Intelligent morning inspection device

Abstract

The application relates to an intelligent morning inspection device, which comprises a device main body, a display screen arranged on the device main body, a first image sensor arranged on the device main body, a structured light sensor and an information collection table, a sealing cover arranged on the information collection table and used for forming a closed space on the information collection table, a second image sensor located in the closed space and a controller, the controller is in data interaction with the first image sensor, the second image sensor and the structured light sensor, and is used for carrying out identity recognition, living body authentication and identification of an abnormal area in a hand image of a feature object, the first image sensor is used for collecting facial information of the feature object, the structured light sensor is used for projecting dot matrix information to two-dimensional information of the face of the feature object, and the second image sensor is used for collecting the hand image of the feature object. The intelligent morning inspection device disclosed by the application realizes the uniqueness of personnel identity and the faster detection passing speed by combining the living body recognition and the local comparison detection mode.

Description

Technical Field

[0001] This application relates to the field of security inspection technology, and in particular to an intelligent morning inspection device. Background Technology

[0002] The canteen morning check equipment is an intelligent terminal designed specifically for the health and hygiene screening of canteen kitchen staff before they start work. Its purpose is to replace manual morning checks and achieve standardized, digitalized and traceable testing. The current testing scope includes identity verification, health screening, data management and control linkage, etc.

[0003] The morning check process can be roughly divided into two parts: identity recognition and security check. Considering the working environment of the canteen, the current identity recognition is mostly done by facial recognition (non-contact), while image inspection is used to find problems (wounds, diseases, stains, etc.).

[0004] The main problems are the use of false information for check-in, inaccurate AI recognition, and slow processing speed. False information for check-in will lead to distortion of the image inspection method, and inaccurate AI recognition will directly affect the accuracy of the judgment, causing unnecessary review work. Slow processing speed will lead to backlog of people, because the morning check-in in the canteen is a scene of concentrated people, and slow processing speed will lead to backlog of people, which will directly affect the progress of subsequent work. Summary of the Invention

[0005] This application provides an intelligent morning inspection device that combines liveness detection and local comparison detection to achieve uniqueness of personnel identity and faster detection speed.

[0006] The above-mentioned objective of this application is achieved through the following technical solution:

[0007] This application provides an intelligent morning health check device, comprising:

[0008] The main body of the equipment is equipped with a display screen.

[0009] Both the first image sensor and the structured light sensor are located on the main body of the device. The first image sensor is used to collect facial information of the feature object, and the structured light sensor is used to project dot matrix information onto the two-dimensional facial information of the feature object.

[0010] The information collection station is located on the main body of the equipment;

[0011] A sealing cover is installed on the information acquisition platform to form a closed space on the information acquisition platform;

[0012] The second image sensor is located on the information acquisition platform and inside the enclosed space. The second image sensor is used to acquire images of the hand of the feature object.

[0013] The controller interacts with the first image sensor, the second image sensor, and the structured light sensor to perform identity recognition, liveness authentication, and identification of abnormal areas in hand images of feature objects.

[0014] In one possible implementation of this application, the structured light sensor projects dot matrix information onto the two-dimensional facial information of the feature object, including:

[0015] Obtain a real random sequence, which is derived from the clock and / or facial information of the feature object acquired by the first image sensor;

[0016] A two-dimensional information projection dot matrix is ​​generated using a real random number sequence. The two-dimensional information projection dot matrix includes multiple array regions, and each array region includes MxN information elements.

[0017] Replace at least one information element in each array region with an identity code to obtain dot matrix information;

[0018] The dot matrix information is projected onto the face of the feature object.

[0019] In one possible implementation of this application, when the feature object performs identity recognition and liveness authentication, dot matrix information is projected onto the two-dimensional information of the feature object's face multiple times;

[0020] Each time two-dimensional information is projected onto a dot matrix, a new real random number sequence is obtained.

