Face temperature measurement method for automatically capturing face based on Fisher algorithm

Abstract

The invention discloses a face temperature measurement method for automatically capturing a face based on a Fisher algorithm, a face temperature measurement terminal comprises a door frame, a gate and a camera, the gate and the camera are arranged on the door frame, and the door frame is provided with a first driving assembly used for driving the gate to move and a second driving assembly used for driving the camera to move. A high-definition camera module, an infrared camera module and a laser ranging module are arranged at the shooting end of the camera, a CPU, a GPU, a storage module and a wireless module are arranged in the camera in an embedded mode, and the first driving assembly, the second driving assembly and the camera are all controlled by a computer; according to the face temperature measurement method, a face detection module, a face recognition module, a body temperature detection module, a UI module, a room temperature automatic calibration module, a distance measurement algorithm and a compensation algorithm are arranged in a CPU, a face detection algorithm, a face recognition algorithm, a living body detection algorithm and an ornament recognition algorithm are configured in a GPU, and the face recognition algorithm is a Fisher algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and particularly relates to a face temperature measurement terminal for automatically capturing faces. Background Art

[0002] Traditional temperature measurement devices such as forehead thermometers only have a single temperature measurement function, and a large amount of manpower is required to intercept and measure temperatures, and the contact also increases the possibility of infection.

[0003] The existing solution is to use a traditional forehead thermometer for close-contact interception and temperature measurement. However, in addition to the single function, it requires a large amount of social human resources, and there are negative temperature measurement and non-standard temperature measurement operations by personnel, increasing the cost of measuring body temperature and increasing the possibility of infection. Summary of the Invention

[0004] In view of the above-mentioned drawbacks of the prior art, the present invention provides a face temperature measurement terminal for automatically capturing faces, which can effectively solve the problems of single function of current forehead thermometers or other temperature measurement devices, manpower consumption, increased disease infection probability, or negative temperature measurement.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A face temperature measurement terminal for automatically capturing faces, comprising a door frame, a gate provided on the door frame, and a camera provided on the door frame; A negative pressure component is provided at the ground at the entrance end of the door frame; A first driving component for driving the movement of the gate is provided on the door frame; A second driving component for driving the movement of the camera is further provided on the door frame. The shooting end of the camera is provided with a high-definition camera module, an infrared camera module, and a laser ranging module. The inside of the camera is embedded with a CPU, a GPU, a storage module, and a wireless module; The first driving component, the second driving component, and the camera are all controlled by a computer.

[0006] Furthermore, protective fences are provided at both ends of the door frame.

[0007] Furthermore, gates are provided at the inner ends of the rods of the door frame perpendicular to the ground; the first driving component includes a hinge seat provided on the door frame and a rotary solenoid valve provided on the hinge seat. The output shaft of the rotary solenoid valve is fixedly connected to the gate installed on the hinge seat, and the central axis of the output shaft of the rotary solenoid valve is perpendicular to the ground; the height of the upper end of the gate in the vertical direction is controlled by a computer; A groove is opened on the ground at the entrance end of the door frame; The negative pressure assembly includes a grid plate installed at the upper notch of the groove, a housing installed at the lower notch of the groove, an air multiplier tube installed inside the housing, an air outlet pipe provided on the housing, and a negative pressure pump provided at the end of the air outlet pipe.

[0008] Furthermore, the gate is L-shaped, and an electric telescopic rod and a set of guiding telescopic rods are arranged side by side on the gate. The tops of the electric telescopic rod and the guiding telescopic rods are both fixed on a baffle plate whose plate surface is parallel to the plate surface of the gate. The air inlet and outlet of the air multiplier tube are respectively at the upper and lower ends in the vertical direction. The air multiplier tube is connected to the output end of a centrifugal air compressor outside the housing through an air inlet pipe. A check valve is also provided at the output end of the negative pressure pump, and a filtering device is provided at the end of the check valve.

[0009] Furthermore, when the electric telescopic rod is in a fully contracted state, the height from the upper end of the baffle plate to the ground is in the range of [0.7m, 1.2m]. When the electric telescopic rod is in a fully extended state, the height from the upper end of the baffle plate to the ground is in the range of (1.2m, 1.6m]. A Tesla check valve structure is provided inside the check valve, and an activated carbon filter element is provided inside the filtering device.

