Face recognition system based on double-camera control ball

Abstract

The application provides a face recognition system based on a double-camera control ball, comprising a patrol module, a face snapshot module, a snapshot adjustment module, a face recognition module and an identity output module; the patrol module is used to set a violation patrol task for the double-camera control ball of a construction site and collect a violation scene image when there is a violation personnel; the face snapshot module is used to determine a first employee face image according to the violation scene image; the snapshot adjustment module is used to determine whether the first employee face image reaches a preset standard and perform employee face snapshot when the preset standard is not reached; the face recognition module is used to obtain a second employee face image and perform face recognition by comparing a face database; and the identity output module is used to determine identity information of the employee according to the face recognition.

Description

Technical Field

[0001] This invention relates to the field of construction supervision technology, and in particular to a face recognition system based on a dual-camera surveillance sphere. Background Technology

[0002] Currently, during construction site operations, network signals can be extremely unstable due to geographical factors, especially at power line construction sites. This is because poles and lines are often located in remote, inaccessible mountainous areas, or in other remote locations where network coverage is nonexistent. Data from equipment used at construction sites for tension, tilt, settlement, and video analytics cannot be transmitted, and measurement results from surveying equipment cannot be synchronized to the cloud. Therefore, for on-site worker monitoring, surveillance cameras are often deployed for monitoring, enabling project supervision and identification of worker violations.

[0003] In current technology, surveillance cameras primarily employ single-camera or multi-camera distributed image fusion for facial recognition and on-site violation capture. Single-camera cameras mainly use visible light cameras. After identifying the target, due to the inherent characteristics of 2D imaging (lacking depth information), the size and distance are proportionally scaled. During facial recognition, this scaling and magnification mean that subsequent facial mapping and localization are based on this single unit. Due to the robustness of the surrounding environment and the sparsity of corner points, accurate facial recognition is impossible. Facial recognition fails at slightly greater distances and requires significant computational resources. Multi-camera distributed image fusion addresses the issue of facial position matching between multiple cameras. Because the intrinsic parameters of each camera differ, the optical axes of different lenses have angular discrepancies, leading to high overlap in facial recognition. This overlap requires multiple offset compensations to achieve facial recognition, thus also demanding substantial computational resources. Furthermore, both of these methods require significant investment of human, material, and network resources in the engineering field, especially in remote engineering sites. In patent CN113593177B, a video alarm linkage method based on high-precision positioning and image recognition, tracking identification, violation detection, and facial recognition can be performed using a surveillance camera. However, it also has a significant drawback: it only uses a built-in facial recognition algorithm for on-site facial recognition, while the surveillance camera needs to capture the entire scene. The camera also lacks depth sensing capabilities, therefore, the accuracy of facial recognition and the clarity of the captured image cannot be assessed.

[0004] Therefore, a new on-site monitoring system is needed to reduce the investment of human, material, and network resources at the construction site. Summary of the Invention

[0005] This invention provides a face recognition system based on a dual-camera surveillance sphere, which addresses the problem in construction sites where the distributed image fusion technology using single-camera and multi-camera systems only processes the original image, lacks depth information, and consumes a lot of manpower, material resources, and network resources for image processing.

[0006] A face recognition system based on a dual-camera PTZ camera includes:

[0007] Inspection module: Used to set up violation inspection tasks for dual-camera surveillance cameras at the construction site, and to collect images of the violation scene when violations are detected; among which,

[0008] The dual-camera PTZ camera is equipped with preset points, which are equipped with a human body capture mechanism. The human body capture mechanism captures the human body image based on the preset points when the cameras of the dual-camera PTZ camera move.

[0009] Face capture module: used to import the violation scene image into the violation feature extraction network to determine the face image of the first employee; wherein,

[0010] The violation feature extraction network includes: violation image scale space, violation behavior Gaussian difference pyramid, violation interpolation, and violation descriptor;

[0011] Capture adjustment module: used to determine whether the first employee's facial image meets the preset standard, and to capture the employee's face if the preset standard is not met;

[0012] Face recognition module: used to acquire the captured image of the second employee's face and compare it with the face database for face recognition;

[0013] Identity output module: used to determine the employee's identity information based on the facial recognition.