[0021] One possible implementation of this application also includes:

[0022] Create an image sequence belonging to the feature object;

[0023] Sequentially acquire facial images of feature objects over time and determine the imaging quality of the facial images;

[0024] Mark facial images of feature objects that meet imaging quality standards as qualified images;

[0025] Determine the type of qualified image, which includes qualified images for recognition and qualified images for authentication;

[0026] Place qualified images into an image sequence, which includes at least one qualified image for recognition and at least one qualified image for authentication;

[0027] Specifically, when the number of qualified recognition images or qualified authentication images in the image sequence meets the requirements, the qualified recognition images or qualified authentication images obtained further in the time series are discarded.

[0028] In one possible implementation of this application, identifying abnormal regions in a hand image includes:

[0029] Obtain a comparative hand image based on the facial information of the feature object;

[0030] The acquired hand images are compared with the comparison hand images to obtain the comparison results;

[0031] Based on the comparison results, the hand image is divided into regions to obtain normal and abnormal regions;

[0032] When an abnormal area is found, an inspection image is generated using the hand image and the abnormal area.

[0033] In one possible implementation of this application, comparing the acquired hand image with a comparison hand image includes:

[0034] The fixed feature points of the hand image are extracted using a feature point matching algorithm, and the fixed feature points of the hand image are compared.

[0035] The hand image is fine-tuned using fixed feature points of the hand image and fixed feature points of the comparison hand image so that the hand image and the comparison hand image coincide.

[0036] Calculate the pixel difference between corresponding pixels in the hand image and the comparison hand image to obtain a pixel difference map;

[0037] Generate comparison results based on the pixel difference map;

[0038] Specifically, when calculating the pixel difference between corresponding pixels in the hand image and the comparison hand image, multiple monochrome channels are used to process the hand image and the comparison hand image to obtain a pixel difference map belonging to each monochrome channel.

[0039] In one possible implementation of this application, after obtaining the abnormal region, the method further includes verifying the abnormal region and deleting the verified abnormal region.

[0040] Verification of abnormal areas includes:

[0041] Determine the edge contour of the abnormal region;

[0042] Use the edge contours of abnormal regions to extract analysis content from hand images and extract reference content from comparative hand images;

[0043] Compare and analyze the grayscale, texture, and shape of the content and the reference content;

[0044] If any of the grayscale comparison, texture comparison, or shape comparison fails, the abnormal region is deleted; otherwise, the abnormal region is retained.

[0045] In one possible implementation of this application, the grayscale values ​​of the comparison analysis content and the reference content include:

[0046] The analysis content and reference content are divided into multiple cells, and each cell contains the same number of pixels;

[0047] Calculate the average pixel value of the cell;

[0048] Compare the average pixel values ​​of adjacent cells and assign a value to the cell based on the comparison result;

[0049] Calculate the similarity between the analyzed content and the reference content and provide the grayscale comparison results.

[0050] In one possible implementation of this application, the textures of the comparative analysis content and the reference content include:

[0051] The analysis content and reference content are divided into multiple cells, and each cell contains the same number of pixels;

[0052] Calculate the average pixel value of the cell;

[0053] Identify the line segments to compare and calculate the difference between adjacent cells on the line segments;

[0054] Generate a comparison sequence based on the differences; the comparison sequence can be either a difference sequence or a quadratic difference sequence.

[0055] Calculate the similarity between two comparison sequences on the same comparison line segment, where the two comparison sequences belong to the analysis content and the reference content, respectively;

[0056] Calculate the similarity between two comparison sequences on all compared line segments and provide the texture comparison results.

[0057] In one possible implementation of this application, the morphological comparison between the comparative analysis content and the reference content includes:

[0058] Obtain the grayscale histogram of the analyzed content and the grayscale histogram of the reference content;

[0059] Select outlier points and suspected outlier points adjacent to the outlier points on the grayscale histogram of the analyzed content.

