An Internet of Things-based intelligent customer data supervision system and method

Abstract

The present invention discloses an intelligent supervision system and method for customer data based on the Internet of Things, which relates to the technical field of data supervision, and includes an image acquisition module, a three-dimensional reconstruction module, a result marking module, and an image cutting module; this intelligent supervision system and method for customer data based on the Internet of Things converts customer data into three-dimensional information, and through the comparison of the three-dimensional information, the changes in customer data can be intuitively found, realizing the intelligent supervision of customer data. At the same time, this setting only requires the three-dimensional reconstruction of the single-sided image to realize the comparison of the three-dimensional information of the corresponding surface with the reference data, avoiding increasing the graphic processing burden and the processing speed of the system for image processing by generating the three-dimensional graphic as a whole when constructing the three-dimensional graphic of the customer data image. At the same time, by cutting the single-sided image marked as unqualified data by the result marking module into single images for processing, the data operation burden and the processing accuracy during the data processing process can be further reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of data supervision, and particularly relates to an intelligent supervision system and method for customer data based on the Internet of Things. Background Art

[0002] Information supervision technology refers to the use of various technical means and methods to monitor and manage information, ensuring the security, accuracy, and legality of information. With the rapid development of information technology, information plays an increasingly important role in social and economic life, and society has higher and higher requirements for information security and supervision.

[0003] Currently, when conducting customer data supervision, three-dimensional image information of customer sensitive data is collected through data integration, and abnormal markings are made by comparing the three-dimensional graphics of the detection data with the reference three-dimensional images of standard customer sensitive data. This will cause an increase in the graphic processing burden when generating the three-dimensional graphics of customer data through the overall data, and increase the processing speed of the system for image processing. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent supervision system and method for customer data based on the Internet of Things to solve the above deficiencies in the prior art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An intelligent supervision system for customer data based on the Internet of Things, including an image acquisition module, a three-dimensional reconstruction module, a result marking module, an image cutting module, a single-image recognition module, a result comparison module, and a result identification module, characterized in that:

[0006] The image acquisition module is used to acquire customer data images, reference data three-dimensional images, and perspective images of reference data based on Internet of Things data;

[0007] The three-dimensional reconstruction module is used to perform single-sided three-dimensional reconstruction based on customer data images;

[0008] The result marking module is used to mark single-sided images with the same comparison result between the single-sided three-dimensional reconstruction result of the single-sided image and the corresponding surface three-dimensional information of the collected reference data as qualified data, and mark single-sided images with different comparison results between the single-sided three-dimensional reconstruction result of the single-sided image and the corresponding surface three-dimensional information of the collected reference data as unqualified data;

[0009] The image cutting module is used to cut the single-sided images marked as unqualified data by the result marking module into single images;

[0010] The single-image recognition module is used to perform single-image recognition on the single images cut by the image cutting module, so as to identify the corresponding single-pixel stroke lines in the single images;

[0011] The result comparison module is used to compare the three-dimensional reconstruction result of the single-sided image with the three-dimensional information of the corresponding surface of the collected reference data, and to compare the identified single-pixel stroke lines with the corresponding positions of the reference data image;

[0012] The result identification module is used to identify the single-pixel stroke lines with different comparison results at the corresponding positions of the identified single-pixel stroke lines and the reference data image as abnormal positions.

[0013] Furthermore, the three-dimensional reconstruction module includes:

[0014] An image extraction module, which extracts the main body of the picture in the single-sided image by using a mask operation;

[0015] An image enhancement module, which uses an exponential adjustment method to perform enhancement processing on the extracted image;

[0016] A surface model restoration module, which reconstructs the surface model of the single-sided image by using gradient field information

[0017] A model macroscopic structure optimization module, which optimizes the macroscopic geometric shape of the single-sided image model by using the image medial axis curve constraint;

[0018] A single-sided image macroscopic structure optimization module, which optimizes the single-sided image model by using a macroscopic model, fuses the single-sided image model and the macroscopic structure model, and generates the final three-dimensional model result of the single-sided image;

[0019] A single-sided image three-dimensional model output module, which is used to output the three-dimensional model of the single-sided image.

[0020] Furthermore, the single-image recognition module includes:

[0021] A hole filling module, which can fill the small holes in the drawing strokes by filling the background with white pixel points, performing a NOT operation on the filled image, and performing an XOR operation on the original image and the image after the NOT operation;

[0022] An error elimination module, which eliminates the isolated small dots still existing in the stroke area through an opening operation;

[0023] A grayscale processing module, which obtains a grayscale image by processing with three different weights of RGB;

[0024] A stroke smoothing module, which uses median filtering to smooth the outline of the strokes and remove noise points;

[0025] A stroke measurement module, which incorporates the original morphological stroke algorithm to construct a thinning algorithm to obtain single-pixel stroke lines without burrs.

