Whiteboard picture correction method and apparatus, and storage medium

Abstract

Disclosed in the present invention are a whiteboard picture correction method and apparatus, and a storage medium, which are applied to a conference camera. The method comprises: in response to a whiteboard picture correction request, acquiring target correction positions of at least four reference pixel points of a whiteboard; performing localization on the current location of each reference pixel point of the whiteboard, so as to obtain an initial position of each reference pixel point; and on the basis of the target correction position and the initial position of each reference pixel point, performing calculation to obtain an affine transformation matrix, and on the basis of the affine transformation matrix, performing affine transformation on a first captured image of a conference camera to obtain a target image, so as to output the target image. Therefore, the accuracy and efficiency of correction performed on a captured picture of a conference camera are improved.

Description

Whiteboard image correction method, device and storage medium Technical Field

[0001] The present invention relates to the technical field of whiteboard detection and positioning, and in particular to a whiteboard image correction method, device and storage medium. Background Art

[0002] During remote conferences or distance learning, the whiteboard function is often used to display the images captured by the conference camera on the whiteboard. In order to better display the whiteboard content to the participants, it is usually necessary to calibrate the whiteboard image.

[0003] Existing whiteboard image correction requires manually adjusting the angle of the camera's footage before a remote meeting or remote teaching session by observing the footage captured by the conference camera. During the meeting, if the footage deviates, the angle also needs to be manually adjusted, resulting in low accuracy in whiteboard image correction.

[0004] Summary of the Invention

[0005] The present invention provides a whiteboard image correction method, device and storage medium to improve the correction accuracy and efficiency of the shooting images of a conference camera.

[0006] The present invention provides a whiteboard image correction method, which is applied to a conference camera. The method comprises:

[0007] In response to a request for whiteboard image correction, obtaining correction target positions of at least four reference pixels of the whiteboard;

[0008] Positioning the current position of each reference pixel of the whiteboard to obtain the initial position of each reference pixel;

[0009] Calculating an affine transformation matrix based on the correction target position and the initial position of each reference pixel point;

[0010] According to the affine transformation matrix, an affine transformation is performed on the first captured image of the conference camera to obtain a target image, so as to output the target image.

[0011] Furthermore, in response to a request for whiteboard image correction, obtaining correction target positions of at least four reference pixels of the whiteboard further includes:

[0012] Boundary information of the whiteboard is obtained, and correction target positions of four reference pixels are updated according to the boundary information.

[0013] Furthermore, the current position of each reference pixel of the whiteboard is located to obtain the initial position of each reference pixel, specifically:

[0014] The whiteboard is photographed to obtain a second photographed image, the second photographed image is input into a positioning detection model, and the positioning coordinates of each reference pixel point are output as the initial position; the positioning detection model is obtained by training the initial positioning detection model according to the training data set.

[0015] Furthermore, the second captured image is input into a positioning detection model, and the positioning coordinates of each reference pixel point are output as the initial position, specifically:

[0016] Inputting the second captured image into a positioning detection model, so that the positioning detection model classifies reference pixels using a softmax function and outputs the positioning coordinates of each reference pixel with the maximum confidence as an initial position; the softmax function converts the output data of the positioning detection model into a probability distribution;

[0017] The expression of the softmax function is:

[0018] Among them, z i is the i-th element of the linear combination, and K is the total number of categories.