[0060] Extract outlier values ​​from the analysis content to obtain the first extraction result, and calculate the quantity of the first extraction result.

[0061] Suspected outlier data points are extracted from both the analyzed content and the reference content to obtain a second extraction result. The similarity of the second extraction result is then calculated.

[0062] The morphological comparison results are given based on the quantity value of the first extraction result and the similarity of the second extraction result. Attached Figure Description

[0063] Figure 1 This is a structural schematic diagram of an intelligent morning inspection device from a first-person perspective, provided in this application.

[0064] Figure 2 This is a structural schematic diagram of an intelligent morning inspection device from a left-hand perspective, as provided in this application.

[0065] Figure 3 This is a schematic diagram of the control principle of a controller provided in this application.

[0066] Figure 4 This is a schematic diagram of a two-dimensional information projection dot matrix including an array region provided in this application.

[0067] Figure 5 This is a schematic diagram of placing a qualified image into an image sequence, as provided in this application.

[0068] Figure 6 This is a schematic diagram illustrating the principle of grayscale comparison analysis content and reference content provided in this application.

[0069] Figure 7 This is a schematic diagram illustrating the principle of obtaining a comparison sequence based on comparing line segments, as provided in this application.

[0070] Figure 8 This is a schematic diagram illustrating the principle of determining abnormal numerical points provided in this application.

[0071] In the diagram, 1 is the main body of the device, 2 is the first image sensor, 3 is the structured light sensor, 4 is the information acquisition platform, 5 is the sealing cover, 6 is the second image sensor, 7 is the controller, and 11 is the display screen. Detailed Implementation

[0072] The technical solutions in this application will be further described in detail below with reference to the accompanying drawings.

[0073] This application discloses an intelligent morning health check device; please refer to [link / reference]. Figures 1 to 3 In some examples, the intelligent morning inspection device disclosed in this application includes a device body 1, a first image sensor 2, a structured light sensor 3, an information acquisition platform 4, a sealing cover 5, a second image sensor 6, a controller 7, and a display screen 11.

[0074] Specifically, the display screen 11 is located on the main body 1 of the device. Its function is to display the facial information of the first image sensor 2 used to collect the facial information of the feature object during identity recognition and liveness authentication, and at the same time provide auxiliary information, such as providing position borders to help the feature object adjust its position and displaying prompts (look up, turn left, turn right).

[0075] Both the first image sensor 2 and the structured light sensor 3 are fixedly installed on the main body 1 of the device. The first image sensor 2 is used to collect facial information of the feature object, and the structured light sensor 3 is used to project dot matrix information onto the two-dimensional facial information of the feature object. The two work together to complete identity recognition and liveness authentication.

[0076] Specifically, the first image sensor 2 acquires facial images of the feature object, and sends the acquired facial images to the controller 7 for comparison to complete identity recognition; the dot matrix information projected by the structured light sensor 3 can perform two-dimensional / three-dimensional identification, avoiding the use of pictures or videos for identity recognition of the feature object.

[0077] An information acquisition station 4 is also installed on the main body 1, and a sealing cover 5 is installed on the information acquisition station 4 to form a closed space on the information acquisition station 4. The second image sensor 6 is installed on the information acquisition station 4 and located inside the closed space.

[0078] The information collection station 4 serves as a temporary support for the hand of the feature object, and the sealing cover 5 provides a relatively stable working environment for the second image sensor 6. This is because strong light, backlight, shadows, and flickering lights at the entrance of the canteen will directly affect the judgment results. Oil stains, water stains, and fog will cause image blurring and texture loss. High temperature and high humidity will affect the clarity and focus of the second image sensor 6.

[0079] The function of the sealing cover 5 is to provide a closed environment in which parameters such as light, temperature and humidity are relatively stable. It can also be equipped with auxiliary means such as supplemental lighting and fan drying.

