenvalues being one, and , is the eigenvalue at each position of the probabilistic face comparison high-dimensional feature map , is dot multiplication by position, is subtraction by position, is the matching probability value, is to calculate the power function with the matching probability value as the exponent for the eigenvalues at each position in the feature map, is addition by position, is the weight as a hyperparameter, is the optimized face comparison high-dimensional feature map.

[0062] ​That is, for the high-dimensional face comparison feature map, based on the convergence limit of the Cauchy-Hadamard formal power series probability distribution, the feature set of the high-dimensional face comparison feature map is represented by a formal power series based on the convergence probability, and a set convergence bounded bias represented as the convergence radius is added for each eigenvalue of the high-dimensional face comparison feature map, so as to avoid the invalid overflow of the classification result caused by the deviation of the distribution pattern of the high-dimensional face comparison feature map during the regression process of the classifier-based identity verifier under the unified probability convergence limit standard, thereby improving the accuracy of the identity verification result obtained by the classifier-based identity verifier. In this way, by utilizing the age information on the ID card, the system can predict and adapt to the time-varying facial features, such as wrinkles and facial contour changes, thus improving the cross-age recognition accuracy. At the same time, an efficient deep learning model and an optimized algorithm are adopted to accurately compare the face reference features and the detected features, improving the processing speed and accuracy of face recognition, thereby ensuring efficient passage in crowded places such as public transportation.

[0063] In summary, the intelligent turnstile system 100 based on the picture recognition technology according to the embodiments of the present application is elucidated. It obtains the face reference image and age information from the ID card recognition image, collects the face detection image by the camera, and uses a data processing algorithm based on deep learning to perform face feature analysis on the face reference image and the face detection image, and encodes the age information, so as to intelligently determine whether the identities match based on the face comparison features obtained by calculating the difference between the face reference correction features and the face detection features assisted by the age information, and control the gate opening instruction. In this way, the accuracy of cross-age identity recognition can be significantly improved. At the same time, thanks to the efficient algorithm for data processing, the processing speed of face recognition is also increased, thereby ensuring efficient passage in crowded places such as public transportation.

[0064] As described above, the intelligent turnstile system 100 based on the picture recognition technology according to the embodiments of the present application can be implemented in various terminal devices, such as a server for an intelligent turnstile based on the picture recognition technology. In one example, the intelligent turnstile system 100 based on the picture recognition technology according to the embodiments of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the intelligent turnstile system 100 based on the picture recognition technology can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the intelligent turnstile system 100 based on the picture recognition technology can also be one of the many hardware modules of the terminal device.

[0065] Alternatively, in another example, the intelligent turnstile system 100 based on the picture recognition technology and the terminal device can also be separate devices, and the intelligent turnstile system 100 based on the picture recognition technology can be connected to the terminal device through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0066] Figure 5 FIG. is a flowchart of an intelligent turnstile method based on picture recognition technology according to an embodiment of the present application. As Figure 5 shown, in the intelligent turnstile method based on picture recognition technology, it includes: S110, obtaining a face reference image and age information from an ID card recognition image; S120, inputting the face reference image into a face feature extractor to obtain a face reference feature map; S130, performing one-hot encoding on the age information to obtain an age information one-hot encoding vector; S140, inputting the age information one-hot encoding vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map; S150, obtaining a face detection image collected by a camera; S160, inputting the face detection image into the face feature extractor to obtain a face detection feature map; S170, calculating a differential feature map between the face detection feature map and the age-assisted face reference corrected feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature; S180, obtaining an identity verification result based on the face comparison high-dimensional feature, and judging whether to generate a gate opening instruction based on the identity verification result.

[0067] Here, those skilled in the art can understand that the specific operations of each step in the above intelligent turnstile method based on picture recognition technology have been introduced in detail in the description of the intelligent turnstile system based on picture recognition technology above, and therefore, its repeated description will be omitted. Figures 1 to 4 The description of the intelligent turnstile system based on picture recognition technology above, and therefore, its repeated description will be omitted.