[0019] Furthermore, the positioning detection model is obtained by training the initial positioning detection model according to the training data set, specifically:

[0020] Acquire several third captured images, annotate the coordinates of reference pixels in each of the third captured images, and use the annotated third captured images as a training data set; the third captured images are obtained by photographing the whiteboard at different angles using a conference camera;

[0021] The training data set is input into an initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

[0022] Furthermore, the training data set is input into the initial positioning detection model, and the initial positioning detection model is trained using the classification loss function and the positioning loss function to obtain a positioning detection model, specifically:

[0023] Obtain the annotated coordinates of the reference pixel points of the training sample, and convert the annotated coordinates into the output layer coordinates of the initial positioning detection model;

[0024] Dividing the area of ​​the output layer of the initial positioning detection model, taking a first area where the output layer coordinates are located as a first positive sample, and taking a preset number of adjacent second areas that are closest to the output layer coordinates as second positive samples;

[0025] The annotated positions of the first positive sample and each of the second positive samples are converted into numerical values ​​in a preset interval, the numerical values ​​are used as regression values ​​of the initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

[0026] Furthermore, a third region adjacent to the first positive sample or each of the second positive samples is taken as a negative sample.

[0027] Furthermore, the initial positioning detection model is trained using the classification loss function and the positioning loss function, specifically:

[0028] Calculating a loss function and using a back-propagation algorithm to update network parameters of the initial positioning detection model to minimize the loss function; the loss function includes a classification loss function and a positioning loss function;

[0029] The expression of the classification loss function is:

[0030] Among them, H represents the classification loss; n represents the number of samples, M represents the number of categories, and y ij Indicates whether the category is the same as the sample i category, 1 if they are the same, 0 if they are different, p ij Represents the predicted probability that the observed sample i belongs to category j;

[0031] The expression of the positioning loss function is:

[0032] Among them, L represents the positioning loss; n represents the number of samples, y i represents the true position of the i-th sample, f(x i ) represents the predicted position of the i-th sample.

[0033] As a preferred solution, the present invention applies the whiteboard image correction method to a conference camera, which can be built into the conference camera as an algorithm. When the algorithm is started, it performs positioning according to the current image captured by the conference camera, and performs correction calculations on the captured image of the conference camera based on the positioning information and the correction target position. The conference camera outputs the corrected image, thereby improving the correction accuracy and efficiency of the captured image of the conference camera.

[0034] Accordingly, the present invention also provides a whiteboard image correction device, which is applied to a conference camera. The device includes: a target information acquisition module, a positioning detection module, a transformation calculation module and a correction output module;

[0035] The target information acquisition module is used to obtain the correction target positions of at least four reference pixels of the whiteboard in response to a request for whiteboard image correction;

[0036] The positioning detection module is used to locate the current position of each reference pixel point of the whiteboard to obtain the initial position of each reference pixel point;

[0037] The transformation calculation module is used to calculate the affine transformation matrix according to the correction target position and the initial position of each reference pixel point;

[0038] The correction output module is used to perform affine transformation on the first captured image of the conference camera according to the affine transformation matrix to obtain a target image, so as to output the target image.

[0039] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute a whiteboard image correction method as described in the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] FIG1 is a schematic flow chart of an embodiment of a whiteboard image correction method provided by the present invention;

[0041] FIG2 is a schematic diagram of an embodiment of a backbone network and an output layer of a positioning detection model of a whiteboard image correction method provided by the present invention;

[0042] FIG3 is a schematic structural diagram of an embodiment of a whiteboard image correction device provided by the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Example 1

[0045] Referring to FIG. 1 , a whiteboard image correction method provided by an embodiment of the present invention is applied to a conference camera, including steps S101 to S103:

[0046] Step S101: In response to a request for whiteboard image correction, obtaining correction target positions of at least four reference pixels of the whiteboard;

[0047] Furthermore, in response to a request for whiteboard image correction, obtaining correction target positions of at least four reference pixels of the whiteboard further includes:

[0048] Boundary information of the whiteboard is obtained, and correction target positions of four reference pixels are updated according to the boundary information.

[0049] In this embodiment, the four corner points of the whiteboard are set as reference pixels, and the correction target positions (tar_tl, tar_tr, tar_bl, tar_br) of the four corner points are obtained. They are usually set to the four corners of the image, and a certain margin (padding) is reserved for the image height (image_height) and image width (image_widht). The updated correction target positions of the four reference pixels are as follows:

[0050] Tar_tl=(padding,padding);

[0051] Tar_tr=(image_widht-padding,padding);

[0052] Tar_bl=(padding,image_height-padding);

[0053] Tar_br=(image_width-padding, image_heght-padding).

