A light rail swiping face recognition method and system based on super-resolution enhancement

Abstract

The application discloses a light rail theft face recognition method and system based on super-resolution enhancement, and the method comprises the following steps: data loading and linkage matching processing: loading transaction data and corresponding standard photos, and linking and matching the transaction data with the monitoring video; key frame extraction: extracting video frames from the matched monitoring video, and then acquiring key frames according to a fixed sampling rate; face detection and intelligent screening: adopting a preset face recognition model to perform multi-target face detection on the key frames, calculating the confidence, and then selecting the optimal face area according to the confidence; directional super-resolution enhancement: performing super-resolution reconstruction on the optimal face area; feature comparison and theft judgment: extracting the face feature vector after the super-resolution enhancement and the feature vector of the corresponding standard photo, calculating the similarity between the two, and judging whether the light rail theft behavior exists or not through a preset threshold; the beneficial effects are that the application can improve the recognition accuracy and reduce the waste of computing resources.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and artificial intelligence, specifically to a method and system for detecting light rail fraud using super-resolution enhancement technology. This method improves the accuracy of facial recognition in low-quality surveillance videos by using super-resolution enhancement technology, thereby enabling the detection of light rail fraud. Background Technology

[0002] Currently, with the rapid development of urban rail transit, light rail has become an important public transportation tool for people's daily travel, and the application of social security cards as a payment credential for travel is becoming increasingly widespread. However, this has also led to frequent cases of light rail card fraud, where unauthorized individuals use other people's social security cards to travel, which not only infringes on the legitimate rights and interests of cardholders but also causes problems for light rail operation and management. To curb such behavior, the industry has gradually introduced facial recognition technology, which compares the faces of people monitored during travel with the faces of social security card holders to detect and issue warnings for fraudulent transactions.

[0003] In existing technologies, facial recognition technology has been applied to some extent in fields such as security and transportation, and related solutions are constantly being optimized. For example, the low-resolution facial recognition method proposed by China Telecom Digital Life (publication number CN119399814A) and the image quality enhancement method based on facial recognition proposed by Zhongshitong (publication number CN119919302A) both solve the problem of facial recognition under low-quality images through technical means. The core idea of ​​these methods is to enhance the low-quality image and then use a conventional facial recognition model for feature extraction and comparison. Among them, super-resolution enhancement technology is one of the key means to improve the clarity of low-quality images. Through this technology, image details can be restored, providing higher-quality image data for subsequent recognition.

[0004] However, existing technologies have many compatibility issues when applied to light rail fraud detection scenarios, as follows:

[0005] 1. Inherent deficiencies in the quality of surveillance videos: The surveillance cameras at light rail stations are installed in fixed positions. Due to factors such as long shooting distance, complex lighting conditions (such as backlighting in the morning and evening, light reflection) and blurring caused by passenger movement, the facial images captured in the videos generally have problems such as low resolution, loss of details and blurriness. This is the core reason for the low accuracy of facial recognition.

[0006] 2. It is difficult to balance the recognition accuracy and false alarm rate: Low-quality face images will result in incomplete feature extraction, leading to a lower face similarity calculation result, making it difficult to accurately determine whether they are the same person; if the recognition threshold is lowered to reduce the false alarm rate, the false alarm rate will increase significantly, affecting operational efficiency.

[0007] 3. Serious waste of computing resources: Existing methods often perform super-resolution processing on all extracted video frames or detected faces. However, super-resolution algorithms have a large computational load, and blind processing will consume a lot of computing resources, resulting in low overall processing efficiency and failing to meet the real-time processing needs of massive transaction data and surveillance videos at light rail stations.

[0008] 4. Poor scenario adaptability: Existing technologies are mostly designed for general face recognition scenarios and have not been optimized to meet the specific needs of light rail fraud detection. They cannot achieve accurate linkage and matching between transaction data and surveillance video, nor have they customized processing strategies for the image characteristics of light rail surveillance. Summary of the Invention

[0009] In view of the technical defects mentioned in the background art, the purpose of this invention is to provide a method and system for facial recognition of light rail fraud based on super-resolution enhancement, which aims to at least solve one of the technical problems in the related art to a certain extent.