[0026] An intelligent supervision method for customer data based on the Internet of Things, comprising the following steps:

[0027] S1, Customer data collection, marking the single-sided images of the collected customer data, where the single-sided images include but are not limited to front views, side views, top views, and bottom views;

[0028] S2, Performing single-sided three-dimensional reconstruction on the single-sided images;

[0029] S3, Comparing the single-sided three-dimensional reconstruction results of the single-sided images with the three-dimensional information of the corresponding surfaces of the collected reference data;

[0030] S4, Marking the single-sided images with the same comparison results in step S3 as qualified data, and marking the single-sided images with different comparison results in step S3 as unqualified data;

[0031] S5, Cutting the single-sided images marked as unqualified data in step S4 into single-frame images;

[0032] S6, Performing single-frame image recognition on the single-frame images completed in cutting in step S5, so as to recognize the corresponding single-pixel stroke lines in the single-frame images;

[0033] S7, Comparing the recognized single-pixel stroke lines with the corresponding positions of the reference data images, and marking the single-pixel stroke lines with different comparison results as abnormal positions.

[0034] Further, the single-sided three-dimensional reconstruction of the single-sided images specifically includes the following steps:

[0035] A1, Inputting the image;

[0036] A2, Performing a masking operation to extract the main body of the image in the single-sided image;

[0037] A3, Image enhancement, where the exponentiation adjustment method is used to perform enhancement processing on the extracted image;

[0038] A4, Surface model restoration, specifically including the following steps:

[0039] A41, Surface model reconstruction;

[0040] A42, Gradient field information extraction, where the pixel value intensity in the enhanced image is used as a clue to calculate the gradient field information of the blades in the image;

[0041] A43, Discrete set processing;

[0042] A44, Surface model restoration, using the gradient field information to reconstruct the surface model of the single-sided image;

[0043] A5. Macroscopic structure optimization of the model. Optimize the macroscopic geometry of the single-sided image model using the image medial axis curve constraint. The specific steps are as follows:

[0044] A51. Macroscopic geometry optimization;

[0045] A52. Fitting the image medial axis curve, where the medial axis curve is taken from the single-sided image;

[0046] A53. Shape function constraint;

[0047] A54. Macroscopic structure of the model;

[0048] A6. Macroscopic structure optimization of the single-sided image. Finally, optimize the single-sided image model with the macroscopic model, fuse the single-sided image model and the macroscopic structure model, and generate the final three-dimensional model result of the single-sided image;

[0049] A7. Output the three-dimensional model of the single-sided image.

[0050] Furthermore, the single-image recognition includes the following working steps:

[0051] B1. To fill the fine holes inside the filled strokes, use the flood fill algorithm to fill the background with white pixel points. Perform image NOT operation on the filled image, and perform image XOR operation on the original image and the image after the NOT operation. Through these three steps, the small holes inside the drawing strokes can be completely filled;

[0052] B2. There are still isolated small dots in the stroke area, which are eliminated by opening operation;

[0053] B3. Process the stroke image with three different weights of RGB to obtain a grayscale image. The calculation formula for grayscale processing is as follows:

[0054] Gray = 0.29 * R + 0.578 * G + 0.123 * B

[0055] where R, G, and B represent the values of the three primary colors - red, green, and blue - of the stroke segmentation image, and "Gray" represents the grayscale value;

[0056] B4. To retain more details of the strokes, use median filtering to smooth the contour of the strokes and remove noise points. The specific steps are as follows:

[0057] B41. The pixel values of the stroke image are divided into levels [1, 2,..., l], ni which is used to represent the number of image pixel values. Therefore, the calculation formula for the total pixel value of the stroke image is as follows:

[0058] ;

[0059] B42, where the frequency of a single pixel in the image pi is calculated as follows:

[0060] ;

[0061] B43, Define two variables as local variables w0 and variable w1 the sum of the frequency values, and their relationship is shown in the following formula:

[0062] ;

[0063] B44, then the foreground pixel frequency u0 and the background pixel frequency u1 of the stroke area image are as follows:

[0064]

[0065] where ;

[0066] B5, Use the Otsu method to perform binary processing on the image;

[0067] B6, In order to measure the stroke length, finally incorporate the original morphological stroke algorithm to construct a thinning algorithm to obtain a single-pixel stroke line without burrs;

[0068] B7, Obtain the result of the single image to be recognized by combining the obtained single-pixel stroke lines.