[0010] Furthermore, the second driving assembly includes a vertical driving assembly, a horizontal driving assembly, and an angular pitching driving assembly. The vertical driving assembly includes track plates provided on both sides of the exit end of the door frame, a first slider slidably connected to the track plates, a screw rod rotatably connected to the track plates and screwed to the first slider, and a servo motor provided at the end of the track plates and driving the screw rod to spin. The horizontal driving assembly includes a guide rail erected between the two first sliders, a second slider slidably connected to the guide rail, and a stepping motor provided on the second slider. A gear installed on the output shaft of the stepping motor meshes with a rack on the guide rail. The angular pitching driving assembly includes a rotating seat provided at the lower end of the second slider and a rotating motor provided on the rotating seat and driving the camera to rotate. The central axis of the output shaft of the rotating motor is parallel to the travel direction of the guide rail.

[0011] Furthermore, a face detection module, a face recognition module, a body temperature detection module, a UI module, a room temperature automatic calibration module, a ranging algorithm, and a compensation algorithm are provided in the CPU. A face detection algorithm, a face recognition algorithm, a live body detection algorithm, and an ornament recognition algorithm are configured in the GPU. The storage module includes a cache unit and a memory unit.

[0012] Furthermore, an auxiliary monitor is provided at the top of the door frame, and the shooting angle of the auxiliary monitor faces one end of the entrance side of the door frame; a display screen is further provided at the top of one end of the entrance side of the door frame, and both the auxiliary monitor and the display screen are controlled by a computer; a monitor is also signal-connected to the computer.

[0013] Furthermore, the computer is connected to the cloud server through a dedicated secure channel.

[0014] Furthermore, a marking line for cooperating with the auxiliary monitor is drawn on the ground on one side of the entrance end of the door frame.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by adding a door frame and a guardrail, the door frame is provided with a first driving component for adjusting the height of the gate, an auxiliary monitor and a display screen are further provided at the top of the door frame, a second driving component for adjusting the spatial position and shooting angle of the camera is also provided on the door frame, the shooting end of the camera is provided with a high-definition camera module, an infrared camera module and a laser ranging module, and a CPU, a GPU, a storage module and a wireless module are embedded in the camera. The CPU is provided with a face detection module, a face recognition module, a body temperature detection module, a UI module, a room temperature automatic calibration module, a ranging algorithm and a compensation algorithm. The GPU is configured with a face detection algorithm, a face recognition algorithm, a live body detection algorithm and an ornament recognition algorithm.

[0016] In this way, the computer can perform preliminary detection on passing personnel through the auxiliary monitor (detection items include but are not limited to: whether there are parallel passing personnel, live body detection, the height of the waiting-to-pass personnel, whether the waiting-to-pass personnel are within the marking line, etc.); then the computer makes the baffle on the gate move to a specified height through the electric telescopic rod (the height of the upper end of the baffle is lower than the shoulder of the waiting-to-pass personnel, so as to ensure that the camera can completely capture the face of the waiting-to-pass personnel), and at the same time the computer commands the second driving motor to make the shooting end of the camera face the face of the waiting-to-pass personnel; then the forehead temperature of the waiting-to-pass personnel is measured through the infrared camera module on the camera (in this process, the computer measures the straight-line distance between the infrared camera module and the forehead of the waiting-to-pass personnel through the laser ranging module, and then brings the measured distance into the compensation algorithm to compensate and correct the temperature value measured by the infrared camera module). If the measured body temperature is normal, the computer commands the gate to rotate and open. If the measured body temperature is abnormal, the computer commands the gate to remain closed, and at the same time the computer commands the display screen and the monitor to alarm; when the waiting-to-pass personnel pass normally, the computer immediately commands the gate to close, and at the same time the electric telescopic rod and the second driving component are reset.

[0017] Thus, the effects of accurate temperature measurement, reduction of labor costs, reduction of infection probability and avoidance of negative temperature measurement are achieved. Description of the Drawings

[0018] Figure 1 is the perspective view of the first perspective of the present invention; Figure 2 is the exploded view of the door frame, the gate and the camera from the second perspective of the present invention; Figure 3 is the exploded view of the second driving component and the camera from the third perspective of the present invention; Figure 4 is Figure 1 the enlarged view of area A in Figure 5 is the internal module relationship diagram of the CPU and the GPU in the present invention; Figure 6 is the exploded view of the negative pressure component from the fourth perspective of the present invention; Figure 7 is the sectional view of the filtering device in the present invention; Figure 8 is the sectional view of the air multiplier in the present invention; Figure 9 is the sectional view of the one-way pipe in the present invention; The reference numerals in the figure respectively represent: 1 - door frame; 2 - gate; 3 - camera; 4 - high-definition camera module; 5 - infrared camera module; 6 - laser ranging module; 7 - CPU; 8 - GPU; 9 - storage module; 10 - wireless module; 11 - computer; 12 - guardrail; 13 - hinge seat; 14 - rotary solenoid valve; 15 - electric telescopic rod; 16 - guiding telescopic rod; 17 - baffle; 18 - track plate; 19 - first slider; 20 - screw; 21 - servo motor; 22 - guide rail; 23 - second slider; 24 - stepping motor; 25 - gear; 26 - rack; 27 - rotating seat; 28 - rotating motor; 29 - auxiliary monitoring; 30 - display screen; 31 - monitor; 32 - marking line; 33 - grid plate; 34 - housing; 35 - air multiplier; 36 - air outlet pipe; 37 - negative pressure pump; 38 - centrifugal air compressor; 39 - one-way pipe; 40 - filtering device; 41 - activated carbon filter element. Detailed Embodiments