[0010] To achieve the above objectives, in a first aspect, the present invention provides a method for facial recognition in light rail fraud prevention based on super-resolution enhancement, the method comprising the following steps:

[0011] (1) Data loading and linkage matching processing: Load social security card transaction data and the standard photos of the corresponding social security card holders, and link and match the transaction data with the monitoring videos of the corresponding sites and transaction periods;

[0012] (2) Keyframe extraction: Extract video frames of a preset duration before and after the transaction time from the matched surveillance video, then obtain keyframes at a fixed sampling rate and perform uniform scaling on the keyframes.

[0013] (3) Face detection and intelligent screening: The key frame is multi-target face detection is performed using a preset face recognition model. At the same time, the detection confidence of each face is calculated, and the optimal face region is selected according to the confidence.

[0014] (4) Targeted super-resolution enhancement: The selected optimal face region is reconstructed using a face restoration converter algorithm;

[0015] (5) Feature comparison and theft determination: The face recognition model is used again to extract the super-resolution enhanced face feature vector and the feature vector of the corresponding standard photo, and the similarity between the two is calculated. The existence of light rail theft is determined by a preset threshold.

[0016] As a specific implementation of this application, the arc surface loss function face model in the deep face framework is used as the face recognition model to perform face detection on the key frame; the model accurately identifies all possible face regions in the key frame and outputs the bounding box coordinates of each face.

[0017] As a specific implementation of this application, during intelligent screening, the face with the highest confidence level is selected as the candidate face, and a comprehensive scoring optimization is performed by combining the face pixel size and the features of the center position of the image.

[0018] As one specific implementation of this application, when selecting the optimal face region...

[0019] Based on the bounding box coordinates of the candidate's face, the bounding box of the face is expanded by a preset number of pixels in all directions (top, bottom, left, and right), and the cropped face image is saved in a preset format for subsequent super-resolution processing.

[0020] As a specific implementation of this application, if it is determined that there is a case of fraudulent use of light rail credit, an alarm mechanism is triggered for the case of fraudulent use based on the business needs of light rail operation.

[0021] Secondly, the present invention also provides a facial recognition system for light rail card skimming based on super-resolution enhancement, the system comprising:

[0022] The data loading module serves as the system's data input source, responsible for loading social security card transaction data and the corresponding standard photos of social security card holders, and for linking and matching transaction data with monitoring videos of the corresponding sites and transaction periods.

[0023] The video processing module is used to receive video data transmitted by the data loading module, extract video frames of preset duration before and after the transaction time from the matched monitoring video, obtain key frames at a fixed sampling rate, and perform uniform scaling on the key frames.

[0024] The face recognition module is used to perform multi-target face detection on keyframes using a preset face recognition model, calculate the detection confidence of each face, and then perform intelligent filtering according to the confidence of the face to select the optimal face region.

[0025] The super-resolution enhancement module is used to perform super-resolution reconstruction on the selected optimal face region using a face restoration converter algorithm;

[0026] The result determination module is used to extract the super-resolution enhanced facial feature vector and the corresponding feature vector of the standard photo using the facial recognition model, calculate the similarity between the two, and determine whether there is a light rail fraudulent transaction by using a preset threshold; and trigger an alarm mechanism for cases determined to be fraudulent transactions.

[0027] This invention, through a collaborative design of intelligent screening and targeted super-resolution enhancement, combined with customized optimization for light rail scenarios, has significant advantages over existing technologies, as detailed below:

[0028] 1. Significantly improved facial recognition accuracy: The CodeFormer facial reconstruction and conversion algorithm is used for targeted super-resolution. This algorithm is optimized for facial features and can accurately restore the facial features and contour details of low-quality faces, solving the core problems of blurry and low-resolution faces in light rail monitoring.

[0029] 2. Significantly optimized computational efficiency and reduced computational resources: Through the intelligent face screening mechanism, super-resolution processing is only performed on the optimal face region with the highest confidence, avoiding the drawback of blindly super-resolution all frames and all faces in the existing technology. This greatly reduces the computational load of the super-resolution algorithm, improving the overall processing efficiency while ensuring accuracy. It can adapt to the real-time processing needs of massive transaction data and surveillance videos.