[0069] Furthermore, the specific steps of the single-sided image cutting the single image include:

[0070] C1, Use the connected component analysis method to perform the first segmentation on the single-sided image to obtain each block;

[0071] C2: Project each block to obtain the length of each block;

[0072] C3: Define the width of a single character and denote it as L. Take this as the threshold, and transfer the block with a projection length greater than the threshold to C4 for the second segmentation;

[0073] C4: Use the initial point selection process to find the initial point, adopt the left descent algorithm, and use the improved penetration rule of the penetration process for segmentation. If the penetration is unsuccessful, it is considered that the penetration has entered a closed loop, and transfer to C5, where the specific steps of the initial point selection process are C41, and the specific steps of the improved penetration process are C42;

[0074] C41(1). First, at the inflection points of the connected digits or the places connected to them, pixels will gather more than other places; these pixels represent the features of the digits, and these are the key points; select the leftmost and rightmost two points, which must be on both sides of the connected digits and represent the features of the two digits respectively;

[0075] C41(2). Assume BB is a connected component, and the distance between the leftmost key point and the rightmost key point is BD; take 1 / 3 of BD to intersect with the connected component. Since the number of intersection points is greater than one, select the topmost point and name it point a;

[0076] C41(3). If the right point or the upper right point of point a is a black point, then this point will replace point a and become the new point a until its right point or upper right point is a white point; at this time, the latest point a is the extreme point of this area;

[0077] C41(4). If the horizontal position of the latest point a is less than 2 / 3 of BD, it means the latest point a is the extreme point of the left part of the connected component; then take the latest point a as the starting point of the penetration process; otherwise, if the horizontal position of the latest point a is greater than 2 / 3 of BD, it means the initial point a is on the connection line of adjacent digits; then take the initial point a as the starting position of the penetration process; C42. The improvement rules of the penetration process are described as follows:

[0078] Rule 1: if n1 = 0 and n3 = 0, then next_pixel = n3

[0079] Rule 2: if n1 = 0 and n2 = 1, then next_pixel = n1

[0080] Rule 3: if n1 = 0, then next_pixel = n1

[0081] Rule 4: if n6 = 1 and n7 = 0, then next_pixel = n7

[0082] Among them, 0 is a white pixel, 1 is a black pixel, n1 is the pixel point on the left side of the current pixel point position, n2 is the pixel point on the lower left side of the current pixel point position, n3 is the pixel point on the lower side of the current pixel point position, n4 is the pixel point on the lower right side of the current pixel point position, n5 is the pixel point on the right side of the current pixel point position, n6 is the pixel point on the upper left side of the current pixel point position, and n7 is the pixel point on the upper side of the current pixel point position;

[0083] C5: The upper part adopts the left - descending algorithm and stops immediately after the first penetration. The trajectory is recorded as P1. The lower part adopts the upper - left - shifting algorithm, which is the same as before, and stops immediately after the first penetration. The trajectory is recorded as P2. The intersection points of the trajectories and the touched number are recorded as p1 and p2. A shortest path connecting p1 and p2 is found on the contour of the touched number, and the connecting trajectory is recorded as P3;

[0084] C6: Finally, a complete image - cutting trajectory includes P1, P2, and P3. Compared with the prior art, an intelligent supervision system and method for customer data based on the Internet of Things provided by the present invention convert customer data into three - dimensional information, and intuitively discover changes in customer data through the comparison of three - dimensional information, realizing the intelligent supervision of customer data. At the same time, only the single - sided image needs to be three - dimensionally reconstructed in this way to realize the comparison of three - dimensional information of the corresponding surface with the reference data, avoiding increasing the graphic - processing burden and the processing speed of the system for image processing by generating a three - dimensional graph as a whole when constructing the three - dimensional graph of the customer - data image. At the same time, the single - sided images marked as unqualified data by the result - marking module are cut into single - frame images for processing, further reducing the data - operation burden and the processing accuracy during the data - processing process. Brief Description of the Drawings

[0085] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0086] Figure 1 It is a block diagram of the system structure provided by the embodiment of the present invention. Detailed Embodiments

[0087] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the following will further introduce the present invention in detail with reference to the drawings.

[0088] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0089] In addition, the terms "first" and "second" are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined. In addition, the terms "mounted", "connected" and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention may be understood according to specific circumstances.

[0090] Example embodiments will be described more fully hereinafter with reference to the accompanying drawings, but the example embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0091] In the case of no conflict, the various embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0092] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0093] The terms used herein are for the purpose of describing particular embodiments only and are not intended to limit the present disclosure. As used herein, the singular forms "a" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms "comprises" and / or "consists of" are used in this specification, it specifies the presence of the stated features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0094] The embodiments described herein may be described with reference to the plan views and / or cross-sectional views by means of the ideal schematic diagrams of the present disclosure. Accordingly, the example illustrations may be modified according to manufacturing techniques and / or tolerances. Therefore, the embodiments are not limited to the embodiments shown in the drawings, but include modifications of configurations formed based on manufacturing processes. Therefore, the regions illustrated in the drawings have schematic attributes, and the shapes of the regions shown in the drawings illustrate the specific shapes of the regions of the elements, but are not intended to be restrictive.