[0019] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention will be further described below with reference to the drawings and embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0020] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the present invention is not limited by the limitations of the specific embodiments disclosed in the following specification.

[0021] A face temperature measurement terminal for automatically capturing faces according to this embodiment, refer to Figure 1-9 : It includes a door frame 1, a gate 2 arranged on the door frame 1, and a camera 3 arranged on the door frame 1.

[0022] The first driving component, the second driving component, and the camera 3 are all controlled by a computer 11. In this embodiment, the computer 11 uses a portable laptop because a portable laptop has obvious portability advantages compared to a desktop computer.

[0023] The computer 11 is connected to the cloud server through a dedicated secure channel, which can effectively ensure that the communication information between the computer 11 and the cloud server is not stolen or leaked. (I) A first driving component for driving the movement of the gate 2 is arranged on the door frame 1.

[0025] (I-I) Gate 2 is provided at the inner ends of the vertical rods of the door frame 1; the first driving component includes a hinge seat 13 arranged on the door frame 1 and a rotary solenoid valve 14 arranged on the hinge seat 13. The output shaft of the rotary solenoid valve 14 is fixedly connected to the gate 2 installed on the hinge seat 13, and the central axis of the output shaft of the rotary solenoid valve 14 is perpendicular to the ground, that is, the two gates 2 on the door frame 1 form a swing gate, and the direction when the gate 2 swings open faces one end of the door frame 1 exit.

[0026] It should be noted that the working states of the two rotary solenoid valves 14 on the door frame 1 are kept synchronized, that is, the states of the two gates 2 always remain the same.

[0027] (I-II) Guardrails 12 are arranged at both ends of the door frame 1, which can effectively regulate the order of people during passage (that is, let the passing people queue up in an orderly manner), and at the same time, the strict queuing order can also improve the travel safety of passengers (because it avoids stampede accidents caused by passenger congestion).

[0028] (I-III) The height of the upper end of the gate 2 in the vertical direction is controlled by the computer 11; such a design allows the computer 11 to adjust the height of the gate 2 according to the height of the passing personnel, so as to facilitate the camera 3 to capture the faces of the personnel to pass without reducing the blocking performance of the gate 2.

[0029] Among them, the implementation method for the adjustable height of the gate 2 is specifically as follows: The gate 2 is in an L shape, and an electric telescopic rod 15 and a set of guiding telescopic rods 16 are arranged side by side on the gate 2. The tops of the electric telescopic rod 15 and the guiding telescopic rods 16 are both fixed on a baffle 17 whose plate surface is parallel to the plate surface of the gate 2. In this embodiment, the number of guiding telescopic rods 16 on each gate 2 is two, and the electric telescopic rod 15 is in the middle of the two guiding telescopic rods 16.

[0030] When the electric telescopic rod 15 is in a fully retracted state, the height of the upper end of the baffle 17 from the ground is within the range of [0.7m, 1.2m]. This is because the age of children with independent walking ability and sufficient cognitive ability is usually around 6 years old, and the height of 6-year-old children is about 1m - 1.3m. Therefore, in order to ensure that the faces of 6-year-old children can be fully exposed (here, exposure means that when the camera 3 is facing the child's face directly, the baffle 17 cannot have any obstruction), it is necessary to subtract the length from the top of the child's head to the neck (usually taken as 0.3m) based on 1m - 1.3m. In this embodiment, when the electric telescopic rod 15 is fully retracted, the height of the upper end of the baffle 17 from the ground is 0.7m.

[0031] When the electric telescopic rod 15 is in a fully extended state, the height of the upper end of the baffle 17 from the ground is within the range of (1.2m, 1.6m]. Because the height of normal adults is generally about 1.6m - 2m, if the height of the baffle 17 is too low, it will cause tall passengers to step over the gate 2. In this embodiment, when the electric telescopic rod 15 is fully extended, the height of the upper end of the baffle 17 from the ground is 1.6m. (Two) The door frame 1 is also provided with a second driving component for driving the movement of the camera 3. The shooting end of the camera 3 is provided with a high-definition camera module 4, an infrared camera module 5, and a laser ranging module 6. The inside of the camera 3 is embedded with a CPU 7, a GPU 8, a storage module 9, and a wireless module 10.