[0030] 3. Significantly reduced false alarm rate, balancing recognition accuracy and operational efficiency: After super-resolution processing, facial features are extracted more completely and similarity calculation is more accurate. There is no need to lower the recognition threshold to reduce the false alarm rate. A stricter threshold standard can be adopted, which effectively reduces false alarms and reduces the ineffective verification costs for light rail operators.

[0031] 4. Strong scenario adaptability and high degree of automation: Customized design for light rail fraud detection scenarios, it realizes accurate linkage and matching of transaction data and monitoring video, and can adapt to complex scenarios with multiple stations and multiple time periods; the whole process is automated, no manual intervention is required, the operation is simple, and it can be directly integrated into the existing light rail operation management system.

[0032] 5. Good feature stability and high recognition reliability: The face restoration converter algorithm (CodeFormer) can maintain the original facial features, expressions and moods during the super-resolution process, avoiding feature distortion caused by over-processing. At the same time, the ArcFace face model with arc surface loss function has strong feature extraction capabilities, further improving the reliability of recognition results. Attached Figure Description

[0033] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below.

[0034] Figure 1 This is a flowchart of a method for identifying unauthorized facial recognition on light rail trains based on super-resolution enhancement, provided in an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of the principle architecture of a light rail facial recognition system based on super-resolution enhancement provided in an embodiment of the present invention;

[0036] Figure 3 This is a comparison chart of experimental results provided in an embodiment of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0039] Super-resolution enhancement: an image processing technique that uses algorithms to restore details in low-resolution images and improve image clarity and resolution;

[0040] DeepFace is a deep learning-based face recognition framework that features face detection and feature extraction.

[0041] ArcFace: An optimized face recognition model that improves feature discrimination and recognition accuracy by introducing an arc loss function.

[0042] CodeFormer is a deep learning super-resolution and inpainting algorithm based on prior knowledge of faces. It is specifically optimized for face images, and can restore details while maintaining feature stability.

[0043] Feature vector: A high-dimensional numerical vector converted from a face image, which can be used to quantify facial features and realize similarity calculation;

[0044] Cosine distance: a metric used to measure the similarity between two feature vectors, with a value range of [0,1]. The smaller the value, the higher the similarity.

[0045] It should be noted that, unless otherwise stated, the technical terms used in this embodiment have the common meaning as understood in the relevant technical field.

[0046] The inventive concept of this invention is as follows: Based on the core ideas of data linkage, precise screening, targeted enhancement, and feature comparison, it first links and matches light rail social security card transaction data with monitoring videos of corresponding stations and time periods to extract key video frames and detect faces; then, it selects the optimal face region through intelligent screening and performs super-resolution enhancement only on this region to avoid invalid calculations; finally, it extracts the feature vectors of the enhanced optimal face and the corresponding standard photo face through a face recognition model, calculates the similarity, and determines whether there is fraudulent behavior, thus achieving dual optimization of accuracy and efficiency.

[0047] Please refer to Figure 1 This invention provides a method for facial recognition in light rail fraud cases based on super-resolution enhancement, the method comprising the following steps:

[0048] S101, Data loading and linkage matching processing: Load social security card transaction data and the standard photos of the corresponding social security card holders, and link and match the transaction data with the monitoring videos of the corresponding sites and transaction periods;

[0049] S102, Keyframe Extraction: Extract video frames of a preset duration before and after the transaction time from the matched surveillance video, then obtain keyframes at a fixed sampling rate and perform uniform scaling on the keyframes.

[0050] S103, Face Detection and Intelligent Screening: A preset face recognition model is used to perform multi-target face detection on keyframes, and the detection confidence of each face is calculated. The optimal face region is then selected by sorting the faces according to their confidence scores.

[0051] S104, Targeted super-resolution enhancement: Super-resolution reconstruction is performed on the selected optimal face region using a face restoration converter algorithm;

[0052] S105, Feature comparison and fraud determination: The face recognition model is used again to extract the super-resolution enhanced face feature vector and the feature vector of the corresponding standard photo, and the similarity between the two is calculated. A preset threshold is used to determine whether there is a light rail fraud.