[0095] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in common dictionaries shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein.

[0096] Please refer to Figure 1 , an Internet of Things-based intelligent customer data supervision system, including an image acquisition module, a three-dimensional reconstruction module, a result marking module, an image cutting module, a single-image recognition module, a result comparison module, and a result identification module, characterized in that:

[0097] The image acquisition module is used to acquire customer data images, reference data three-dimensional images, and perspective images of reference data based on Internet of Things data;

[0098] The three-dimensional reconstruction module is used to perform single-sided three-dimensional reconstruction based on the customer data image;

[0099] The result marking module is used to mark the single-sided images with the same comparison result between the single-sided three-dimensional reconstruction result of the single-sided image and the three-dimensional information of the corresponding surface of the collected reference data as qualified data, and mark the single-sided images with different comparison results between the single-sided three-dimensional reconstruction result of the single-sided image and the three-dimensional information of the corresponding surface of the collected reference data as unqualified data;

[0100] The image cutting module is used to cut the single-sided images marked as unqualified data by the result marking module into single images;

[0101] The single-image recognition module is used to perform single-image recognition on the single images cut by the image cutting module, so as to recognize the corresponding single-pixel stroke lines in the single images;

[0102] The result comparison module is used to compare the single-sided three-dimensional reconstruction result of the single-sided image with the three-dimensional information of the corresponding surface of the collected reference data, and compare the recognized single-pixel stroke lines with the corresponding positions of the reference data images;

[0103] The result identification module is used to identify the single-pixel stroke lines with different comparison results between the recognized single-pixel stroke lines and the corresponding positions of the reference data images as abnormal positions.

[0104] Collect customer data images, reference data three-dimensional images, and images of each perspective of the reference data through the image acquisition module, mark the single-sided images of the collected customer data, perform single-sided three-dimensional reconstruction on the single-sided images through the three-dimensional reconstruction module, and compare the single-sided three-dimensional reconstruction results of the single-sided images with the corresponding surface three-dimensional information of the collected reference data through the result comparison module. Mark the single-sided images with the same comparison results as qualified data, and mark the single-sided images with different comparison results as unqualified data. Cut the single-sided images marked as unqualified data into single images through the image cutting module, and perform single-image recognition on the cut single images through the single-image recognition module, so as to identify the corresponding single-pixel stroke lines in the single images. Compare the recognized single-pixel stroke lines with the corresponding positions of the reference data images through the result comparison module, and mark the single-pixel stroke lines with different comparison results as abnormal positions through the result identification module. By converting customer data into three-dimensional information, the comparison of three-dimensional information can intuitively discover the changes in customer data, realizing the intelligent supervision of customer data. At the same time, such a setting only requires performing three-dimensional reconstruction on single-sided images to achieve the comparison with the corresponding surface three-dimensional information of the reference data, avoiding increasing the graphic processing burden and the processing speed of the system for image processing by generating a three-dimensional graphic as a whole when constructing the three-dimensional graphic of the customer data image. At the same time, by processing the single-sided images marked as unqualified data by the result marking module into single images, the data operation burden and the processing accuracy during data processing can be further reduced.

[0105] The three-dimensional reconstruction module includes:

[0106] An image extraction module, which extracts the main body of the image in the single-sided image by using a masking operation;

[0107] An image enhancement module, which performs enhancement processing on the extracted image by using an exponential adjustment method;

[0108] A surface model restoration module, which reconstructs the surface model of the single-sided image by using gradient field information

[0109] A model macroscopic structure optimization module, which optimizes the macroscopic geometry of the single-sided image model by using the image central axis curve constraint;

[0110] A single-sided image macroscopic structure optimization module, which optimizes the single-sided image model by using the macroscopic model, fuses the single-sided image model and the macroscopic structure model, and generates the final single-sided image three-dimensional model result;

[0111] A single-sided image three-dimensional model output module, which is used to output the single-sided image three-dimensional model.

[0112] The single-image recognition module includes:

[0113] A hole filling module, which can fill small holes inside the drawing strokes by filling the background with white pixels, performing an image NOT operation on the filled image, and performing an image XOR operation on the original image and the image after the NOT operation.

[0114] An error elimination module, which eliminates isolated small dots still existing in the stroke area through opening operation.

[0115] A grayscale processing module, which obtains a grayscale image by processing with three different weights of RGB.

[0116] A stroke smoothing module, which uses median filtering to smooth the contour of the strokes and remove noise points.

[0117] A stroke measurement module, which incorporates the original morphological stroke algorithm to construct a thinning algorithm to obtain a single-pixel stroke line without burrs.

[0118] An intelligent supervision method for customer data based on the Internet of Things, including the following steps:

[0119] S1, Customer data collection, marking the single-sided images of the collected customer data, where the single-sided images include but are not limited to front views, side views, top views, and bottom views.