[0033] Among them, the high-definition camera module 4 is used to shoot the faces of passengers so as to facilitate the capture and analysis of the facial features of passengers.

[0034] Among them, the infrared camera module 5 is used to measure the body temperature of passengers. It should be noted that the light spot of the infrared rays emitted by the infrared camera module 5 on the passenger's body is located on the forehead of the passenger.

[0035] Among them, the laser ranging module 6 is used to accurately measure the straight-line distance between the infrared camera module 5 and the passenger's forehead. In this embodiment, in order to ensure that the laser emitted by the laser ranging module 6 will not cause damage to the passenger's body, the wavelength range of the laser emitted by the laser ranging module 6 is 8μm - 14μm.

[0036] (Two - One) The second driving assembly includes a vertical driving assembly, a horizontal driving assembly, and an angular pitch driving assembly.

[0037] The vertical driving assembly includes track plates 18 arranged on both sides of the exit end of the doorframe 1, a first slider 19 slidably connected to the track plates 18, a screw rod 20 rotatably connected to the track plates 18 and screwed to the first slider 19, and a servo motor 21 arranged at the end of the track plates 18 and driving the screw rod 20 to rotate. In this way, the height of the camera 3 in the vertical direction can be adjusted by the vertical driving assembly, so that the height of the camera 3 and the forehead of the person to pass through are the same in the vertical direction. It should be noted that the operating states of the two servo motors 21 in the vertical driving assembly are kept synchronized.

[0038] The horizontal driving assembly includes a guide rail 22 erected between the two first sliders 19, a second slider 23 slidably connected to the guide rail 22, and a stepping motor 24 arranged on the second slider 23. A gear 25 installed on the output shaft of the stepping motor 24 meshes with a rack 26 on the guide rail 22. In this way, the position of the camera 3 in the horizontal direction can be adjusted by the horizontal driving assembly, so that the camera 3 and the forehead of the person to pass through are in the same horizontal direction (the horizontal direction mentioned here is parallel to the passing direction of the doorframe 1).

[0039] The angular pitch driving assembly includes a rotating seat 27 arranged at the lower end of the second slider 23 and a rotating motor 28 arranged on the rotating seat 27 and driving the camera 3 to rotate. The central axis of the output shaft of the rotating motor 28 is parallel to the stroke direction of the guide rail 22. Since the heights of some passengers may exceed the stroke of the vertical driving assembly, the angular pitch driving assembly is needed to adjust the pitch angle of the camera 3, so that the shooting end of the camera 3 is aligned with the forehead of the person to pass through.

[0040] In summary, the function of the second driving assembly is to align the shooting end of the camera 3 with the forehead of the person to pass through.

[0041] (Two-two) The CPU7 is provided with a face detection module (for detecting whether there is a face within the shooting field of view of the high-definition camera module 4), a face recognition module (for recognizing the face captured by the high-definition camera module 4 and determining whether it is a real person, and at the same time identifying the identity of the captured face), a body temperature detection module (for calculating the data captured by the infrared camera module 5 to obtain the forehead temperature of the passenger), a UI module (for the man-machine interaction interface displayed on the computer 11 to facilitate the operation and viewing of the temperature measurement personnel), a room temperature automatic calibration module (for automatically calibrating the infrared camera module 5 to the room temperature when the camera 3 is started, so as to ensure that the difference between the measured body temperature and the actual body temperature is within the allowable error range), a ranging algorithm (for processing the data of the laser ranging module 6 to obtain the straight-line distance between the camera 3 and the passenger's forehead), and a compensation algorithm (for cooperating with the ranging algorithm and the body temperature detection module to compensate and correct the temperature measurement result, so that the final temperature measurement result is consistent with the actual body temperature of the passenger).

[0042] Among them, the compensation algorithm establishes the relationship between the initial temperature value and the compensation temperature value by measuring the actual temperature value of the target multiple times and the initial temperature value of the target measured by infrared light, so as to give the correction compensation formula. The compensation algorithm belongs to the prior art, and its algorithm model can refer to the Chinese patent document CN106017690.

[0043] The GPU8 is configured with a face detection algorithm, a face recognition algorithm, a live detection algorithm, and an ornament recognition algorithm (detecting whether there are obstacles such as masks on the passenger's face. If so, the computer 11 issues a prompt through the display screen 30 to let the passenger remove the mask).

[0044] It should be noted that: Ⅰ. Face detection algorithm: Currently, the face detection algorithms based on deep learning are roughly divided into three categories: The first is the cascade-based face detection algorithm (such as Cascade CNN, MTCNN). Although it has the advantages of fast running speed and moderate detection performance, it can only be applied to scenarios with limited computing power, simple background, and few faces.