[0014] Furthermore, the face database includes the following construction methods:

[0015] Obtain information on construction site employees; among which,

[0016] The employee information includes: identity information, job information, and job type;

[0017] The employee information is divided into identity information data, job title data, and job type data;

[0018] The identity information data includes: employee facial image, name, and contact information;

[0019] The job data includes: employee work location and employee job title;

[0020] The job type data includes: the employee's job type and the employee's working hours;

[0021] Based on the identity information data, job title data, and job type data, a three-tiered data table is generated;

[0022] A face recognition database is generated based on the three-layer data tables.

[0023] Furthermore: the inspection module includes:

[0024] Task formulation unit: used to set up dual-camera surveillance cameras at the construction site, and to conduct site inspections using these cameras to determine whether employees are engaging in any violations; wherein,

[0025] The dual-camera PTZ camera includes a main camera and an auxiliary camera;

[0026] The main camera is used to acquire scene images;

[0027] The auxiliary camera is used to enhance the employees in the scene image;

[0028] Violation Scene Acquisition Unit: Used to acquire violation scene images based on the violation behavior; wherein,

[0029] A database of violations is pre-set. By comparing the violation images in the database with the scene images captured by the dual-camera surveillance system, it is determined whether any violations exist.

[0030] The dual-camera control ball is equipped with a steering gimbal, which is used to acquire scene images from different directions.

[0031] Employee determination unit: used to determine whether a violating employee exists in the violation scene image based on the violation scene image; wherein,

[0032] If no employee is found violating the rules in the image of the violation scene, an alarm will be triggered for the violation at the location of the violation.

[0033] If a violating employee is present in the violation scene image, the violating employee will be marked as having violated the rules.

[0034] Image optimization unit: used to filter the violation scene images according to the violation markers, and determine the target images containing employee facial images; wherein,

[0035] The target image includes images of the employee's face.

[0036] Furthermore: the task formulation unit includes:

[0037] Feature extraction subunit: used to acquire historical violation data, classify violations, and determine the characteristics of different violation types;

[0038] Framework construction subunit: used to construct a decision tree-based behavior integration framework based on the characteristics of the violation; wherein,

[0039] The integrated framework includes a structured behavior violation template, a set of behavior image samples, and a behavior feature database. The structured behavior violation template, the set of behavior image samples, and the behavior feature database are all equipped with comparison tools for comparative analysis.

[0040] Detection setting subunit: used to set detection mechanisms for different violations based on the behavior integration framework; wherein,

[0041] The detection mechanism is based on behavioral benchmarks and behavioral characteristics of different behaviors within the behavioral integration framework;

[0042] Process determination subunit: used to determine the detection process for different violations based on the detection mechanism;

[0043] Route setting subunit: used to acquire the site map of the construction site, and set the inspection route of different dual-camera surveillance balls according to the site map;

[0044] Task setting subunit: used to set up violation inspection tasks based on the inspection route and detection process.

[0045] Furthermore: the face capture module includes:

[0046] Event recording unit: Based on the violation scene image, determines the violation type, and generates a violation event record by acquiring the image through the dual-camera surveillance camera; wherein,

[0047] The types of violations include immediate violations and evolutionary violations;

[0048] Analysis Unit: Based on the recorded violations, the unit performs event analysis and image analysis using the edge detection device built into the dual-camera surveillance sphere; wherein,

[0049] The event analysis is used to analyze the severity of violations and the methods used to trigger alerts, and to determine the time of the violation.

[0050] The image analysis is used to determine whether there is a facial image of a violating employee in the image of the violation scene;

[0051] Video recording unit: used to retrieve violation video footage through the dual-camera surveillance camera based on the violation event, and associate the violation video footage with the violation event record;

[0052] When the violation type is an immediate violation, capture the real-time violation image from the violation video recording;

[0053] When the violation type is an evolving violation, an immediate alarm is triggered and an alarm signal is generated. The evolving violation image is captured from the violation video recording.

[0054] Recognition unit: used to extract the image of the employee's face region from the real-time violation image and the violation evolution image, and generate a first employee face image.

[0055] Furthermore: the snapshot adjustment module includes:

[0056] Image analysis unit: used to analyze and process the facial image of the first employee to determine image information; wherein,

[0057] The image information includes: image clarity information and information on the employee's face displayed in the image;

[0058] Image compliance judgment unit: used to determine whether the image information meets the face recognition standard by using a preset image standard, and to determine the missing information when the face recognition standard is not met;

[0059] Missing information adjustment unit: used to control the dual-camera control ball to re-capture the employee's face based on the missing information, and generate a second employee face image.