[0080] The controller 7 interacts with the first image sensor 2, the second image sensor 6, and the structured light sensor 3 to perform identity recognition, liveness authentication, and identification of abnormal areas in hand images of feature objects.

[0081] Specifically, the functions of controller 7 are mainly divided into two parts: identifying abnormal areas and identity recognition (liveness authentication). The following is a further introduction to these two parts.

[0082] The specific method by which the structured light sensor 3 projects dot matrix information onto the two-dimensional facial information of the feature object is as follows:

[0083] Obtain a real random sequence, which is derived from the clock and / or facial information of the feature object acquired by the first image sensor 2;

[0084] A two-dimensional information projection dot matrix is ​​generated using a real random number sequence. The two-dimensional information projection dot matrix includes multiple array regions, and each array region includes MxN information elements.

[0085] Replace at least one information element in each array region with an identity code to obtain dot matrix information;

[0086] The dot matrix information is projected onto the face of the feature object.

[0087] The core purpose of the above method is to avoid using images or videos for facial recognition of the subject. It should be understood that the general method of facial comparison is to collect the facial information of the subject by the first image sensor 2. If an image or video is used to replace the real subject, the first image sensor 2 cannot distinguish it. Therefore, auxiliary actions (opening the mouth, turning the head, and looking down) are involved. However, the auxiliary actions can be implemented by video, which still poses certain security risks. At the same time, there are also problems of prolonging the operation time and repeated operation after failure, making it difficult to apply to densely populated scenarios.

[0088] If a dot projection method is used, there is a possibility that infrared noise in sunlight will completely obscure the dot matrix. In addition, because the dot matrix image is too small, a high-precision first image sensor 2 and complex algorithms are required for processing, which will also prolong the processing time and increase the recognition time.

[0089] In addition, the single method of dot matrix projection makes it susceptible to forgery.

[0090] This application uses a real random number sequence to generate a two-dimensional information projection dot matrix, and then projects this two-dimensional information projection dot matrix onto the face of the feature object. Specifically, the real random number sequence comes from a clock (e.g., the clock inside the controller 7) and / or facial information of the feature object collected by the first image sensor 2 (e.g., random quantities such as the values ​​of a few randomly collected pixels), or the current temperature value or noise value (requiring the configuration of a corresponding sensor), etc. The real random number sequence is untraceable.

[0091] Two-dimensional information projection arrays include multiple array regions, such as Figure 4 As shown, each array region includes MxN information elements, which are composed of pixels. Here, the two-dimensional information projection dot matrix is ​​regarded as a QR code. At this time, an identity code also needs to be added to the two-dimensional information projection dot matrix.

[0092] In some possible implementations, the real random number sequence is extended by replication, and a fixed number of real random numbers are assigned to each array region in sequence.

[0093] The specific process involves replacing at least one information element in each array region with an identity code. The identity code represents the location of the array region, facilitating rapid comparison processing by the controller 7. It should be understood that when there are multiple array regions, it is necessary to determine the location of the array regions through successive comparisons. However, with the addition of the identity code, the identity code can be read directly.

[0094] It should be noted that when projecting two-dimensional information dot matrix, the area of ​​the projection region must be larger than the face of the feature object. The reason for adding an identity code to each array region is that it is impossible to determine which array region's identity code has a higher recognition rate.

[0095] In some implementations, when a feature object performs identity recognition and liveness authentication, dot matrix information is projected onto the two-dimensional information of the feature object's face multiple times;

[0096] Each time two-dimensional information is projected onto a dot matrix, a new real random number sequence is obtained.

[0097] For image acquisition, the following method is used:

[0098] Create an image sequence belonging to the feature object;

[0099] Sequentially acquire facial images of feature objects over time and determine the imaging quality of the facial images;

[0100] Mark facial images of feature objects that meet imaging quality standards as qualified images;

[0101] Determine the type of qualified image, which includes qualified images for recognition and qualified images for authentication;

[0102] Place qualified images into an image sequence, which includes at least one qualified image for recognition and at least one qualified image for authentication;

[0103] Specifically, when the number of qualified recognition images or qualified authentication images in the image sequence meets the requirements, the qualified recognition images or qualified authentication images obtained further in the time series are discarded.