[0045] The second is the two-stage face detection algorithm, which is generally based on the Faster-RCNN framework. In the first stage, candidate regions are generated, and then in the second stage, the candidate regions are classified and regressed. Representative methods include Face R-CNN, ScaleFace, and FDNet. Although it has the advantage of high detection accuracy, it still has the prominent disadvantage of slow detection speed.

[0046] The third type is the single-stage face detection algorithm, which is mainly based on the classification and regression of anchors. It is usually optimized based on classical frameworks (such as SSD and RetinaNet). Its detection speed is faster than that of the two-stage method, and its detection performance is better than that of the cascade method. It is an algorithm that balances detection performance and speed.

[0047] In summary, in this embodiment, the face detection algorithm adopts the above-mentioned third method, Retinaface.

[0048] Ⅱ. Face recognition algorithm: Currently, there are three classic face recognition algorithms: The first one is the Eigenface method. The specific steps are as follows: First, a batch of face images are converted into a set of feature vectors, called "Eigenfaces", that is, "feature faces", which are the basic components of the initial training image set. The recognition process is to project a new image onto the eigenface subspace, and the determination and recognition are carried out based on the position of its projection point in the subspace and the length of the projection line. After the image is transformed into another space, images of the same category will converge together, and images of different categories will be farther apart. In the original pixel space, it is difficult to separate images of different categories with a simple line or plane. After being transformed into another space, they can be well separated. The spatial transformation method selected by Eigenfaces is PCA (Principal Component Analysis). By using PCA, the main components of the face distribution are obtained. The specific implementation is to perform eigenvalue decomposition on the covariance matrix of all face images in the training set to obtain the corresponding eigenvectors, and these eigenvectors are the "feature faces". Each eigenvector or feature face is equivalent to capturing or describing a change or characteristic between faces. This means that each face can be represented as a linear combination of these feature faces.

[0049] The second one is Local Binary Patterns (LBP). This is a visual operator used for classification in the field of computer vision. LBP is an operator used to describe the texture features of an image. Its core idea is to use the gray value of the central pixel as a threshold and compare it with its neighborhood to obtain the corresponding binary code to represent the local texture features. LBP extracts local features as the discrimination basis. The significant advantage of the LBP method is that it is insensitive to illumination, but it still does not solve the problems of pose and expression. However, compared with the Eigenface method, the recognition rate of LBP has been greatly improved.

[0050] The third is the Fisherface algorithm, Fisher linear discriminant analysis (LDA): The linear discriminant problem for two classes can be regarded as all samples being projected onto a direction (or a dimensional space), and then a classification threshold is determined in this space. The hyperplane passing through this threshold point and perpendicular to the projection direction is the classification surface. The discrimination idea is to select the projection direction so that the two classes are as far apart as possible after projection, and the samples within the class are as clustered as possible (the between-class variance is the largest, and the within-class variance is the smallest).

[0051] Its definition formula is as follows: ; represents the distance between the projection centers of different classifications, and the larger its value, the better.

[0052] , this formula is called the scatter matrix, representing the scatter value after projection of the same classification, that is, the aggregation degree of the projection points. The smaller its value, the more aggregated the projection points are.

[0053] Combining the above two formulas, with the first formula as the numerator and the second formula as the denominator, we can get: , where the larger the value of , the better the dimensionality reduction performance of

[0054] The specific process of the Fisherfaces method is as follows: PCA dimensionality reduction (performing PCA processing on the original samples to obtain new samples after PCA processing) and LDA feature extraction (using the Fisher linear discriminant method for the samples after dimensionality reduction to determine an optimal projection direction, constructing a one-dimensional feature space (this is called Fisherfaces), projecting the multi-dimensional face images into the Fisherfaces feature space, and forming a set of feature vectors using the within-class sample data. This set of feature vectors represents the features of the face).

[0055] To sum up, in this embodiment, the above-mentioned third Fisherface algorithm is adopted because it combines the advantages of PCA and LDA.

[0056] Ⅲ. Liveness detection algorithm: Liveness detection technologies are generally divided into two categories: The first is collaborative liveness detection (the most common liveness detection method). The specific method is as follows: through collaborative combined actions such as blinking, opening the mouth, shaking the head, nodding, or even reading random numbers, using technologies such as facial key point localization and face tracking to verify whether the user is a real live person operating.

[0057] The second is non-collaborative liveness detection (or silent liveness detection technology). It does not require the user to perform additional actions and can directly distinguish forged face attacks such as paper photos, screen imaging, and face masks.