[0120] S2, Performing single-sided 3D reconstruction on the single-sided images.

[0121] S3, Comparing the single-sided 3D reconstruction results of the single-sided images with the 3D information of the corresponding sides of the collected reference data.

[0122] S4, Marking the single-sided images with the same comparison results in step S3 as qualified data, and marking the single-sided images with different comparison results in step S3 as unqualified data.

[0123] S5, Cutting the single-sided images marked as unqualified data in step S4 into single images; single images can be obtained by cutting the single-sided images along the complete cutting trajectory, and the cut single images are transmitted to step S6 for single-image recognition, so as to recognize the corresponding single-pixel stroke lines in the single images.

[0124] S6, Performing single-image recognition on the single images cut in step S5, so as to recognize the corresponding single-pixel stroke lines in the single images.

[0125] S7. Compare the single-sided 3D reconstruction result of the single-sided image with the 3D information of the corresponding surface of the collected reference data through the result comparison model. When the comparison result between the single-sided 3D reconstruction result of the single-sided image and the 3D information of the corresponding surface of the collected reference data is the same, compare the single-pixel stroke lines identified in step S6 with the corresponding positions of the reference data image through the result comparison model, and mark the single-pixel stroke lines with different comparison results between the single-pixel stroke lines identified in step S6 and the corresponding positions of the reference data image as abnormal positions.

[0126] Among them, by modifying the single-pixel stroke lines marked as abnormal positions, data correction can be achieved, and then the correct 3D graph of the customer data image can be obtained. By converting the customer data into 3D information, the changes in the customer data can be intuitively found through the comparison of 3D information, realizing the intelligent supervision of customer data. At the same time, such a setting only requires 3D reconstruction of the single-sided image to achieve the comparison with the 3D information of the corresponding surface of the reference data, avoiding increasing the graphic processing burden and the processing speed of the system for image processing by generating the 3D graph as a whole when constructing the 3D graph of the customer data image.

[0127] The specific steps for single-sided 3D reconstruction of a single-sided image are as follows:

[0128] A1. Input the image;

[0129] A2. Extract the main body of the picture in the single-sided image through mask operation;

[0130] A3. Image enhancement, in which the exponential adjustment method is used to enhance the extracted image;

[0131] A4. Surface model restoration, specifically including the following steps:

[0132] A41. Surface model reconstruction;

[0133] A42. Gradient field information extraction, in which the pixel value intensity in the enhanced image is used as a clue to calculate the gradient field information of the blades in the image;

[0134] A43. Discrete set processing;

[0135] A44. Surface model restoration, using the gradient field information to reconstruct the surface model of the single-sided image;

[0136] A5. Model macrostructure optimization, using the image medial axis curve constraint to optimize the macro geometric shape of the single-sided image model, specifically including the following steps:

[0137] A51. Macro geometric optimization;

[0138] A52. Image medial axis curve fitting, in which the medial axis curve is taken from the single-sided image;

[0139] A53, Shape function constraint;

[0140] A54, Macrostructure of the model;

[0141] A6, Optimization of the macrostructure of the single-sided image. Finally, the macro model optimizes the single-sided image model, fuses the single-sided image model and the macrostructure model, and generates the final three-dimensional model result of the single-sided image;

[0142] A7, Output the three-dimensional model of the single-sided image.

[0143] Single-image recognition includes the following working steps:

[0144] B1, To fill the fine holes in the filled strokes, use the flood fill algorithm to fill the background with white pixels, perform image NOT operation on the filled image, and perform image XOR operation on the original image and the image after the NOT operation. Through these three steps, the small holes in the drawing strokes can be completely filled;

[0145] B2, There are still isolated small dots in the stroke area, which are eliminated by opening operation;

[0146] B3, The stroke image is processed with three different weights of RGB to obtain a grayscale image. The calculation formula for grayscale processing is as follows:

[0147] Gray = 0.29 * R + 0.578 * G + 0.123 * B

[0148] Where R, G, and B represent the values of the three primary colors - red, green, and blue - of the stroke segmentation image, and "Gray" represents the grayscale value;

[0149] B4, To retain more details of the strokes, use median filtering to smooth the outline of the strokes and remove noise points, which specifically includes the following steps:

[0150] B41, The pixel values of the stroke image are divided into [1, 2,..., l] levels, ni which is used to represent the number of image pixel values. Therefore, the calculation formula for the total pixel value of the stroke image is as follows:

[0151] ;

[0152] B42, Among them, the frequency of a single pixel in the image pi is calculated as follows:

[0153] ;

[0154] B43, Define two variables as local variables w0 and variable w1The sum of the frequency values, and their relationship is shown in the following formula:

[0155] ;

[0156] For B44, the foreground pixel frequency of the stroke area image u0 and the background pixel frequency u1 are as follows:

[0157]

[0158] where ;

[0159] For B5, use the Otsu method to perform binary processing on the image;

[0160] For B6, in order to measure the stroke length, finally integrate the original morphological stroke algorithm to construct a thinning algorithm to obtain a single-pixel stroke line without burrs;

[0161] For B7, obtain the result of the single image to be recognized by combining the obtained single-pixel stroke lines.