[0068] In summary, the intelligent turnstile method based on the embodiment of the present application is clarified. It obtains a face reference image and age information from an ID card recognition image, collects a face detection image by a camera, and uses a data processing algorithm based on deep learning to perform face feature analysis on the face reference image and the face detection image, and perform information encoding on the age information. Thus, it intelligently judges whether the identities match based on the face reference corrected feature obtained with the assistance of age information and the face comparison feature calculated by taking the difference between the face detection features, and controls the gate opening instruction. In this way, the accuracy of cross-age identity recognition can be significantly improved. At the same time, due to the efficient algorithm for data processing, the processing speed of face recognition is also increased, thereby ensuring efficient passage in crowded places such as public transportation.

[0069] It should be noted that the information involved in this application (including but not limited to user device information, user personal information, etc.), data (including but not limited to data for analysis, stored data, displayed data, etc.), and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards in relevant countries and regions. For example, the relevant information or data such as "face", "face recognition", "facial" involved in this disclosure are all obtained under full authorization.

[0070] The basic principles of this application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in this application are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of this application. In addition, the specific details of the above application are only for the purpose of illustration and facilitating understanding, rather than limitations. The above details do not limit this application to necessarily adopt the above specific details for implementation.

Claims

1. An intelligent gate system based on image recognition technology, characterized in that: include: A face reference image age information acquisition module is used to obtain a face reference image and age information from an ID card recognition image; A face reference feature extraction module, used for inputting the face reference image into a face feature extractor to obtain a face reference feature map; An age information one-hot encoding module, used for performing one-hot encoding on the age information to obtain an age information one-hot encoding vector; An age information-face reference feature joint encoding module, used for inputting the age information one-hot encoding vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map; A face detection image acquisition module is used to obtain a face detection image acquired by a camera; A face detection feature extraction module, used for inputting the face detection image into the face feature extractor to obtain a face detection feature map; A face feature difference calculation module, used for calculating a difference feature map between the face detection feature map and the age-assisted face reference correction feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature; An identity authentication result generating module, used to obtain an identity authentication result based on the face comparison high-dimensional features, and based on the identity authentication result, determine whether to generate a gate opening instruction; Wherein, the age information-face reference feature joint encoding module further comprises: An age information feature linear transformation and reshaping unit is used to perform linear transformation and feature reshaping on the age information one-hot encoding vector to obtain an age information one-hot encoding modulation matrix; A facial reference feature linear transformation unit, configured to perform a first linear transformation and a second linear transformation on the facial reference feature map to obtain a first linear transformation facial reference feature map and a second linear transformation facial reference feature map; A face reference feature linear transformation reshaping unit, used for reshaping the first linear transformation face reference feature map and the second linear transformation face reference feature map to obtain a first linear transformation face reference feature matrix and a second linear transformation face reference feature matrix; an age information one-hot encoding attention calculation unit, configured to use the first linear transformation face reference feature matrix and the second linear transformation face reference feature matrix as a key matrix and a value matrix and use the age information one-hot encoding modulation matrix as a query matrix, input the query matrix, the key matrix and the value matrix into a Transformer structure to obtain an age information one-hot encoding attention query face reference matrix; An age information feature encoding and reshaping unit is used to reshape and perform point convolution encoding on the age information one-hot encoding attention query face reference matrix to obtain an age information one-hot encoding attention query face reference feature map; The age-assisted facial feature correction unit is used to perform position point multiplication on the age information one-hot encoding attention query face reference feature map and the face reference feature map to obtain the age-assisted face reference correction feature map.

2. The intelligent gate system based on image recognition technology according to claim 1 is characterized in that: The face reference feature extraction module is specifically used to: input the face reference image into a face feature extractor based on a hole convolutional neural network model to obtain the face reference feature map.

3. The intelligent gate system based on image recognition technology according to claim 2 is characterized in that: The age information feature linear transformation shape reshaping unit is specifically used for: After performing matrix multiplication of the age information one-hot encoding vector and the age information modulation matrix to obtain the age information modulation vector, the age information modulation vector and the age information bias vector are added according to position to obtain the age information one-hot encoding modulation bias vector; Inputting the age information one-hot coded modulation bias vector into a Sigmoid function for activation processing to obtain an age information one-hot coded modulation activation vector; The age information one-hot coded modulation activation vector is reshaped to obtain the age information one-hot coded modulation matrix.