[0053] In this embodiment, the transaction data includes key information such as card number, transaction date, transaction time, and exit station; a standard photo of the corresponding social security card holder is loaded and stored in the constructed reference face database, ensuring that the photo is a clear, frontal image; then, based on the date, time, and station information in the transaction record, it is precisely matched with the monitoring video files of each light rail station to locate the monitoring video of the time period in which the corresponding transaction occurred.

[0054] In step S102, video frames within a preset duration before and after the transaction time (e.g., 30 seconds before and after the transaction time) are extracted from the matched surveillance video to reduce the range of invalid frames; key frames are extracted at a fixed sampling rate (e.g., 3 frames per second) to balance the amount of data and the recognition accuracy; the extracted key frames are uniformly scaled to adjust the size to the preset specifications to balance the face recognition speed and feature preservation effect.

[0055] In step S103, the ArcFace face model with arc loss function in the DeepFace framework is used to detect faces in keyframes. This model can accurately identify all possible face regions within the frame and output the bounding box coordinates of each face (x is the horizontal coordinate of the top left corner, y is the vertical coordinate of the top left corner, w is the width, and h is the height). At the same time, the detection confidence of each face is calculated. The higher the confidence value, the greater the probability that the region is a real face.

[0056] Furthermore, during intelligent screening, the face with the highest confidence level is selected as the candidate face, and a comprehensive scoring optimization is performed by combining the face pixel size and the features of the center position of the image.

[0057] Specifically, all detected faces are sorted in descending order based on detection confidence; the face with the highest confidence is selected as the candidate face, which is the face with the best quality in the current frame and most likely to correspond to the transaction person; optionally, a comprehensive score is performed by combining features such as face size (which must meet a preset pixel size threshold) and position (prioritizing faces in the center of the image) to further optimize the accuracy of candidate face selection.

[0058] When selecting the optimal face region, the bounding box coordinates of the candidate face are used to expand the bounding box of the face by a preset number of pixels in all directions (top, bottom, left, and right), and the cropped face image is saved in a preset format for subsequent super-resolution processing.

[0059] Specifically, in order to preserve the contextual information around the face and avoid feature loss, the bounding box is appropriately expanded, such as by expanding it by a preset number of pixels in the top, bottom, left, and right.

[0060] In step S104, the CodeFormer algorithm is used to perform super-resolution reconstruction on the expanded and cropped face region. This algorithm is a deep learning super-resolution model based on prior knowledge of the face, which can selectively restore the detailed features of the face, such as facial features and contours, while keeping the original features, expressions and demeanor of the face unchanged, avoiding feature distortion caused by over-processing. The super-resolution factor can be adjusted according to the actual needs of the scene (such as 2x or 4x) to balance processing efficiency and enhancement effect.

[0061] In step S105, the ArcFace face model with the DeepFace framework and arc loss function is used to extract feature vectors from the corresponding social security card holder photos in the reference face database, as well as feature vectors from the super-resolution enhanced face image. The cosine distance between the two feature vectors is calculated; the smaller the distance value, the higher the similarity between the two faces. The distance value is converted into an intuitive similarity score using the formula: Similarity = 1 - Cosine Distance.

[0062] A preset similarity threshold (e.g., 0.32) is set, and the calculated similarity score is compared with this threshold. If the score exceeds the threshold, it is determined that the person is the same, and there is no fraudulent activity. If the score is lower than the threshold, it is determined that the person is not the same, and there is suspicion of fraudulent activity. The specific identification result is output, including the similarity score, and an alarm mechanism is triggered for cases that are determined to be fraudulent activities, based on the business needs of light rail operation.

[0063] The above solution, through the collaborative design of intelligent screening and targeted super-resolution enhancement, combined with customized optimization for light rail scenarios, has significant advantages over existing technologies, as detailed below:

[0064] 1. Significantly improved facial recognition accuracy: The CodeFormer facial reconstruction and conversion algorithm is used for targeted super-resolution. This algorithm is optimized for facial features and can accurately restore the facial features and contour details of low-quality faces, solving the core problems of blurry and low-resolution faces in light rail monitoring.