[0060] Furthermore, the image analysis unit includes the following analysis steps:

[0061] Sharpness analysis subunit: used to obtain the three-dimensional vector of the face in the first employee's facial image; wherein,

[0062] The three-dimensional vector is a vector generated based on the three-dimensional information of facial feature points in the first employee's facial image. The blurriness value of the first employee's facial image is calculated based on the three-dimensional vector and a preset calculation rule.

[0063] The calculation rule is a data processing rule that calculates the blur value by weighting the three-dimensional vector, and determines whether the first employee's facial image meets the preset clarity condition, wherein the clarity condition is that the blur value is less than the preset threshold.

[0064] Face display subunit: used to pre-set a face puzzle template, fill the face image of the first employee into the face puzzle template, determine whether it can be fully filled, and output the employee's face display information when it cannot be fully filled.

[0065] Furthermore: the face recognition module includes:

[0066] Feature extraction unit: used to extract the facial features to be identified from the second employee's facial image;

[0067] Feature model unit: used to generate a facial feature model based on the facial features to be identified; wherein,

[0068] The facial feature model is a comparison model used for facial feature comparison;

[0069] Recognition unit: used to perform facial recognition on the second employee's facial image based on the facial database and the facial feature model; wherein,

[0070] The facial recognition includes facial matching and identity information retrieval.

[0071] Furthermore: the feature model unit includes:

[0072] Model building unit: Used to build a similarity comparison model of facial features using neural networks;

[0073] Training unit: used to generate a training set of employee facial images based on the facial database; wherein,

[0074] The training sample set includes positive samples, regular samples, and negative samples of employee facial images;

[0075] The positive samples are employee facial images with high clarity and a number of facial features exceeding a first preset threshold;

[0076] The negative samples are employee facial images with low resolution and fewer than a first preset threshold number of facial features;

[0077] The standard samples are employee facial images whose number of facial features falls within the first preset range.

[0078] The first preset threshold is the standard range of identifiable features of an employee's face;

[0079] The similarity comparison model is trained using the training set to obtain a facial feature model.

[0080] Furthermore: the identity output module includes:

[0081] Identity statistics unit: used to determine the corresponding target employee information in the face database based on the facial recognition;

[0082] Violation statistics unit: used to determine the violation information of the target employee based on the violation scene image and mark the violation;

[0083] Output unit: Used to generate violation logs based on the target employee information and violation markers, and upload them to the cloud control center via dual-camera surveillance cameras.

[0084] The beneficial effects of this invention are as follows: This invention can perform facial recognition and output facial information during the inspection of construction sites through the inspection mechanism. Therefore, this invention can also provide theft warnings. When the recognized face is not the employee information known to the company, it can realize remote alarm and automatic supervision of the construction site.

[0085] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0086] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0087] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0088] Figure 1 This invention provides a face recognition system based on a dual-camera PTZ camera.

[0089] Figure 2 This is a flowchart illustrating the construction process of the face database in an embodiment of the present invention. Detailed Implementation

[0090] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0091] A face recognition system based on a dual-camera PTZ camera includes:

[0092] Inspection module: Used to set up violation inspection tasks for dual-camera surveillance cameras at the construction site, and to collect images of the violation scene when violations are detected; among which,

[0093] The dual-camera PTZ camera is equipped with preset points, which are equipped with a human body capture mechanism. The human body capture mechanism captures the human body image based on the preset points when the cameras of the dual-camera PTZ camera move.

[0094] Face capture module: used to import the violation scene image into the violation feature extraction network to determine the face image of the first employee; wherein,

[0095] The violation feature extraction network includes: violation image scale space, violation behavior Gaussian difference pyramid, violation interpolation, and violation descriptor;

[0096] Capture adjustment module: used to determine whether the first employee's facial image meets the preset standard, and to capture the employee's face if the preset standard is not met;

[0097] Face recognition module: used to acquire the captured image of the second employee's face and compare it with the face database for face recognition;

[0098] Identity output module: used to determine the employee's identity information based on the facial recognition.

[0099] The principle of the above technical solution is as follows: (see attached) Figure 1 As shown, this invention relates to facial recognition at construction sites. In this invention, dual-camera monitoring cameras are deployed at the construction site for inspection. These cameras detect violations through inspection. During the violation detection process, the image may be unclear due to factors such as the shooting angle or dust contamination of the camera lens, meaning it may not meet the preset standards. In this case, the invention will adjust the image and re-acquire the facial image of the person violating the rules, thereby confirming the violation. Similarly, this invention's system can also be used as a time clock / check-in tool.