[0162] The specific steps for cutting a single-sided image into a single image include:

[0163] C1, use the connected component analysis method to perform the first segmentation on the single-sided image to obtain each block;

[0164] C2: Project each block to obtain the length of each block;

[0165] C3: Define the width of a single character and denote it as L. Take this as the threshold, and transfer the block with a projection length greater than the threshold to C4 for the second segmentation;

[0166] C4: Use the initial point selection process to find the initial point, adopt the left descent algorithm, and use the improved seepage rule in the seepage process for segmentation. If the seepage is unsuccessful, it is considered that the seepage has entered a closed loop and transfer to C5, where the specific steps of the initial point selection process are C41, and the specific steps of the improved seepage process are C42;

[0167] C41(1), first, at the inflection point of the connected digits or the places connected to them, pixels will gather more than other places; these pixels represent the characteristics of the digits, and these are called key points; select the leftmost and rightmost two points, which must be on both sides of the connected digits, and represent the characteristics of these two digits respectively;

[0168] C41(2), assume that BB is a connected component, and the distance between the leftmost key point and the rightmost key point is BD; take 1 / 3 of BD to intersect with the connected component. Since the number of intersection points is greater than one, select the topmost point and name it point a;

[0169] C41(3), if the right or upper - right point of point a is a black point, then this point will replace point a to become the new point a until its right or upper - right point is a white point; at this time, the latest point a is the extreme point of this area;

[0170] C41(4), if the horizontal position of the latest point a is less than 2 / 3 of BD, it means that the latest point a is the extreme point of the left part of the connected component; then take the latest point a as the starting point of the penetration process; otherwise, if the horizontal position of the latest point a is greater than 2 / 3 of BD, it means that the initial point a is on the connection line of adjacent digits; then take the initial point a as the starting position of the penetration process;

[0171] C42, the penetration - process improvement rule is described as:

[0172] Rule 1: if n1 = 0 and n3 = 0, then next_pixel = n3

[0173] Rule 2: if n1 = 0 and n2 = 1, then next_pixel = n1

[0174] Rule 3: if n1 = 0, then next_pixel = n1

[0175] Rule 4: if n6 = 1 and n7 = 0, then next_pixel = n7

[0176] Where 0 represents a white pixel, 1 represents a black pixel, n1 is the pixel point to the left of the current pixel - point position, n2 is the pixel point to the lower - left of the current pixel - point position, n3 is the pixel point below the current pixel - point position, n4 is the pixel point to the lower - right of the current pixel - point position, n5 is the pixel point to the right of the current pixel - point position, n6 is the pixel point to the upper - left of the current pixel - point position, and n7 is the pixel point above the current pixel - point position;

[0177] C5: The upper part uses the left - descending algorithm and stops immediately after the first penetration, and the trajectory is recorded as P1. The lower part uses the upper - left - moving algorithm, which is the same as before, and stops immediately after the first penetration, and the trajectory is recorded as P2. The intersection points of the trajectories and the contacted digits are recorded as p1 and p2. Find the shortest path connecting p1 and p2 on the contour of the contacted digit, and the connecting trajectory is recorded as P3;

[0178] C6: Finally, a complete image - cutting trajectory includes P1, P2, and P3.