[0065] 2. Significantly optimized computational efficiency and reduced computational resources: Through the intelligent face screening mechanism, super-resolution processing is only performed on the optimal face region with the highest confidence, avoiding the drawback of blindly super-resolution all frames and all faces in the existing technology. This greatly reduces the computational load of the super-resolution algorithm, improving the overall processing efficiency while ensuring accuracy. It can adapt to the real-time processing needs of massive transaction data and surveillance videos.

[0066] 3. Significantly reduced false alarm rate, balancing recognition accuracy and operational efficiency: After super-resolution processing, facial features are extracted more completely and similarity calculation is more accurate. There is no need to lower the recognition threshold to reduce the false alarm rate. A stricter threshold standard can be adopted, which effectively reduces false alarms and reduces the ineffective verification costs for light rail operators.

[0067] 4. Strong scenario adaptability and high degree of automation: Customized design for light rail fraud detection scenarios, it realizes accurate linkage and matching of transaction data and monitoring video, and can adapt to complex scenarios with multiple stations and multiple time periods; the whole process is automated, no manual intervention is required, the operation is simple, and it can be directly integrated into the existing light rail operation management system.

[0068] 5. Good feature stability and high recognition reliability: The face restoration converter algorithm (CodeFormer) can maintain the original facial features, expressions and moods during the super-resolution process, avoiding feature distortion caused by over-processing. At the same time, the ArcFace face model with arc surface loss function has strong feature extraction capabilities, further improving the reliability of recognition results.

[0069] Based on the same inventive concept, this invention also provides a facial recognition system for light rail card skimming based on super-resolution enhancement, referring to... Figure 2 The system includes:

[0070] The data loading module serves as the system's data input source, responsible for loading social security card transaction data and the corresponding standard photos of social security card holders, and for linking and matching transaction data with monitoring videos of the corresponding sites and transaction periods.

[0071] The video processing module is used to receive video data transmitted by the data loading module, extract video frames of preset duration before and after the transaction time from the matched monitoring video, obtain key frames at a fixed sampling rate, and perform uniform scaling on the key frames.

[0072] The face recognition module is used to perform multi-target face detection on keyframes using a preset face recognition model, calculate the detection confidence of each face, and then perform intelligent filtering according to the confidence of the face to select the optimal face region.

[0073] The super-resolution enhancement module is used to perform super-resolution reconstruction on the selected optimal face region using a face restoration converter algorithm;

[0074] The result determination module is used to extract the super-resolution enhanced facial feature vector and the corresponding feature vector of the standard photo using the facial recognition model, calculate the similarity between the two, and determine whether there is a light rail fraudulent transaction by using a preset threshold; and trigger an alarm mechanism for cases determined to be fraudulent transactions.

[0075] Furthermore, the arc surface loss function face model in the deep face framework is used as the face recognition model to perform face detection on the key frame; the model accurately identifies all possible face regions in the key frame and outputs the bounding box coordinates of each face.

[0076] During intelligent screening, the face with the highest confidence level is selected as the candidate face, and a comprehensive score is optimized by combining the face pixel size and the features of the center position of the image.

[0077] When selecting the optimal face region

[0078] Based on the bounding box coordinates of the candidate's face, the bounding box of the face is expanded by a preset number of pixels in all directions (top, bottom, left, and right), and the cropped face image is saved in a preset format for subsequent super-resolution processing.

[0079] It should be noted that for more specific details of the system implementation and the description of the working process, please refer to the aforementioned method implementation section, which will not be repeated here.

[0080] To better understand this solution, a scenario of detecting social security card fraud at a light rail station is used as an example to explain the specific implementation process of this invention in detail. The specific steps are as follows:

[0081] Data loading and linkage matching processing: The social security card transaction records are loaded through the data loading module. One target record is: card number 0310xxxxxxxxxxx691, transaction date 20xx year x month x day, transaction time 07:48:00, exit station xx station; At the same time, a clear frontal photo of the cardholder is loaded and stored in the reference face database; Based on the transaction date, time and station information, the corresponding time period of the surveillance video file of xx station is matched (the video coverage time is from 07:46:01 to 07:49:59 on 20xx year x month x day).

[0082] Keyframe extraction: The video processing module extracts video frames from the matched video between 07:47:30 and 07:48:30, extracts keyframes at a sampling rate of 3 frames per second, and obtains a total of 90 keyframes; all keyframes are scaled to a uniform size of 1920×1080 pixels and transmitted to the face recognition module.