[0100] The beneficial effects of the above technical solution are as follows: the present invention can perform facial recognition and output facial information during the inspection of construction sites through the inspection mechanism. Therefore, the present invention can also provide theft warning. When the recognized face is not the employee information known to the company, the present invention can realize remote alarm and automatic supervision of the construction site.

[0101] Furthermore, the face database includes the following construction methods:

[0102] Obtain information on construction site employees; among which,

[0103] The employee information includes: identity information, job information, and job type;

[0104] The identification information includes the employee's name and facial information;

[0105] The employee information is divided into identity information data, job title data, and job type data;

[0106] The identity information data includes: employee facial image, name, and contact information;

[0107] The job data includes: employee work location and employee job title;

[0108] The job type data includes: the employee's job type and the employee's working hours;

[0109] Based on the identity information data, job title data, and job type data, a three-tiered data table is generated;

[0110] A face recognition database is generated based on the three-layer data tables.

[0111] The principle of the above technical solution is as follows: (see attached) Figure 2 As shown, in the process of building the face database, this invention collects employee information, including identity information, job information and job type; the different identity information is divided into three layers, each layer is a data table, and finally the face recognition database is formed by these data tables.

[0112] The beneficial effects of the above technical solution are as follows: compared with the prior art, the present invention can know the specific information of each person on the construction site more clearly, and the identification of violations and facial recognition are clearer through this specific information.

[0113] Furthermore: the inspection module includes:

[0114] Task formulation unit: used to set up dual-camera surveillance cameras at the construction site, and to conduct site inspections using these cameras to determine whether employees are engaging in any violations; wherein,

[0115] The dual-camera PTZ camera includes a main camera and an auxiliary camera;

[0116] The main camera is used to acquire scene images;

[0117] The auxiliary camera is used to enhance the employees in the scene image;

[0118] Violation Scene Acquisition Unit: Used to acquire violation scene images based on the violation behavior; wherein,

[0119] A database of violations is pre-set. By comparing the violation images in the database with the scene images captured by the dual-camera surveillance system, it is determined whether any violations exist.

[0120] The dual-camera control ball is equipped with a steering gimbal, which is used to acquire scene images from different directions.

[0121] Employee determination unit: used to determine whether a violating employee exists in the violation scene image based on the violation scene image; wherein,

[0122] If no employee is found violating the rules in the image of the violation scene, an alarm will be triggered for the violation at the location of the violation.

[0123] If a violating employee is present in the violation scene image, the violating employee will be marked as having violated the rules.

[0124] Image optimization unit: used to filter the violation scene images according to the violation markers, and determine the target images containing employee facial images; wherein,

[0125] The target image includes images of the employee's face.

[0126] The principle of the above technical solution is as follows: (see attached) Figure 1 As shown, in the inspection module of this invention, a dual-camera surveillance sphere, including a main camera and an auxiliary camera, captures scene images and enhances the images of employees within the scene, making violations clearer and facilitating facial recognition of violators. The dual-camera surveillance sphere of this invention includes a pan-tilt unit (PTZ), allowing for changes in the shooting direction. When a violating employee is present, they are marked, and finally, the target image is obtained from the violation scene image.

[0127] In practice, the dual-camera surveillance system uses the main camera at the bottom to capture images of the entire scene, identifying violations. During facial recognition, the auxiliary camera at the top focuses and enhances the recognition effect, improving image clarity and thus enabling facial recognition functionality.

[0128] The beneficial effects of the above technical solution are: the present invention can obtain clearer photos and perform finer screening based on this.

[0129] Furthermore: the task formulation unit includes:

[0130] Feature extraction subunit: used to acquire historical violation data, classify violations, and determine the characteristics of different violation types;

[0131] Framework construction subunit: used to construct a decision tree-based behavior integration framework based on the characteristics of the violation; wherein,

[0132] The integrated framework includes a structured behavior violation template, a set of behavior image samples, and a behavior feature database. The structured behavior violation template, the set of behavior image samples, and the behavior feature database are all equipped with comparison tools for comparative analysis.

[0133] The comparison tool is used to compare and classify violations.