4. The intelligent gate system based on image recognition technology according to claim 3 is characterized in that: The face reference feature linear transformation unit is specifically used for: Each feature matrix along the channel dimension in the face reference feature map is multiplied by position with the first face reference modulation matrix, and then added by position with the first face reference bias matrix to obtain a first face reference bias feature map; Inputting the first face reference bias feature map into the Sigmoid function for activation processing to obtain the first linear transformation face reference feature map; Multiplying each feature matrix along the channel dimension in the face reference feature map by position with the second face reference modulation matrix, and then adding the matrix by position with the second face reference bias matrix to obtain a second face reference bias feature map; The second face reference offset feature map is input into the Sigmoid function for activation processing to obtain the second linear transformation face reference feature map.

5. The intelligent gate system based on image recognition technology according to claim 4 is characterized in that: The age information unique hot encoding attention calculation unit is specifically used for: Calculating the matrix product between the transposed matrix of the age information one-hot encoding modulation matrix and the first linear transformation face reference feature matrix and inputting the matrix product into a softmax function to obtain an age information-face reference feature association matrix; The age information-face reference feature association matrix is ​​used as a weight matrix, and the matrix product between the weight matrix and the second linear transformation face reference feature matrix is ​​calculated to obtain the age information one-hot encoding attention query face reference matrix.

6. The intelligent gate system based on image recognition technology according to claim 5 is characterized in that: The face detection feature extraction module is specifically used to: input the face detection image into the face feature extractor based on the hollow convolutional neural network model to obtain the face detection feature map.

7. The intelligent gate system based on image recognition technology according to claim 6 is characterized in that: The identity authentication result generating module further comprises: A face feature analysis unit, used for inputting the face comparison high-dimensional features into a classifier-based identity verification device to obtain the identity verification result, wherein the identity verification result is used to indicate whether the identity matches; The gate opening instruction generating unit is used to generate a gate opening instruction in response to the identity authentication result being an identity match.

8. A smart gate method based on image recognition technology, using the smart gate system based on image recognition technology according to claim 1, characterized in that: include: Obtain a face reference image and age information from the ID card recognition image; Inputting the face reference image into a face feature extractor to obtain a face reference feature map; One-hot encoding the age information to obtain a one-hot encoding vector of the age information; Inputting the age information one-hot encoding vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map; Obtaining a face detection image captured by a camera; Inputting the face detection image into the face feature extractor to obtain a face detection feature map; Calculating a differential feature map between the face detection feature map and the age-assisted face reference correction feature map to obtain a face comparison high-dimensional feature map as a face comparison high-dimensional feature; Based on the face comparison high-dimensional features, obtaining an identity verification result, and based on the identity verification result, determining whether to generate a gate opening instruction; The step of inputting the age information one-hot encoding vector and the face reference feature map into a modality-assisted heterogeneous Transformer cross-domain joint encoder to obtain an age-assisted face reference corrected feature map further comprises: Performing linear transformation and feature shape reshaping on the age information one-hot encoding vector to obtain an age information one-hot encoding modulation matrix; Performing a first linear transformation and a second linear transformation on the face reference feature map to obtain a first linear transformation face reference feature map and a second linear transformation face reference feature map; Reshaping the first linear transformation human face reference feature map and the second linear transformation human face reference feature map to obtain a first linear transformation human face reference feature matrix and a second linear transformation human face reference feature matrix; Taking the first linear transformation face reference feature matrix and the second linear transformation face reference feature matrix as a key matrix and a value matrix and taking the age information one-hot encoding modulation matrix as a query matrix, inputting the query matrix, the key matrix and the value matrix into a Transformer structure to obtain an age information one-hot encoding attention query face reference matrix; Reshaping and point convolution encoding the age information one-hot encoding attention query face reference matrix to obtain an age information one-hot encoding attention query face reference feature map; The age information one-hot encoded attention query face reference feature map is point-multiplied by the face reference feature map to obtain the age-assisted face reference correction feature map.

9. The intelligent gate method based on image recognition technology according to claim 8 is characterized in that: Inputting the face reference image into a face feature extractor to obtain a face reference feature map, including: inputting the face reference image into a face feature extractor based on a hole convolutional neural network model to obtain the face reference feature map.

Images