[0054] Step S102: locating the current position of each reference pixel of the whiteboard to obtain the initial position of each reference pixel;

[0055] Furthermore, the current position of each reference pixel of the whiteboard is located to obtain the initial position of each reference pixel, specifically:

[0056] The whiteboard is photographed to obtain a second photographed image, the second photographed image is input into a positioning detection model, and the positioning coordinates of each reference pixel point are output as the initial position; the positioning detection model is obtained by training the initial positioning detection model according to the training data set.

[0057] Furthermore, the second captured image is input into a positioning detection model, and the positioning coordinates of each reference pixel point are output as the initial position, specifically:

[0058] Inputting the second captured image into a positioning detection model, so that the positioning detection model classifies reference pixels using a softmax function and outputs the positioning coordinates of each reference pixel with the maximum confidence as an initial position; the softmax function converts the output data of the positioning detection model into a probability distribution;

[0059] The expression of the softmax function is:

[0060] Among them, z i is the i-th element of the linear combination, and K is the total number of categories.

[0061] In this embodiment, the whiteboard detection and positioning problem is transformed into the detection and positioning of the four corners of the whiteboard. A deep learning method is used to detect and position the four corners of the whiteboard.

[0062] In this embodiment, the captured image of the input model is an RGB image, and the RGB color should be saved in three channels.

[0063] Furthermore, the positioning detection model is obtained by training the initial positioning detection model according to the training data set, specifically:

[0064] Acquire several third captured images, annotate the coordinates of reference pixels in each of the third captured images, and use the annotated third captured images as a training data set; the third captured images are obtained by photographing the whiteboard at different angles using a conference camera;

[0065] The training data set is input into an initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

[0066] Furthermore, the training data set is input into the initial positioning detection model, and the initial positioning detection model is trained using the classification loss function and the positioning loss function to obtain a positioning detection model, specifically:

[0067] Obtain the annotated coordinates of the reference pixel points of the training sample, and convert the annotated coordinates into the output layer coordinates of the initial positioning detection model;

[0068] Dividing the area of ​​the output layer of the initial positioning detection model, taking a first area where the output layer coordinates are located as a first positive sample, and taking a preset number of adjacent second areas that are closest to the output layer coordinates as second positive samples;

[0069] The annotated positions of the first positive sample and each of the second positive samples are converted into numerical values ​​in a preset interval, the numerical values ​​are used as regression values ​​of the initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

[0070] Furthermore, a third region adjacent to the first positive sample or each of the second positive samples is taken as a negative sample.

[0071] In this example, a dataset was created by collecting photos of a whiteboard taken from different angles. Four corners were annotated: the upper left, upper right, lower left, and lower right corners of the whiteboard. Each corner was labeled with the following information: corner type, x-coordinate, and y-coordinate. The dataset was randomly divided into two parts: a training set and a test set.

[0072] The backbone network can be a ResNet or Mobilenet model based on the deployment computing power. The output layer has a step of 16, so the output layer width and height = int((image width and height + 1) / step). The backbone network consists of an input layer and an output layer.

[0073] In this embodiment, the annotated coordinates (x, y) of the four reference pixels of the whiteboard are converted from image coordinates to output layer coordinates. The conversion relationship is:

[0074] xt = float(x) / step;

[0075] yt=float(y) / step;

[0076] Please refer to Figure 2. The output layer is divided into 8×7 regions. The location of the corner point and the three locations closest to the corner point (a total of 4) are selected as positive samples (as shown in the dark gray area in the figure), and the 12 samples around the positive sample are used as negative samples (as shown in the light gray area in the figure).