[0134] Detection setting subunit: used to set detection mechanisms for different violations based on the behavior integration framework; wherein,

[0135] The detection mechanism is based on behavioral benchmarks and behavioral characteristics of different behaviors within a behavioral integration framework; behavioral benchmarks are the standard work actions and behaviors of employees. Behavioral characteristics represent indicative information about employee violations or standard work behaviors during work.

[0136] Process determination subunit: used to determine the detection process for different violations based on the detection mechanism;

[0137] Route setting subunit: used to acquire the site map of the construction site, and set the inspection route of different dual-camera surveillance balls according to the site map;

[0138] Task setting subunit: used to set up violation inspection tasks based on the inspection route and detection process.

[0139] The principle of the above technical solution is as follows: This invention will formulate tasks based on historical violations, and the violations are the benchmark for determining the inspectors. By integrating the decision tree of the violations, the invention can compare and identify different violations, thereby realizing construction site inspection.

[0140] The beneficial effects of the above technical solution are as follows: the present invention can automatically set inspection tasks, and based on the inspection tasks, realize on-site inspection of the construction site and determine whether there are any violations.

[0141] Furthermore: the face capture module includes:

[0142] Event recording unit: Based on the violation scene image, determines the violation type, and generates a violation event record by acquiring the image through the dual-camera surveillance camera; wherein,

[0143] The types of violations include immediate violations and evolutionary violations;

[0144] Immediate violations are behaviors that can be directly detected and are currently in progress; evolving violations are behaviors that users have not yet violated, but may lead to violations if not corrected, such as employees performing high-standard work when they are physically abnormal or in a poor mental state.

[0145] Analysis Unit: Based on the recorded violations, the unit performs event analysis and image analysis using the edge detection device built into the dual-camera surveillance sphere; wherein,

[0146] The event analysis is used to analyze the severity of violations and the methods used to trigger alerts, and to determine the time of the violation.

[0147] The image analysis is used to determine whether there is a facial image of a violating employee in the image of the violation scene;

[0148] Video recording unit: used to retrieve violation video footage through the dual-camera surveillance camera based on the violation event, and associate the violation video footage with the violation event record;

[0149] When the violation type is an immediate violation, capture the real-time violation image from the violation video recording;

[0150] When the violation type is an evolving violation, an immediate alarm is triggered and an alarm signal is generated. The evolving violation image is captured from the violation video recording.

[0151] Recognition unit: used to extract the image of the employee's face region from the real-time violation image and the violation evolution image, and generate a first employee face image.

[0152] The principle of the above technical solution is as follows: (see attached) Figure 1 As shown, when performing face capture, this invention can analyze violations based on the image of the violation scene. However, this invention identifies two types of violations: "immediate violations and evolving violations," that is, situations where violations are in progress and situations that are likely to develop into violations. For example, if it is too hot on a construction site, workers may have to not wear helmets. For these violations, behavioral analysis and behavioral evolution are performed to ensure that all possible violations are highly likely to be detected.

[0153] The beneficial effects of the above technical solution are as follows: the present invention can identify different types of violations in different ways, thereby handling the violations.

[0154] Furthermore: the snapshot adjustment module includes:

[0155] Image analysis unit: used to analyze and process the facial image of the first employee to determine image information; wherein,

[0156] The image information includes: image clarity information and information on the employee's face displayed in the image;

[0157] Image compliance judgment unit: used to determine whether the image information meets the face recognition standard by using a preset image standard, and to determine the missing information when the face recognition standard is not met;

[0158] Missing information adjustment unit: used to control the dual-camera control ball to re-capture the employee's face based on the missing information, and generate a second employee face image.

[0159] The principle of the above technical solution is as follows: (see attached) Figure 1 As shown, this invention can capture and analyze images of violations. This is because the images obtained during violation identification are insufficient to determine the specific identity information of the violator, thus adjusting the dual-camera control ball to determine the employee's facial information.

[0160] The beneficial effect of the above technical solution is that when the image of the violation identification is found to be defective, the present invention will re-capture a clearer and more distinct facial image of the violation person to prevent failure to identify.