[0058] Non-collaborative liveness detection is generally divided into three technical routes: infrared images, 3D structured light, and RGB images according to different imaging sources. These three routes have their own advantages and disadvantages according to different application scenarios.

[0059] ① For liveness detection of infrared images, an infrared camera is required. Infrared images filter out light in a specific wavelength band and are inherently resistant to screen-based fake face attacks. Whether it is visible light or infrared light, they are essentially electromagnetic waves. Object imaging is related to the reflection characteristics of its surface material. The reflection characteristics of a real human face and attack media such as paper, screen, and three-dimensional mask are different, so the imaging effects are also different. This surface material difference is more obvious in the reflection of infrared waves. When a face on the screen appears in front of the infrared camera, the infrared imaging shows only a white expanse, and even the face cannot be displayed, so the attack cannot succeed.

[0060] ② 3D structured light liveness detection uses a structured light / TOF depth camera, introducing the concept of "depth information", which can easily distinguish fake face attacks of 2D media such as paper photos and screens. 3D structured light requires a 3D camera. In this way, when taking a picture of a human face, 3D data of the face area can be obtained, and further analysis can be performed based on these data to finally determine whether the face is from a live person or a non-live person. The non-live range that 3D structured light can detect is relatively wide, including photos and videos on electronic screens, photos printed on different materials (including bending, folding, cutting, and punching, etc.). The key to this detection path lies in how to select the most discriminative features based on the 3D face data of live and non-live people to train a classifier, and use the trained classifier to distinguish live and non-live people.

[0061] ③ RGB monocular liveness detection only requires a common RGB camera. By analyzing the collected moiré patterns, imaging deformities, reflectivity and other portrait flaws, the recognition information required for liveness detection can be obtained, and the accuracy of recognition is ensured through multi-dimensional recognition bases.

[0062] In summary: The detection cost of infrared image liveness detection is medium, with excellent defense capabilities against attacks using screens and paper, and good to medium defense capabilities against mask attacks; the cost of 3D structured light liveness detection is the highest, with the best effect and excellent defense capabilities against attacks using screens, paper, and masks; the detection cost of RGB images is relatively low, with good defense capabilities against attacks using screens and paper, and average defense capabilities against mask attacks. Therefore, in this embodiment, considering ensuring the detection performance of the product of the present invention while also controlling costs, a method combining two detection methods of RGB and infrared images is selected.

[0063] Ⅳ. Ornament recognition algorithm: Face recognition is a biometric identification technology that identifies a person's identity based on the person's facial feature information. Cameras or webcams are used to collect images or video streams containing human faces, and the faces in the images are automatically detected and tracked. Subsequently, a series of related technologies for the detected faces are carried out, which is usually also called portrait recognition or facial recognition. Although it has a development history of many years, face recognition technology is still affected by various factors such as lighting, perspective, occlusion, and age during the actual application process; among them, ornaments worn on the face by personnel (such as masks, sunglasses, face towels, headscarves, bangs, etc.) need to first detect whether there are ornaments that block the facial features of the personnel. If so, the personnel are prompted to remove these ornaments.

[0064] In this embodiment, a detection method based on PCA analysis is adopted, which includes two key steps: in the analysis stage, the occluded face image is projected onto the face feature space, and the face is reconstructed using the projection coefficients; in the detection stage, the occluded face image is compared with the reconstructed face image. The greater the difference, the greater the possibility of being determined as occluded. The occluded area is estimated based on the difference between the reconstructed face and the original occluded face.

[0065] The storage module 9 includes a cache unit and a memory unit.

[0066] Among them, the memory unit is used to store the face data of passing passengers, and upload it to the cloud server through a dedicated secure channel, and then access the public security system through the cloud server, so as to facilitate the epidemic prevention department to query the movement trajectories of suspected cases, and at the same time facilitate the epidemic prevention department to query the dense personnel related to the suspected cases. (Three) An auxiliary monitor 29 is also provided at the top of the doorframe 1. The shooting angle of the auxiliary monitor 29 faces one end of the entrance side of the doorframe 1, and the auxiliary monitor 29 is used to detect whether there are passengers at the entrance of the doorframe 1.

[0068] A display screen 30 (for prompting whether the passenger is allowed to pass) is also provided at the top of one end of the entrance side of the doorframe 1. Both the auxiliary monitor 29 and the display screen 30 are controlled by the computer 11.

[0069] A display 31 is also signal-connected to the computer 11. Among them, the display 31 faces the passengers, so that the staff on the other side of the guardrail 12 can conveniently see the temperature measurement results of the passengers.