[0077] The position of the positive sample on the output layer is as follows:

[0078] Positive sample 1: int(xt), int(yt);

[0079] Positive sample 2: xt_near, int(yt);

[0080] Positive sample 3: int(xt),yt_near;

[0081] Positive sample 4: xt_near, yt_near;

[0082] Wherein, if xt-int(xt)>=0.5, xt_near=int(xt)+1, otherwise xt_near=int(xt)-1;

[0083] If yt-int(yt)>=0.5, yt_near=int(yt)+1, otherwise yt_near=int(yt)-1;

[0084] In this embodiment, for each positive sample (xp, yp), the annotation position is shifted to the interval [-1, 1), and the conversion relationship is:

[0085] F(xt)=(xt-(xp+0.5));

[0086] F(yt)=(yt-(yp+0.5));

[0087] Each positive example is labeled with an image coordinate. The values ​​originally range from 0 to the image size. Shifting the data to the range [-1, 1) allows for better training results. The transformed values, F(xt) and F(yt), are the values ​​that the network will regress.

[0088] Furthermore, the initial positioning detection model is trained using the classification loss function and the positioning loss function, specifically:

[0089] Calculating a loss function and using a back-propagation algorithm to update network parameters of the initial positioning detection model to minimize the loss function; the loss function includes a classification loss function and a positioning loss function;

[0090] The expression of the classification loss function is:

[0091] Where H represents the classification loss; n represents the number of samples, M represents the number of categories, in this embodiment, M=4, y ij Indicates whether the category is the same as the sample i category, 1 if they are the same, 0 if they are different, p ij Represents the predicted probability that the observed sample i belongs to category j;

[0092] The expression of the positioning loss function is:

[0093] Among them, L represents the positioning loss; n represents the number of samples, y i represents the true position of the i-th sample, f(x i ) represents the predicted position of the i-th sample.

[0094] For the output data of the model, softmax is used for classification, and only the result with the highest confidence score is selected for each corner point. The corner point position (x, y) is calculated by reverse calculation of the process described in 4, as follows:

[0095] x=(x_output+(xp+0.5))*step;

[0096] y=(y_output+(yp+0.5))*step.

[0097] Step S103: Calculate an affine transformation matrix based on the corrected target position and initial position of each reference pixel point; perform affine transformation on the first captured image of the conference camera based on the affine transformation matrix to obtain a target image, and output the target image.

[0098] In this embodiment, the conference camera transforms the captured image and outputs a corrected image.

[0099] The implementation of the present invention has the following effects:

[0100] The present invention applies the whiteboard image correction method to a conference camera, which can be built into the conference camera as an algorithm. When the algorithm is started, positioning is performed according to the current image captured by the conference camera, and correction calculations are performed on the captured image of the conference camera based on the positioning information and the correction target position. The conference camera outputs the corrected image, thereby improving the correction accuracy and efficiency of the captured image of the conference camera.

[0101] Example 2

[0102] 2 , which shows a whiteboard image correction device provided by an embodiment of the present invention, which is applied to a conference camera. The device includes: a target information acquisition module 201 , a positioning detection module 202 , a transformation calculation module 203 , and a correction output module 204 ;

[0103] The target information acquisition module is used to obtain the correction target positions of at least four reference pixels of the whiteboard in response to a request for whiteboard image correction;

[0104] The positioning detection module is used to locate the current position of each reference pixel point of the whiteboard to obtain the initial position of each reference pixel point;

[0105] The transformation calculation module is used to calculate the affine transformation matrix according to the correction target position and the initial position of each reference pixel point;

[0106] The correction output module is used to perform affine transformation on the first captured image of the conference camera according to the affine transformation matrix to obtain a target image, so as to output the target image.

[0107] The target information acquisition module includes an updating unit;

[0108] The updating unit is used to obtain boundary information of the whiteboard and update the correction target positions of the four reference pixels according to the boundary information.