[0179] After step C6 is completed, the single-sided image can be cut according to the complete cutting trajectory obtained in step C6 to obtain a single image. The cut single image is transmitted to the single-image recognition module for single-image recognition, so as to recognize the corresponding single-pixel stroke line in the single image. The single-sided three-dimensional reconstruction result of the single-sided image is compared with the three-dimensional information of the corresponding surface of the collected reference data through the result comparison module. When the comparison result of the single-sided three-dimensional reconstruction result of the single-sided image and the three-dimensional information of the corresponding surface of the collected reference data is the same, based on the result comparison module, the single-pixel stroke line recognized in step S6 is compared with the corresponding position of the reference data image, and the single-pixel stroke line with different comparison results between the single-pixel stroke line recognized in step S6 and the corresponding position of the reference data image is marked as an abnormal position. By modifying the single-pixel stroke line marked as an abnormal position, data correction is realized, and then the correct three-dimensional graph of the customer data image is obtained. By converting the customer data into three-dimensional information, the change of the customer data can be intuitively found through the comparison of the three-dimensional information, realizing the intelligent supervision of the customer data. At the same time, such a setting only requires three-dimensional reconstruction of the single-sided image to realize the comparison with the three-dimensional information of the corresponding surface of the reference data, avoiding increasing the graphic processing burden and the processing speed of the system for image processing by generating a three-dimensional graph as a whole when constructing the three-dimensional graph of the customer data image. The above only describes some exemplary embodiments of the present invention by way of illustration. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A customer data intelligent supervision system based on the Internet of Things, comprising an image acquisition module, a three-dimensional reconstruction module, a result marking module, an image cutting module, a single image recognition module, a result comparison module and a result identification module, characterized in that: The image acquisition module is used to acquire customer data images, reference data three-dimensional images, and images of reference data at various viewing angles based on IoT data; The 3D reconstruction module is used to perform single-sided 3D reconstruction based on the customer data image. The 3D reconstruction module includes an image extraction module, an image enhancement module, a surface model recovery module, a model macrostructure optimization module, a single-sided image macrostructure optimization module, and a single-sided image 3D model output module. The image extraction module uses a mask operation to extract the main body of the single-sided image. The image enhancement module uses an exponential adjustment method to enhance the extracted image. The surface model recovery module uses gradient field information to reconstruct the surface model of the single-sided image. The model macrostructure optimization module uses the image central axis curve constraint to optimize the macro-geometric shape of the single-sided image model. The single-sided image macrostructure optimization module fuses the single-sided image model and the macrostructure model to generate a final single-sided image 3D model result. The single-sided image 3D model output module is used to output the single-sided image 3D model. The result marking module is used to mark the single-side 3D reconstruction result of the single-side image and the single-side image with the same comparison result as qualified data, and mark the single-side image with different comparison results as unqualified data; The image cutting module is used to cut the single-sided image marked as unqualified data by the result marking module into a single image; The single image recognition module is used to perform single image recognition on the single image cut by the image cutting module, so as to recognize the corresponding single pixel stroke line in the single image; The result comparison module is used to compare the single-side 3D reconstruction result of the single-side image with the corresponding face 3D information of the collected reference data, and to compare the recognized single-pixel stroke line with the corresponding position of the reference data image; The result identification module is used to compare the identified single-pixel stroke line with the corresponding position of the reference data image, and the single-pixel stroke line with different results is identified as an abnormal position.

2. According to claim 1, a customer data intelligent supervision system based on the Internet of Things is characterized by: The single image recognition module includes a hole filling module, an error elimination module, a grayscale processing module, a stroke smoothing module and a stroke measurement module. The hole filling module fills the background with white pixels, performs image non-operation on the filled image, and performs image XOR operation on the original image and the image after the non-operation operation to fill the small holes in the drawing strokes. The error elimination module eliminates the isolated small dots that still exist in the stroke area through opening operation. The grayscale processing module obtains a grayscale image by processing with three different weights of RGB. The stroke smoothing module uses median filtering to smooth the outline of the stroke and remove noise points. The stroke measurement module integrates the original morphological stroke algorithm to construct a thinning algorithm, so as to obtain a single-pixel stroke line without burrs.

3. A method for intelligent supervision of customer data based on the Internet of Things, which is applicable to the intelligent supervision system for customer data based on the Internet of Things according to any one of claims 1-2, characterized in that: The following steps are involved: S1, collecting customer data, marking the collected single-sided images of customer data, the single-sided images including but not limited to the front view, side view, top view and bottom view; S2, performing single-side 3D reconstruction on the single-side image, wherein the single-side 3D reconstruction on the single-side image specifically comprises the following steps: A1, input image; A2, mask operation to extract the main body of the picture in the single-sided image; A3, image enhancement, using the exponential adjustment method to enhance the extracted image; A4, surface model recovery, specifically includes the following steps: A41, surface model reconstruction; A42, gradient field information extraction, uses the pixel value intensity in the enhanced image as a clue to calculate the gradient field information of the leaves in the image; A43, discrete set processing; A44, surface model recovery, uses gradient field information to reconstruct the surface model of a single-sided image; A5, model macro structure optimization, using the image medial axis curve constraint to optimize the macro geometry of the single-sided image model, specifically including the following steps: A51, macro geometry optimization; A52, image medial axis curve fitting, the medial axis curve is taken from a single-sided image; A53, shape function constraint; A54, model macrostructure; A6, single-sided image macrostructure optimization, integrating the single-sided image model with the macrostructure model to generate the final single-sided image 3D model result; A7, output single-sided image 3D model; S3, comparing the single-side 3D reconstruction result of the single-side image with the corresponding surface 3D information of the collected reference data; S4, marking the single-side images with the same comparison results as qualified data, and marking the single-side images with different comparison results as unqualified data; S5, cutting the single-sided image marked as unqualified data into single images; S6, performing single image recognition on the cut single image, so as to recognize the corresponding single pixel stroke line in the single image; S7, comparing the identified single-pixel stroke line with the corresponding position of the reference data image, and marking the single-pixel stroke line with different comparison results as an abnormal position.