[0161] Furthermore, the facial recognition module will also assess the clarity of the offender's facial image. If the clarity is insufficient, the facial image of the offender will be re-captured. The specific process is as follows:

[0162] Step 1: Acquire video data and divide the video data into multiple frame images;

[0163] Step 2: Calculate the average pixel value for each frame of the image:

[0164]

[0165] in, Indicates the pixel mean; x j，i y represents the color temperature value of the i-th pixel in the j-th image; j，i L(x) represents the grayscale value of the i-th pixel in the j-th image; i∈n, n represents the total number of pixels in the frame image; j∈m, m represents the total number of frames when the video data is divided into frames; L(x) j，i ,y j，i () represents the pixel function of the i-th pixel in the j-th image;

[0166] Step 3: Calculate the variance of each frame image:

[0167]

[0168] Step 4: Based on the variance and pixel mean of each frame image, construct a brightness detection model and determine whether the brightness of the frame image reaches the preset brightness.

[0169]

[0170] in, This indicates the expected average color temperature. P represents the expected average grayscale value; when P < 1, it means the video data has not reached the preset brightness; when P ≥ 1, it means the video data has reached the preset brightness.

[0171] The specific principle of the above scheme is as follows: During the face recognition process, this invention also determines the video brightness. By judging the video brightness, it determines whether the video data meets the requirements for face recognition. Moreover, this invention is based on frame-by-frame image segmentation for individual recognition, so that when some videos are unclear or lack sufficient brightness, individual images with sufficient brightness are recognized individually. In step 1, this invention calculates the pixel mean, which is the average color temperature of each image after the video data is segmented into frames. The calculation of the average color temperature involves measuring the color and grayscale values ​​in the video data. The grayscale value determines the brightness of the image. In step 2, the variance of each image is calculated based on the color temperature, also considering the overall color temperature and brightness of the video data. In step 3, a brightness detection model is constructed. This brightness detection model is based on the actual color temperature and grayscale values. After subtracting the expected color temperature and grayscale values, the model calculates the correlation between the mean and variance of the actual color temperature and grayscale values ​​and the expected color temperature and grayscale values. If, after subtracting the expected color temperature and grayscale values ​​(i.e., after achieving the target color temperature and grayscale values), the correlation variance and the correlation between the pixel mean and the expected color temperature and grayscale values ​​are still greater than or equal to 1, that is, the correlation is greater than 1, it means that the preset brightness has been achieved; otherwise, the preset brightness has not been achieved.

[0172] Furthermore, the image analysis unit includes the following analysis steps:

[0173] Sharpness analysis subunit: used to obtain the three-dimensional vector of the face in the first employee's facial image; wherein,

[0174] The three-dimensional vector is a vector generated based on the three-dimensional information of facial feature points in the first employee's facial image. The blurriness value of the first employee's facial image is calculated based on the three-dimensional vector and a preset calculation rule.

[0175] The calculation rule is a data processing rule that calculates the blur value by weighting the three-dimensional vector, and determines whether the first employee's facial image meets the preset clarity condition, wherein the clarity condition is that the blur value is less than the preset threshold.

[0176] Face display subunit: used to pre-set a face puzzle template, fill the face image of the first employee into the face puzzle template, determine whether it can be fully filled, and output the employee's face display information when it cannot be fully filled.

[0177] The principle of the above technical solution is as follows: This invention can analyze and three-dimensionally vectorize the facial image of the first employee. Through the three-dimensional vector, the blurriness of the first employee's facial image can be calculated, thereby determining whether the image meets the clarity requirements for facial recognition. Furthermore, it can determine whether the facial image can completely fill the facial mosaic template, thus determining whether the employee's facial information is displayed completely.

[0178] The beneficial effects of the above technical solution are as follows: the present invention can determine the direction that needs to be adjusted, that is, the missing part of the first facial image, and then adjust it by means of a dual-camera control ball, and then take another picture by means of a dual-camera control ball to determine the facial image.

[0179] Furthermore: the face recognition module includes:

[0180] Feature extraction unit: used to extract the facial features to be identified from the second employee's facial image;

[0181] Feature model unit: used to generate a facial feature model based on the facial features to be identified; wherein,

[0182] The facial feature model is a comparison model used for facial feature comparison;

[0183] Recognition unit: used to perform facial recognition on the second employee's facial image based on the facial database and the facial feature model; wherein,

[0184] The facial recognition includes facial matching and identity information retrieval.

[0185] The principle of the above technical solution is as follows: When performing face recognition, the present invention constructs a face feature model, and then uses the face feature model to match face information based on comparison to determine the corresponding identity information.

[0186] The beneficial effect of the above technical solution is that the present invention can perform face recognition through special matching to determine specific face information.