[0109] The positioning detection module includes: a positioning detection unit and a model training unit;

[0110] The positioning detection unit is used to photograph the whiteboard to obtain a second photographed image, input the second photographed image into the positioning detection model, and output the positioning coordinates of each reference pixel point as an initial position;

[0111] The second captured image is input into the positioning detection model, and the positioning coordinates of each reference pixel point are output as the initial position, specifically:

[0112] Inputting the second captured image into a positioning detection model, so that the positioning detection model classifies reference pixels using a softmax function and outputs the positioning coordinates of each reference pixel with the maximum confidence as an initial position; the softmax function converts the output data of the positioning detection model into a probability distribution;

[0113] The expression of the softmax function is:

[0114] Among them, z i is the i-th element of the linear combination, and K is the total number of categories.

[0115] The model training unit is configured to obtain a plurality of third captured images, annotate the coordinates of reference pixels in each of the third captured images, and use the annotated third captured images as a training data set; the third captured images are obtained by photographing the whiteboard at different angles using a conference camera;

[0116] The training data set is input into an initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

[0117] Furthermore, the training data set is input into the initial positioning detection model, and the initial positioning detection model is trained using the classification loss function and the positioning loss function to obtain a positioning detection model, specifically:

[0118] Obtain the annotated coordinates of the reference pixel points of the training sample, and convert the annotated coordinates into the output layer coordinates of the initial positioning detection model;

[0119] Dividing the area of ​​the output layer of the initial positioning detection model, taking a first area where the output layer coordinates are located as a first positive sample, and taking a preset number of adjacent second areas that are closest to the output layer coordinates as second positive samples;

[0120] The annotated positions of the first positive sample and each of the second positive samples are converted into numerical values ​​in a preset interval, the numerical values ​​are used as regression values ​​of the initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

[0121] Furthermore, a third region adjacent to the first positive sample or each of the second positive samples is taken as a negative sample.

[0122] Furthermore, the initial positioning detection model is trained using the classification loss function and the positioning loss function, specifically:

[0123] Calculating a loss function and using a back-propagation algorithm to update network parameters of the initial positioning detection model to minimize the loss function; the loss function includes a classification loss function and a positioning loss function;

[0124] The expression of the classification loss function is:

[0125] Among them, H represents the classification loss; n represents the number of samples, M represents the number of categories, and y ij Indicates whether the category is the same as the sample i category, 1 if they are the same, 0 if they are different, p ij Represents the predicted probability that the observed sample i belongs to category j;

[0126] The expression of the positioning loss function is:

[0127] Among them, L represents the positioning loss; n represents the number of samples, y i represents the true position of the i-th sample, f(x i ) represents the predicted position of the i-th sample.

[0128] The whiteboard image correction device described above can implement the whiteboard image correction method of the above-mentioned method embodiment. The optional options in the above-mentioned method embodiment also apply to this embodiment and will not be described in detail here. The remaining contents of the embodiment of this application can be referred to the contents of the above-mentioned method embodiment and will not be repeated in this embodiment.

[0129] Example 3

[0130] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the whiteboard image correction method described in any one of the above embodiments.

[0131] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0132] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0133] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0134] The memory can be used to store the computer programs and / or modules, and the processor realizes various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile terminal, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0135] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0136] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A whiteboard image correction method, characterized in that: Applied to a conference camera, the method includes: In response to a request for whiteboard image correction, obtaining correction target positions of at least four reference pixels of the whiteboard; Positioning the current position of each reference pixel of the whiteboard to obtain the initial position of each reference pixel; Calculating an affine transformation matrix based on the correction target position and the initial position of each reference pixel point; According to the affine transformation matrix, an affine transformation is performed on the first captured image of the conference camera to obtain a target image, so as to output the target image.

2. A whiteboard image correction method according to claim 1, characterized in that: The step of obtaining the correction target positions of at least four reference pixels of the whiteboard in response to the request for whiteboard image correction further includes: Boundary information of the whiteboard is obtained, and correction target positions of four reference pixels are updated according to the boundary information.