[0104] Here, a continuous image acquisition method is used to capture images of the feature object. The facial images of the feature object are first judged for their imaging quality, and the indicators here include sharpness, noise, brightness, contrast, and exposure uniformity.

[0105] There are several ways to improve image sharpness, including applying Laplacian variance to the image, applying Laplacian filtering to the image, calculating the variance (the larger the variance, the sharper the edges), Brenner gradient (calculating the sum of squared differences between adjacent pixels), and Tenegrad gradient (calculating the sum of the magnitudes of Sobel x and y gradients).

[0106] The noise was handled by sliding 3×3 and 5×5 windows across the entire image; calculating the standard deviation within each window; and averaging the standard deviations of all windows.

[0107] The brightness is processed by calculating the average gray value of the entire image; the higher the result, the brighter the image.

[0108] Contrast is handled by calculating the standard deviation of image gray levels; the larger the standard deviation, the higher the contrast.

[0109] The method for processing exposure uniformity is to divide the image into blocks and calculate the difference in brightness between each block. The smaller the difference, the more uniform the exposure.

[0110] Each facial image of the feature object acquired sequentially in the time series needs to be judged. Qualified facial images are used as qualified images. Qualified images are first classified into recognition qualified images and authentication qualified images.

[0111] Then it is placed onto the image sequence, according to the following rules:

[0112] Place the qualified images into the image sequence, which includes at least one qualified image for recognition and at least one qualified image for authentication, such as... Figure 5 As shown.

[0113] Here, the number of qualified images for recognition and qualified images for authentication is generally set to two, which means there are four empty slots in the image sequence. After all four slots are occupied, the facial images of the feature objects are sequentially acquired in the time series.

[0114] The advantage of this method is that it can guarantee the acquisition of qualified recognition and authentication images, and can dynamically adjust the time consumed. In a stable environment, it is generally possible to complete the acquisition of qualified recognition and authentication images in 5-8 shots.

[0115] The specific method for identifying abnormal areas in hand images is as follows:

[0116] Obtain a comparative hand image based on the facial information of the feature object;

[0117] The acquired hand images are compared with the comparison hand images to obtain the comparison results;

[0118] Based on the comparison results, the hand image is divided into regions to obtain normal and abnormal regions;

[0119] When an abnormal area is found, an inspection image is generated using the hand image and the abnormal area.

[0120] This method determines whether there are abnormal areas on a hand image by comparing today's image with yesterday's image. The advantage of this comparison method is its speed. It should be understood that AI recognition has a certain degree of accuracy, but it requires massive amounts of standard data for training and repeated verification. Most current AI models are general-purpose models with insufficient generalization ability and a certain probability of misidentification. In addition, considering local security and local computing power deployment, most models are simplified models that have been distilled, further reducing their recognition ability.

[0121] The recognition method used in this application uses yesterday's image as a reference. The advantage of this method is that it can directly identify abnormal areas. At the same time, it uses hand images and abnormal areas to generate inspection images to improve detection efficiency.

[0122] Specifically, after one feature object completes the acquisition of a hand image, other feature objects can continue the work. The purpose of the inspection image is to provide it to the verification personnel, who will further confirm any abnormal areas. If there is a safety risk in the abnormal area, appropriate measures will be taken. If there is no safety risk in the abnormal area, the feature object can enter the area to work.

[0123] The method for comparing the acquired hand image with the comparison hand image is as follows:

[0124] The fixed feature points of the hand image are extracted using a feature point matching algorithm, and the fixed feature points of the hand image are compared.