[0083] Multi-target face detection: The face recognition module uses the ArcFace face model with the arc loss function of the DeepFace framework to detect each key frame. In one key frame, three faces are detected, with corresponding detection confidence scores of 0.85, 0.72 and 0.58, respectively. The bounding box coordinates of each face are also output.

[0084] Intelligent face selection: Sort by confidence in descending order, with 0.85 being the highest value, and select the face as the candidate target; at the same time, verify the face size (boundary box width 150 pixels, height 180 pixels) and position (located in the center area of ​​the image). If the preset conditions are met, determine it as the optimal face.

[0085] Face region cropping: Based on the bounding box coordinates of the face (x=120, y=80, w=150, h=180), the face region is cropped from the keyframe; to preserve context information, the bounding box is extended by 100 pixels in all directions to obtain a face image with a size of 350×380 pixels, which is then saved and transmitted to the super-resolution enhancement module.

[0086] Super-resolution enhancement: The super-resolution enhancement module uses the face restoration converter algorithm (CodeFormer) to perform 2x super-resolution processing on the captured face image to obtain a high-definition face image of 700×760 pixels. This image completely preserves the original face features and expressions, while restoring facial details without distortion, and is then sent back to the face recognition module.

[0087] Facial feature extraction and comparison: The facial recognition module uses the ArcFace face model with arc surface loss function to extract the feature vector A of the holder's photo in the reference face database and the feature vector B of the super-resolution high-definition face image. The cosine distance between vector A and vector B is calculated to be 0.504, and the similarity score is obtained by formula conversion as 1-0.504=0.496.

[0088] Similarity determination: The preset similarity threshold is 0.32. If the similarity is 0.496, it is determined to be the same person and there is no fraudulent activity. The result determination module outputs the recognition result (similarity 0.496, no fraudulent activity) and does not trigger the alarm mechanism.

[0089] To verify the effectiveness of this invention, five groups of low-quality face samples from light rail surveillance (covering typical scenarios such as low light, motion blur, and long-distance shooting) were selected. Face recognition comparison experiments were conducted using the method of this invention (including super-resolution enhancement) and existing conventional methods (without super-resolution enhancement). Some experimental data are shown in Table 1 below. Figure 3 As shown:

[0090] Table 1 Comparative Experimental Data

[0091] Sample number Similarity before super-resolution (conventional method) Similarity after super-resolution (method of this invention) Increase 1 0.072 0.248 243% 2 0.254 0.496 95% 3 0.298 0.453 52% 4 0.147 0.354 141% 5 0.158 0.224 42%

[0092] The experimental data shows that the average improvement in face recognition similarity of the 5 groups of samples reached 115%, and the similarity of all samples after super-score was significantly improved. Some samples that could not be accurately identified in the first place (similarity below the threshold) can be accurately identified after being processed by the method of this invention, which verifies the improvement effect of this invention on accuracy. At the same time, the average processing time per sample of the method of this invention is reduced by 40% compared with the prior art, which reflects the advantage of computational efficiency.

[0093] The intelligent face screening mechanism is based on sorting by detection confidence and combined with face size and location to assist in screening. It can exclude low-quality and non-target faces and ensure that only the best face is super-scored. This is the key to improving efficiency and avoiding waste of resources.

[0094] The Face Reconstruction Converter Algorithm (CodeFormer) differs from general super-resolution algorithms in that it incorporates prior facial knowledge to specifically repair facial details while maintaining feature stability and avoiding distortion caused by over-enhancement. This is the core of improving recognition accuracy.

[0095] The system first quickly filters out the best faces, then performs targeted super-resolution for accurate recognition, achieving a balance between speed and accuracy, and is suitable for the real-time detection needs of light rail.

[0096] In another embodiment of the present invention, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a method for detecting unauthorized use of light rail cards based on super-resolution enhancement as described in the preceding embodiments.

[0097] It should be noted that if the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0098] In the several embodiments provided in this application, it should be understood that the described systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0099] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The scope of the preferred embodiments of this application includes other implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which the embodiments of this application pertain.