[0070] On the ground on one side of the entrance end of the doorframe 1, there is also a marking line 32 that cooperates with the auxiliary monitor 29. In this way, the computer 11 can detect whether the passenger stands within the marking line 32 through the auxiliary monitor 29. If so, the computer 11 instructs the second driving component and the camera 3 to work; otherwise, both the second driving component and the camera 3 are in the standby state (wherein, in the standby state: the first slider 19 in the second driving component is at the uppermost end of the track plate 18, the second slider 23 is in the middle of the guide rail 22, and the camera 3 is horizontal; in the first driving component, the electric telescopic rod 15 is in the maximum extended state; the gates 2 are all in the closed state). (Four) A groove is formed in the ground at the entrance end of the doorframe 1, and the groove is located between the marking line 32 and the doorframe 1.

[0072] The negative pressure component includes a grid plate 33 installed at the upper slot opening of the groove, a housing 34 installed at the lower slot opening of the groove, an air multiplier tube 35 installed inside the housing 34, an air outlet pipe 36 provided on the housing 34, and a negative pressure pump 37 provided at the end of the air outlet pipe 36. Moreover, the suction port and the air outlet of the air multiplier tube 35 are respectively at the upper and lower ends in the vertical direction. The air multiplier tube 35 is connected to the output end of a centrifugal air compressor 38 outside the housing 34 through an air inlet pipe.

[0073] In this way, through the cooperation of the housing 34, the air multiplier tube 35, the centrifugal air compressor 38, and the negative pressure pump 37, the air above the grid plate 33 can be sucked, so as to remove the air exhaled by the passengers standing on the grid plate 33 and the surrounding air from top to bottom (that is, a low-pressure area will be generated above the grid plate 33 when the negative pressure component works), thus avoiding the aerosol (containing pathogenic bacteria) ejected outward by the passengers who are being identified and temperature-measured by the camera 3 (the passengers in this state have taken off their masks) due to actions such as breathing, sneezing, and exhaling from splashing on the people or equipment around them, thereby avoiding cross-infection.

[0074] There is a circular crack with a width of only 1.3 millimeters on the inner side of the body of the air multiplier tube 35. Between these small cracks, while the air flow clings to the inner wall, due to the Coanda effect, it drives the surrounding air to flow about 15 times and then "blows" out a refreshing cool breeze with a speed of up to 35 kilometers per hour (that is, it adopts the principle of a jet pump (injector). By using a high-speed moving air jet to generate a negative pressure area behind the nozzle to suck the surrounding air).

[0075] In addition, the negative pressure component can also clean the soles of passengers to a certain extent, that is, suck out a certain amount of dust on the soles.

[0076] A one-way pipe 39 with a Tesla one-way valve structure inside is also provided at the output end of the negative pressure pump 37; this can effectively ensure that the sucked gas flows unidirectionally in the negative pressure component (that is, prevent the backflow of dirty gas); there are no moving parts in Tesla, no internal mechanical movement is required, the spatial structure is used to push the gas flow, and the physical structure is used to accelerate the gas flow, reducing the energy loss of the gas during transportation. Therefore, it can make up for the shortcomings of traditional valves that are easily damaged due to the need for movable parts.

[0077] A filtering device 40 with an activated carbon filter element 41 inside is provided at the end of the one-way pipe 39; in this way, the gas to be discharged by the negative pressure pump 37 can be filtered and disinfected through the activated carbon filter element 41, so as to remove harmful substances such as dust and aerosol in the gas.

[0078] The above are only the preferred embodiments of the present invention, and are not limitations on the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A face temperature measurement method for automatically capturing human faces based on the Fisherface algorithm, including a face temperature measurement terminal for automatically capturing human faces, characterized in that: The face temperature measurement terminal includes a door frame, a gate provided on the door frame, and a camera provided on the door frame; A negative pressure component is provided at the ground at the entrance end of the door frame; A first driving component for driving the movement of the gate is provided on the door frame; A second driving component for driving the movement of the camera is further provided on the door frame. The shooting end of the camera is provided with a high-definition camera module, an infrared camera module, and a laser ranging module. The interior of the camera is embedded with a CPU, a GPU, a storage module, and a wireless module; The first driving component, the second driving component, and the camera are all controlled by a computer; Gateways are provided at the inner ends of the rods of the door frame perpendicular to the ground; the first driving component includes a hinge seat provided on the door frame and a rotary solenoid valve provided on the hinge seat. The output shaft of the rotary solenoid valve is fixedly connected to the gate on the mounting hinge seat, and the central axis of the output shaft of the rotary solenoid valve is perpendicular to the ground; the height of the upper end of the gate in the vertical direction is controlled by a computer; A groove is formed in the ground at the entrance end of the door frame; The negative pressure component includes a grid plate installed at the upper notch of the groove, a housing installed at the lower notch of the groove, an air multiplier installed inside the housing, an air outlet pipe provided on the housing, and a negative pressure pump provided at the end of the air outlet pipe; The face temperature measurement method for automatically capturing faces based on the Fisherface algorithm includes: A face detection module, a face recognition module, a body temperature detection module, a UI module, a room temperature automatic calibration module, a ranging algorithm, and a compensation algorithm provided in the CPU; And a face detection algorithm, a face recognition algorithm, a living body detection algorithm, and an ornament recognition algorithm configured in the GPU; The storage module includes a cache unit and a memory unit; The face recognition algorithm is the Fisherface algorithm.