[0125] The hand image is fine-tuned using fixed feature points of the hand image and fixed feature points of the comparison hand image so that the hand image and the comparison hand image coincide.

[0126] Calculate the pixel difference between corresponding pixels in the hand image and the comparison hand image to obtain a pixel difference map;

[0127] Generate comparison results based on the pixel difference map;

[0128] Specifically, when calculating the pixel difference between corresponding pixels in the hand image and the comparison hand image, multiple monochrome channels are used to process the hand image and the comparison hand image to obtain a pixel difference map belonging to each monochrome channel.

[0129] The above method determines the comparison result by the difference between pixels. For example, if the difference between pixels is required to be greater than 3, then the location of pixels with a difference greater than 3 will be shown in the comparison result. At the same time, considering the color tendency of different types of abnormal areas, the hand image and the comparison hand image are converted into monochrome images (red image, green image, blue image) and compared separately. At this time, a pixel difference map belonging to each monochrome channel is obtained, that is, three monochrome channel pixel difference maps. These three monochrome channel pixel difference maps are fused, and the pixel differences at the same position are accumulated.

[0130] After obtaining the abnormal region, the process also includes verifying the abnormal region and deleting the abnormal region that passes the verification.

[0131] Verification of abnormal areas includes:

[0132] Determine the edge contour of the abnormal region;

[0133] Use the edge contours of abnormal regions to extract analysis content from hand images and extract reference content from comparative hand images;

[0134] Compare and analyze the grayscale, texture, and shape of the content and the reference content;

[0135] If any of the grayscale comparison, texture comparison, or shape comparison fails, the abnormal region is deleted; otherwise, the abnormal region is retained.

[0136] The grayscale method for comparing and analyzing the content and the reference content is as follows:

[0137] The analysis content and reference content are divided into multiple cells, and each cell contains the same number of pixels;

[0138] Calculate the average pixel value of the cell;

[0139] Compare the average pixel values ​​of adjacent cells and assign a value to the cell based on the comparison result;

[0140] Calculate the similarity between the analyzed content and the reference content and provide the grayscale comparison results.

[0141] The advantage of this method is that it simplifies the comparison content and, by assigning values ​​to cells, converts the analysis and reference content into a display using 0s and 1s, such as... Figure 6 As shown, the black area represents 1 and the white area represents 0. At this time, the comparison only needs to determine whether the values ​​at the same position are equal or unequal, which greatly shortens the processing time.

[0142] When the ratio of the number of unequal results to the total number of results is greater than a set value, the grayscale comparison result fails.

[0143] The textures of the compared and analyzed content and the reference content are analyzed as follows:

[0144] The analysis content and reference content are divided into multiple cells, and each cell contains the same number of pixels;

[0145] Calculate the average pixel value of the cell;

[0146] Identify the line segments to compare and calculate the difference between adjacent cells on the line segments;

[0147] Generate a comparison sequence based on the differences; the comparison sequence can be either a difference sequence or a quadratic difference sequence.

[0148] Calculate the similarity between two comparison sequences on the same comparison line segment, where the two comparison sequences belong to the analysis content and the reference content, respectively;

[0149] Calculate the similarity between two comparison sequences on all compared line segments and provide the texture comparison results.

[0150] The comparison segment here is a segment of fixed length, and the comparison segment is in the same position in both the analysis content and the reference content, such as... Figure 7 As shown, the similarity between two comparative sequences on the same comparative line segment is calculated. The methods used here include Euclidean distance, Pearson correlation coefficient, cosine similarity, and Manhattan distance. These methods all calculate a numerical value, so a reference value needs to be set here.

[0151] The advantage of this approach is that it offers greater flexibility in the deployment of line segments. For example, it can employ dense deployment (where all analysis content and reference content participate in the analysis) or interval deployment (where only a portion of the analysis content and reference content participates in the analysis). The spacing of interval deployment can be adjusted according to actual needs, with larger spacing resulting in less data processing.