[0100] The integrated modules described above can be implemented in either hardware or software. When used, each module will only collect and store user information with the user's full authorization and in compliance with relevant laws and regulations, protecting the security and privacy of user data and strictly prohibiting unauthorized access. Data processing will be conducted within the scope stipulated by law and will not exceed the purposes and scope authorized by the user. Users have the right to access, correct, delete, restrict, and refuse processing of their personal data, and must strictly comply with applicable laws and regulations and undergo compliance reviews.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for facial recognition in light rail fraud prevention based on super-resolution enhancement, characterized in that, The method includes the following steps: (1) Data loading and linkage matching processing: Load social security card transaction data and the standard photos of the corresponding social security card holders, and link and match the transaction data with the monitoring videos of the corresponding sites and transaction periods; (2) Keyframe extraction: Extract video frames of a preset duration before and after the transaction time from the matched surveillance video, then obtain keyframes at a fixed sampling rate and perform uniform scaling on the keyframes. (3) Face detection and intelligent screening: The key frame is multi-target face detection is performed using a preset face recognition model. At the same time, the detection confidence of each face is calculated, and the optimal face region is selected according to the confidence. (4) Targeted super-resolution enhancement: The selected optimal face region is reconstructed using a face restoration converter algorithm; (5) Feature comparison and theft determination: The face recognition model is used again to extract the super-resolution enhanced face feature vector and the feature vector of the corresponding standard photo, and the similarity between the two is calculated. The existence of light rail theft is determined by a preset threshold.

2. The method as described in claim 1, characterized in that, The arc surface loss function face model in the deep face framework is used as the face recognition model to detect faces in key frames; the model accurately identifies all possible face regions in the key frames and outputs the bounding box coordinates of each face.

3. The method as described in claim 2, characterized in that, During intelligent screening, the face with the highest confidence level is selected as the candidate face, and a comprehensive score is optimized by combining the face pixel size and the features of the center position of the image.

4. The method as described in claim 3, characterized in that, When selecting the optimal face region, the bounding box coordinates of the candidate face are used to expand the bounding box of the face by a preset number of pixels in all directions (top, bottom, left, and right), and the cropped face image is saved in a preset format for subsequent super-resolution processing.

5. The method according to any one of claims 1 to 4, characterized in that, If it is determined that there is fraudulent use of light rail credit, an alarm mechanism will be triggered for the case based on the operational needs of the light rail.

6. A facial recognition system for light rail card skimming based on super-resolution enhancement, characterized in that, The system includes: The data loading module serves as the system's data input source, responsible for loading social security card transaction data and the corresponding standard photos of social security card holders, and for linking and matching transaction data with monitoring videos of the corresponding sites and transaction periods. The video processing module is used to receive video data transmitted by the data loading module, extract video frames of preset duration before and after the transaction time from the matched monitoring video, obtain key frames at a fixed sampling rate, and perform uniform scaling on the key frames. The face recognition module is used to perform multi-target face detection on keyframes using a preset face recognition model, calculate the detection confidence of each face, and then perform intelligent filtering according to the confidence of the face to select the optimal face region. The super-resolution enhancement module is used to perform super-resolution reconstruction on the selected optimal face region using a face restoration converter algorithm; The result determination module is used to extract the super-resolution enhanced facial feature vector and the corresponding feature vector of the standard photo using the facial recognition model, calculate the similarity between the two, and determine whether there is light rail fraud by using a preset threshold. An alarm mechanism is triggered for cases determined to be fraudulent transactions.

7. The system as described in claim 6, characterized in that, The arc surface loss function face model in the deep face framework is used as the face recognition model to detect faces in key frames; the model accurately identifies all possible face regions in the key frames and outputs the bounding box coordinates of each face.

8. The system as described in claim 7, characterized in that, During intelligent screening, the face with the highest confidence level is selected as the candidate face, and a comprehensive score is optimized by combining the face pixel size and the features of the center position of the image.

9. The system as described in claim 8, characterized in that, When selecting the optimal face region Based on the bounding box coordinates of the candidate's face, the bounding box of the face is expanded by a preset number of pixels in all directions (top, bottom, left, and right), and the cropped face image is saved in a preset format for subsequent super-resolution processing.

Images