[0187] Furthermore: the feature model unit includes:

[0188] Model building unit: Used to build a similarity comparison model of facial features using neural networks;

[0189] Training unit: used to generate a training set of employee facial images based on the facial database; wherein,

[0190] The training sample set includes positive samples, regular samples, and negative samples of employee facial images;

[0191] The positive samples are employee facial images with high clarity and a number of facial features exceeding a first preset threshold;

[0192] The negative samples are employee facial images with low resolution and fewer than a first preset threshold number of facial features;

[0193] The standard samples are employee facial images whose number of facial features falls within the first preset range.

[0194] The first preset threshold is the standard range of identifiable features of an employee's face;

[0195] The similarity comparison model is trained using the training set to obtain a facial feature model.

[0196] Furthermore: the identity output module includes:

[0197] Identity statistics unit: used to determine the corresponding target employee information in the face database based on the facial recognition;

[0198] Violation statistics unit: used to determine the violation information of the target employee based on the violation scene image and mark the violation;

[0199] Output unit: Used to generate violation logs based on the target employee information and violation markers, and upload them to the cloud control center via dual-camera surveillance cameras.

[0200] The principle of the above technical solution is as follows: (see attached) Figure 1 As shown, when outputting identity information, this invention will collect statistics on the identity information of employees who violate the rules and the violations themselves, and then send the information to the cloud management center. The cloud management center will then send this information to the corresponding client.

[0201] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A face recognition system based on a dual-camera PTZ camera, characterized in that, include: Inspection module: Used to set up violation inspection tasks for dual-camera surveillance cameras at the construction site, and to collect images of the violation scene when violations are detected; among which, The dual-camera surveillance sphere is equipped with preset points, which are equipped with a human body capture mechanism. When the cameras of the dual-camera surveillance sphere move, the human body capture mechanism captures the human body image based on the preset points. Face capture module: used to import the violation scene image into the violation feature extraction network to determine the face image of the first employee; wherein, The violation feature extraction network includes: violation image scale space, violation behavior Gaussian difference pyramid, violation interpolation, and violation descriptor; Capture adjustment module: used to determine whether the first employee's facial image meets the preset standard, and to capture the employee's face if the preset standard is not met; Face recognition module: used to acquire the captured image of the second employee's face and compare it with the face database for face recognition; Identity output module: used to determine the employee's identity information based on the facial recognition; The inspection module includes: Task formulation unit: used to set up dual-camera surveillance cameras at the construction site and conduct site inspections through these cameras to determine whether employees have engaged in any violations; wherein, The dual-camera PTZ camera includes a main camera and an auxiliary camera; The main camera is used to acquire scene images; The auxiliary camera is used to enhance the employees in the scene image; The snapshot adjustment module includes: Image analysis unit: used to analyze and process the facial image of the first employee to determine image information; wherein, The image information includes: image clarity information and information on the employee's face displayed in the image; Image compliance judgment unit: used to determine whether the image information meets the face recognition standard by using a preset image standard, and to determine the missing information when the face recognition standard is not met; Missing information adjustment unit: used to control the dual-camera control ball to re-capture the employee's face based on the missing information, and generate a second employee face image.

2. The face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The face database includes the following construction methods: Obtain information on construction site employees; among which, The employee information includes: identity information, job information, and job type; The employee information is divided into identity information data, job title data, and job type data; The identity information data includes: employee facial image, name, and contact information; The job data includes: employee work location and employee job title; The job type data includes: the employee's job type and the employee's working hours; Based on the identity information data, job title data, and job type data, a three-tiered data table is generated; A face recognition database is generated based on the three-layer data tables.

3. The face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The inspection module also includes: Violation Scene Acquisition Unit: Used to acquire violation scene images based on the violation behavior; wherein, A database of violations is pre-set. By comparing the violation images in the database with the scene images captured by the dual-camera surveillance system, it is determined whether any violations exist. The dual-camera control ball is equipped with a steering gimbal, which is used to acquire scene images from different directions. Employee determination unit: used to determine whether a violating employee exists in the violation scene image based on the violation scene image; wherein, If no employee is found violating the rules in the image of the violation scene, an alarm will be triggered for the violation at the location of the violation. If a violating employee is present in the violation scene image, the violating employee will be marked as having violated the rules. Image optimization unit: used to filter the violation scene images according to the violation markers, and determine the target images containing employee facial images; wherein, The target image includes images of the employee's face.