[0152] When the similarity between two comparison sequences is less than a set reference value, the texture comparison result fails.

[0153] The comparative analysis content and the reference content are compared in the following ways:

[0154] Obtain the grayscale histogram of the analyzed content and the grayscale histogram of the reference content;

[0155] Select outlier points and suspected outlier points adjacent to the outlier points on the grayscale histogram of the analyzed content.

[0156] Extract outlier values ​​from the analysis content to obtain the first extraction result, and calculate the quantity of the first extraction result.

[0157] Suspected outlier data points are extracted from both the analyzed content and the reference content to obtain a second extraction result. The similarity of the second extraction result is then calculated.

[0158] The morphological comparison results are given based on the quantity value of the first extraction result and the similarity of the second extraction result.

[0159] This method uses grayscale histograms to identify outlier values. Outlier values ​​are grayscale values ​​that appear only on the analysis content. Suspected outlier values ​​that are adjacent to outlier values ​​are grayscale values ​​that appear on both the analysis content and the reference content.

[0160] like Figure 8 As shown, the grayscale histogram is represented by a curve. The grayscale histogram of the analyzed content is represented by a dashed line, while the grayscale histogram of the reference content is represented by a solid line. The non-overlapping areas on the two lines represent outlier points (multiple). The quantity value of the first extraction result corresponds to an area value, and the similarity of the second extraction result refers to the ratio of the area of ​​the non-overlapping area to the area of ​​the entire region.

[0161] Specifically, the suspected abnormal value points adjacent to the abnormal value point are generally the value points to the left and right of the abnormal value point, and the number of suspected abnormal value points is generally 2-3 times that of the abnormal value point.

[0162] The main purpose here is to determine whether the suspected outlier data points are related to the outlier data points. If they are related, the similarity score of the second extraction result will decrease; otherwise, the similarity score of the second extraction result will increase.

[0163] The first extraction result is an area value, and the second result is a ratio value. Both of these values ​​also need to be set with a reference value. When the area value or ratio value is greater than the corresponding reference value, the comparison result between the analysis content and the reference content is "not passed".

[0164] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An intelligent morning check-up device, characterized in that, include: The main body of the equipment (1) is equipped with a display screen (11). The first image sensor (2) and the structured light sensor (3) are both located on the main body (1) of the device. The first image sensor (2) is used to collect facial information of the feature object, and the structured light sensor (3) is used to project dot matrix information onto the two-dimensional facial information of the feature object. Information collection station (4) is located on the main body of the equipment (1); A sealing cover (5) is installed on the information collection station (4). The sealing cover (5) is used to form a closed space on the information collection station (4). The second image sensor (6) is installed on the information acquisition station (4) and located inside the enclosed space. The second image sensor (6) is used to acquire images of the hand of the feature object. The controller (7) interacts with the first image sensor (2), the second image sensor (6) and the structured light sensor (3) to identify the feature object, perform liveness authentication and identify abnormal areas in the hand image; The structured light sensor (3) projects dot matrix information onto the two-dimensional facial information of the feature object, including: Obtain a real random number sequence, which is derived from the facial information of the feature object collected by the clock and / or the first image sensor (2); A two-dimensional information projection dot matrix is ​​generated using a real random number sequence. The two-dimensional information projection dot matrix includes multiple array regions, and each array region includes MxN information elements. Replace at least one information element in each array region with an identity code to obtain dot matrix information; Projecting dot matrix information onto the face of the feature object; When performing identity recognition and liveness authentication on a feature object, dot matrix information is projected onto the two-dimensional information of the feature object's face multiple times; Each time two-dimensional information is projected onto a dot matrix, a new real random number sequence is obtained.