2. The face temperature measurement method for automatically capturing human faces based on the Fisherface algorithm according to claim 1, wherein Guardrails are provided at both ends of the door frame.

3. The face temperature measurement method for automatically capturing human faces based on the Fisherface algorithm according to claim 1, wherein The gate is L-shaped, and an electric telescopic rod and a set of guiding telescopic rods are provided side by side on the gate. The tops of the electric telescopic rod and the guiding telescopic rods are both fixed on a baffle plate whose plate surface is parallel to the plate surface of the gate; The air inlets and outlets of the air multiplier are respectively at the upper and lower ends in the vertical direction. The air multiplier is connected to the output end of a centrifugal air compressor outside the housing through an air inlet pipe. A one-way pipe is further provided at the output end of the negative pressure pump, and a filtering device is provided at the end of the one-way pipe.

4. The face temperature measurement method for automatically capturing a human face based on the Fisherface algorithm according to claim 3, characterized in that, When the electric telescopic rod is in a fully retracted state, the height of the upper end of the baffle plate from the ground is in the range of [0.7m, 1.2m]. When the electric telescopic rod is in a fully extended state, the height of the upper end of the baffle plate from the ground is in the range of (1.2m, 1.6m]; A Tesla one-way valve structure is provided inside the one-way pipe, and an activated carbon filter element is provided inside the filtering device.

5. The face temperature measurement method for automatically capturing a human face based on the Fisherface algorithm according to claim 1, characterized in that, The second driving component includes a vertical direction driving component, a horizontal direction driving component, and an angle pitching driving component; The vertical direction driving component includes track plates provided on both sides at the outlet end of the door frame, a first slider slidably connected to the track plates, a screw rod rotatably connected to the track plates and screwed to the first slider, and a servo motor provided at the end of the track plate and driving the screw rod to spin; The horizontal driving component includes a guide rail installed between two first sliders, a second slider slidably connected to the guide rail, and a stepping motor disposed on the second slider. A gear mounted on the output shaft of the stepping motor meshes with a rack on the guide rail; The angular pitch driving component includes a rotating seat disposed at the lower end of the second slider and a rotating motor disposed on the rotating seat and driving the camera to rotate. The central axis of the output shaft of the rotating motor is parallel to the stroke direction of the guide rail.

6. The face temperature measurement method for automatically capturing a human face based on the Fisherface algorithm according to claim 1, characterized in that, An auxiliary monitor is further provided at the top of the door frame, and the shooting angle of the auxiliary monitor faces one end of the entrance side of the door frame; a display screen is further provided at the top of one end of the entrance side of the door frame. Both the auxiliary monitor and the display screen are controlled by a computer; a monitor is also signal-connected to the computer.

7. The face temperature measurement method for automatically capturing a human face based on the Fisherface algorithm according to claim 7, wherein The computer is connected to the cloud server through a dedicated secure channel.

8. An automatic face temperature measurement method for capturing faces based on the Fisherface algorithm according to claim 7, characterized in that, A marking line matching the auxiliary monitor is further drawn on the ground on one side of the entrance end of the door frame.

9. The face temperature measurement method for automatically capturing a human face based on the Fisherface algorithm according to claim 1, wherein The definition formula of the Fisherface algorithm is as follows: ； Represents the distance of different classification projection centers, and the larger the value, the better; , this formula is called the hash value, representing the hash value after the same classification projection, that is, the aggregation degree of the projection points. The smaller its value, the more aggregated the projection points are; Combining the above two formulas, with the first formula as the numerator and the second formula as the denominator, we can obtain: , where the larger the value of the better the dimensionality reduction performance is. The specific process includes determining the optimal projection direction and determining the classification threshold in this direction. The specific process is as follows: Perform PCA processing on the original samples to obtain new samples after PCA processing. Use the Fisher linear discriminant method for the samples after dimensionality reduction to determine an optimal projection direction, construct a one-dimensional feature space, project the multi-dimensional face images into the Fisherfaces feature space, and form a set of feature vectors using the within-class sample data. This set of feature vectors represents the features of the face.

Images