4. The method for intelligent supervision of customer data based on the Internet of Things according to claim 3 is characterized in that: The single image recognition comprises the following working steps: B1, to fill the tiny holes in the strokes, use the flood fill algorithm to fill the background with white pixels, perform image negation on the filled image, and perform image XOR operation on the original image and the image after the negation operation. Through the above three steps, the tiny holes in the drawing strokes are completely filled; B2, there are still isolated small dots in the stroke area, which are eliminated through opening operation; B3, the stroke image is processed with three different weights of RGB to obtain a grayscale image. The calculation formula for grayscale processing is as follows: Gray=0.29*R+0.578*G+0.123*B; Among them, R, G, B represent the values ​​of the three primary colors of the stroke segmentation image - red, green, and blue, and "Gray" represents the grayscale value; B4, in order to retain more stroke details, use median filtering to smooth the stroke outline and remove noise points, which specifically includes the following steps: B41, the pixel values ​​of the stroke image are divided into [1, 2, ..., l] levels, It is used to represent the number of image pixel values. Therefore, the calculation formula for the total pixel value of the stroke image is as follows: ； B42, where the frequency of a single pixel in the image The calculation formula is as follows: ； B43, define two variables as local variables and local variables The sum of the frequency values, the relationship between the two is shown in the following formula: ； B44, the foreground pixel frequency of the stroke area image and background pixel frequency As shown below: ； in ; B5, use the maximum inter-class variance method to perform binary processing on the image; B6, in order to measure the stroke length, the original morphological stroke algorithm is finally integrated to construct a thinning algorithm to obtain a single-pixel stroke line without burrs; B7, the single-pixel stroke lines obtained by combining are used to obtain the result of the single image to be identified.

5. The method for intelligent supervision of customer data based on the Internet of Things according to claim 4 is characterized in that: The specific steps of cutting a single image from a single-sided image include: C1, the connected region analysis method is used to perform the first segmentation of the single-face image to obtain each block; C2: Project each block to get the length of each block; C3: Define the width of a single character and record it as L. Use this as the threshold and transfer the blocks whose projection length is greater than the threshold to C4 for the second segmentation. C4: Use the initial point selection process to find the initial point, use the left descent algorithm, and use the improved permeation rule of the permeation process to segment. If the permeation is unsuccessful, it is considered that the permeation has entered a closed loop and go to C5, where the specific steps of the initial point selection process are C41, and the specific steps of the permeation process improvement are C42; C41 (1), first, at the inflection point of the glued digit or where it is connected, the pixels will be more concentrated than other places; these pixels represent the characteristics of the digits, and they are called key points; select the leftmost and rightmost points, which must be located on both sides of the glued digit, representing the characteristics of the two digits respectively; C41 (2), assuming that BB is a connected component, and the distance between the leftmost point and the rightmost point is BD; take 1 / 3 of BD to intersect with the connected component. Since the number of intersection points is greater than one, select the top point and name it point a; C41 (3), if the right point or upper right point of point a is a black point, then this point will replace point a and become the new point a, until its right point or upper right point is a white point; at this time, the latest point a is the extreme point of this area; C41 (4), if the horizontal position of the latest point a is less than 2 / 3 of BD, it means that the latest point a is the extreme point of the left part of the connected component; then the latest point a is used as the starting point of the permeation process; otherwise, if the horizontal position of the latest point a is greater than 2 / 3 of BD, it means that the initial point a is on the connecting line of adjacent numbers; then the initial point a is used as the starting position of the permeation process; C42, describes the penetration process improvement rules as: Rule 1: if n1=0 and n3 =0, then next_pixel = n3 Rule 2: if n1=0 and n2 =1, then next_pixel = n1 Rule 3: if n1=0, then next_pixel = n1 Rule 4: if n6=1 and n7 =0, then next_pixel = n7 Where 0 is a white pixel, 1 is a black pixel, n1 is the pixel to the left of the current pixel, n2 is the pixel to the lower left of the current pixel, n3 is the pixel below the current pixel, n4 is the pixel to the lower right of the current pixel, n5 is the pixel to the right of the current pixel, n6 is the pixel to the upper left of the current pixel, and n7 is the pixel above the current pixel; C5: The upper part uses the left-down algorithm and stops immediately after the first penetration. The trajectory is recorded as P1. The lower part uses the left-upward algorithm. Same as before, it stops immediately after the first penetration. The trajectory is recorded as P2. The intersection points of the trajectory and the contact number are recorded as p1 and p2. A shortest path connecting p1 and p2 is found on the contour of the contact number. The connecting trajectory is recorded as P3. C6: The final complete image cutting trajectory includes P1, P2 and P3.

Images