2. The intelligent morning inspection device according to claim 1, characterized in that, Also includes: Create an image sequence belonging to the feature object; Sequentially acquire facial images of feature objects over time and determine the imaging quality of the facial images; Mark facial images of feature objects that meet imaging quality standards as qualified images; Determine the type of qualified image, which includes qualified images for recognition and qualified images for authentication; Place qualified images into an image sequence, which includes at least one qualified image for recognition and at least one qualified image for authentication; Specifically, when the number of qualified recognition images or qualified authentication images in the image sequence meets the requirements, the qualified recognition images or qualified authentication images obtained further in the time series are discarded.

3. The intelligent morning inspection device according to claim 1, characterized in that, Identifying abnormal regions in hand images includes: Obtain a comparative hand image based on the facial information of the feature object; The acquired hand images are compared with the comparison hand images to obtain the comparison results; Based on the comparison results, the hand image is divided into regions to obtain normal and abnormal regions; When an abnormal area is found, an inspection image is generated using the hand image and the abnormal area.

4. The intelligent morning inspection device according to claim 3, characterized in that, The process of comparing the acquired hand images with the comparison hand images includes: The fixed feature points of the hand image are extracted using a feature point matching algorithm, and the fixed feature points of the hand image are compared. The hand image is fine-tuned using fixed feature points of the hand image and fixed feature points of the comparison hand image so that the hand image and the comparison hand image coincide. Calculate the pixel difference between corresponding pixels in the hand image and the comparison hand image to obtain a pixel difference map; Generate comparison results based on the pixel difference map; Specifically, when calculating the pixel difference between corresponding pixels in the hand image and the comparison hand image, multiple monochrome channels are used to process the hand image and the comparison hand image to obtain a pixel difference map belonging to each monochrome channel.

5. The intelligent morning inspection device according to claim 4, characterized in that, After obtaining the abnormal region, the process also includes verifying the abnormal region and deleting the abnormal region that passes the verification. Verification of abnormal areas includes: Determine the edge contour of the abnormal region; Use the edge contours of abnormal regions to extract analysis content from hand images and extract reference content from comparative hand images; Compare and analyze the grayscale, texture, and shape of the content and the reference content; If any of the grayscale comparison, texture comparison, or shape comparison fails, the abnormal region is deleted; otherwise, the abnormal region is retained.

6. The intelligent morning inspection device according to claim 5, characterized in that, The grayscale values ​​for the comparative analysis content and the reference content include: The analysis content and reference content are divided into multiple cells, and each cell contains the same number of pixels; Calculate the average pixel value of the cell; Compare the average pixel values ​​of adjacent cells and assign a value to the cell based on the comparison result; Calculate the similarity between the analyzed content and the reference content and provide the grayscale comparison results.

7. The intelligent morning inspection device according to claim 5, characterized in that, The textures of the comparative analysis content and the reference content include: The analysis content and reference content are divided into multiple cells, and each cell contains the same number of pixels; Calculate the average pixel value of the cell; Identify the line segments to compare and calculate the difference between adjacent cells on the line segments; Generate a comparison sequence based on the differences; the comparison sequence can be either a difference sequence or a quadratic difference sequence. Calculate the similarity between two comparison sequences on the same comparison line segment, where the two comparison sequences belong to the analysis content and the reference content, respectively; Calculate the similarity between two comparison sequences on all compared line segments and provide the texture comparison results.

8. The intelligent morning inspection device according to claim 5, characterized in that, The comparative analysis of the content and the comparison of the reference content includes the following: Obtain the grayscale histogram of the analyzed content and the grayscale histogram of the reference content; Select outlier points and suspected outlier points adjacent to the outlier points on the grayscale histogram of the analyzed content. Extract outlier values ​​from the analysis content to obtain the first extraction result, and calculate the quantity of the first extraction result. Suspected outlier data points are extracted from both the analyzed content and the reference content to obtain a second extraction result. The similarity of the second extraction result is then calculated. The morphological comparison results are given based on the quantity value of the first extraction result and the similarity of the second extraction result.

Images