3. A whiteboard image correction method according to claim 1, characterized in that: Positioning the current position of each reference pixel of the whiteboard to obtain the initial position of each reference pixel is specifically as follows: Photographing the whiteboard to obtain a second photographed image, inputting the second photographed image into a positioning detection model, and outputting the positioning coordinates of each reference pixel point as an initial position; The positioning detection model is obtained by training the initial positioning detection model according to the training data set.

4. A whiteboard image correction method according to claim 3, characterized in that: The step of inputting the second captured image into the positioning detection model and outputting the positioning coordinates of each reference pixel point as the initial position is specifically as follows: The second captured image is input into the positioning detection model so that the positioning detection model adopts The softmax function classifies the reference pixels and outputs the positioning coordinates of each reference pixel with the maximum confidence as the initial position; the softmax function converts the output data of the positioning detection model into a probability distribution; The expression of the softmax function is: Among them, z i is the i-th element of the linear combination, and K is the total number of categories.

5. A whiteboard image correction method according to claim 3, characterized in that: The positioning detection model is obtained by training the initial positioning detection model according to the training data set, specifically: Acquire several third captured images, annotate the coordinates of reference pixels in each of the third captured images, and use the annotated third captured images as a training data set; the third captured images are obtained by photographing the whiteboard at different angles using a conference camera; The training data set is input into an initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

6. A whiteboard image correction method according to claim 5, characterized in that: The training data set is input into the initial positioning detection model, and the classification loss function and the positioning loss function are used to train the initial positioning detection model to obtain the positioning detection model, specifically: Obtain the annotated coordinates of the reference pixel points of the training sample, and convert the annotated coordinates into the output layer coordinates of the initial positioning detection model; Dividing the area of the output layer of the initial positioning detection model, taking a first area where the output layer coordinates are located as a first positive sample, and taking a preset number of adjacent second areas that are closest to the output layer coordinates as second positive samples; The annotated positions of the first positive sample and each of the second positive samples are converted into numerical values in a preset interval, the numerical values are used as regression values of the initial positioning detection model, and the initial positioning detection model is trained using a classification loss function and a positioning loss function to obtain a positioning detection model.

7. A whiteboard image correction method according to claim 6, characterized in that: Also includes: A third region adjacent to the first positive sample or each of the second positive samples is taken as a negative sample.

8. A whiteboard image correction method according to claim 5, characterized in that: The initial positioning detection model is trained using the classification loss function and the positioning loss function, specifically: Calculating a loss function and using a back-propagation algorithm to update network parameters of the initial positioning detection model to minimize the loss function; the loss function includes a classification loss function and a positioning loss function; The expression of the classification loss function is: Among them, H represents the classification loss; n represents the number of samples, M represents the number of categories, and y ij Indicates whether the category is the same as the sample i category, 1 if they are the same, 0 if they are different, p ij Represents the predicted probability that the observed sample i belongs to category j; The expression of the positioning loss function is: Among them, L represents the positioning loss; n represents the number of samples, y i represents the true position of the i-th sample, f(x i ) represents the predicted position of the i-th sample.

9. A whiteboard image correction device, characterized in that: Applied to conference cameras, the device includes: a target information acquisition module, a positioning detection module, a transformation calculation module and a correction output module; The target information acquisition module is used to obtain the correction target positions of at least four reference pixels of the whiteboard in response to a request for whiteboard image correction; The positioning detection module is used to locate the current position of each reference pixel point of the whiteboard to obtain the initial position of each reference pixel point; The transformation calculation module is used to calculate the affine transformation matrix according to the correction target position and the initial position of each reference pixel point; The correction output module is used to correct the first image of the conference camera according to the affine transformation matrix. The captured image is affine transformed to obtain a target image, so as to output the target image.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to execute the whiteboard image correction method according to any one of claims 1 to 8.

Images