4. A face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The task formulation unit includes: Feature extraction subunit: used to acquire historical violation data, classify violations, and determine the characteristics of different violation types; Framework construction subunit: used to construct a decision tree-based behavior integration framework based on the characteristics of the violation; wherein, The behavior integration framework includes a structured behavior violation template, a set of behavior image samples, and a behavior feature database. The structured behavior violation template, the set of behavior image samples, and the behavior feature database are all equipped with comparison tools for comparative analysis. Detection setting subunit: used to set detection mechanisms for different violations based on the behavior integration framework; wherein, The detection mechanism is based on behavioral benchmarks and behavioral characteristics of different behaviors within the behavioral integration framework; Process determination subunit: used to determine the detection process for different violations based on the detection mechanism; Route setting subunit: used to acquire the site map of the construction site, and set the inspection route of different dual-camera surveillance balls according to the site map; Task setting subunit: used to set up violation inspection tasks based on the inspection route and detection process.

5. A face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The face capture module includes: Event recording unit: Based on the violation scene image, determines the violation type, and generates a violation event record by acquiring the image through the dual-camera surveillance camera; wherein, The types of violations include immediate violations and evolutionary violations; Analysis Unit: Based on the recorded violations, the unit performs event analysis and image analysis using the edge detection device built into the dual-camera surveillance sphere; wherein, The event analysis is used to analyze the severity of violations and the methods used to trigger alerts, and to determine the time of the violation. The image analysis is used to determine whether there is a facial image of a violating employee in the image of the violation scene; Video recording unit: used to retrieve violation video recordings through the dual-camera surveillance camera based on the violation event records, and associate the violation video recordings with the violation event records; When the violation type is an immediate violation type, capture the real-time violation image from the violation recording; When the violation type is an evolutionary violation, an immediate alarm is triggered and an alarm signal is generated. The violation evolution image is captured from the violation video recording. Recognition unit: used to extract the image of the employee's face region from the real-time violation image and the violation evolution image, and generate a first employee face image.

6. A face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The image analysis unit includes the following analysis steps: Sharpness analysis subunit: used to obtain the three-dimensional vector of the face in the first employee's facial image; wherein, The three-dimensional vector is a vector generated based on the three-dimensional information of facial feature points in the first employee's facial image. The blur value of the first employee's facial image is calculated based on the three-dimensional vector and a preset calculation rule. The calculation rule is a data processing rule that calculates the blur value by weighting the three-dimensional vector, and determines whether the first employee's facial image meets the preset clarity condition, wherein the clarity condition is that the blur value is less than the preset threshold. Face display subunit: used to pre-set a face puzzle template, fill the face image of the first employee into the face puzzle template, determine whether it can be fully filled, and output the employee's face display information when it cannot be fully filled.

7. A face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The face recognition module includes: Feature extraction unit: used to extract the facial features to be identified from the second employee's facial image; Feature model unit: used to generate a facial feature model based on the facial features to be identified; wherein, The facial feature model is a comparison model used for facial feature comparison; Recognition unit: used to perform facial recognition on the second employee's facial image based on the facial database and the facial feature model; wherein, The facial recognition includes facial matching and identity information retrieval.

8. A face recognition system based on a dual-camera PTZ camera as described in claim 7, characterized in that, The feature model unit includes: Model building unit: Used to build a similarity comparison model of facial features using neural networks; Training unit: used to generate a training sample set of employee facial images based on the face database; wherein, The training sample set includes positive samples, regular samples, and negative samples of employee facial images; The positive samples are employee facial images with high clarity and a number of facial features exceeding a first preset threshold; The negative samples are employee facial images with low clarity and fewer than a first preset threshold number of facial features; The regular samples are employee facial images where the number of facial features is within the first preset threshold. The first preset threshold is the standard range of identifiable features of an employee's face; The similarity comparison model is trained using the training sample set to obtain a facial feature model.

9. A face recognition system based on a dual-camera PTZ camera as described in claim 1, characterized in that, The identity output module includes: Identity statistics unit: used to determine the corresponding target employee information in the face database based on the facial recognition; Violation statistics unit: used to determine the violation information of the target employee based on the violation scene image and mark the violation; Output unit: Used to generate violation logs based on the target employee information and violation markers, and upload them to the cloud control center via dual-camera surveillance cameras.

Images