A method, apparatus, and computer-readable storage medium for dual-target accuracy optimization

Abstract

This application discloses a method, apparatus, and computer-readable storage medium for optimizing the calibration accuracy of a binocular vision system, used to improve the stability of the system. The method includes: controlling the binocular vision system to capture images of a target display screen when preset monitoring conditions are met, and performing geometric pose correction on the images captured by the front-view camera and the oblique-view camera respectively to obtain a first dot matrix image and a second dot matrix image, wherein the first dot matrix image corresponds to the oblique-view camera and the second dot matrix image corresponds to the front-view camera; extracting coordinate sets from the first and second dot matrix images respectively to obtain a first dot matrix coordinate set and a second dot matrix coordinate set; transforming the first dot matrix coordinate set using a pre-stored projection transformation matrix to obtain a third dot matrix coordinate set; calculating the mapping accuracy error between the second and third dot matrix coordinate sets; and determining whether to recalibrate the binocular vision system based on the mapping accuracy error.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus and computer-readable storage medium for optimizing the accuracy of dual-target positioning. Background Technology

[0002] With the widespread application of dot matrix displays in industrial inspection, visual calibration, and other fields, how to stably and accurately extract dot matrix regions from acquired images has become a key issue in improving the overall recognition accuracy and reliability of the system. In existing technologies, a monocular vision system is typically used to photograph the target display screen, and traditional methods such as image grayscale processing and edge detection are used to identify the dot matrix boundary regions. Common processing steps in these methods include: obtaining image gradient information using the Sobel or Canny operator to extract edge points; then fitting boundary curves using Hough transform or contour analysis methods; and finally performing perspective transformation on the image to obtain a regularized dot matrix region image.

[0003] However, boundary detection methods based on the above monocular image processing are highly sensitive to factors such as illumination changes, screen tilt angle, and image noise, and are easily affected by the following problems: On the one hand, gradient extraction is prone to missing boundary information or generating false boundaries under strong reflection or image blur conditions, resulting in unstable dot region recognition; on the other hand, since boundary judgment is based solely on a single-view image, it cannot effectively solve the problems of image distortion and partial boundary occlusion, and the fitted boundary contour is prone to deviation, which in turn affects the accuracy of image correction and subsequent dot region extraction.

[0004] Furthermore, in actual deployment, the position, angle, and background conditions of the display screen often change. Traditional methods lack an adaptive adjustment mechanism based on spatial redundancy information, resulting in significant fluctuations in recognition results and insufficient stability in complex environments. Therefore, there is an urgent need for an image processing scheme that can integrate multi-view image information and improve the anti-interference capability of boundary recognition, thereby enhancing the stability of dot matrix region extraction. Summary of the Invention

[0005] This application provides a method, apparatus, and computer-readable storage medium for optimizing binocular positioning accuracy, which can improve the stability of binocular vision systems.

[0006] The first aspect of this application provides a method for optimizing the accuracy of a dual-target positioning system, including:

[0007] When the preset monitoring conditions are met, the binocular vision system is controlled to capture images of the target display screen, and the images captured by the front-view camera and the oblique-view camera are geometrically pose corrected to obtain a first dot matrix image and a second dot matrix image, wherein the first dot matrix image corresponds to the oblique-view camera and the second dot matrix image corresponds to the front-view camera.

[0008] Coordinate sets are extracted from the first dot matrix image and the second dot matrix image respectively to obtain the first dot matrix coordinate set and the second dot matrix coordinate set;

[0009] The first set of point coordinates is transformed by a pre-stored projection transformation matrix to obtain the third set of point coordinates.

[0010] Calculate the mapping accuracy error between the second matrix coordinate set and the third matrix coordinate set;

[0011] Whether the binocular vision system needs to be recalibrated is determined based on the mapping accuracy error.

[0012] Optionally, before controlling the binocular vision system to capture images of the target display screen, the method further includes:

[0013] The fourth and fifth dot matrix images obtained by the binocular vision system from the target display screen are acquired, wherein the fourth dot matrix image corresponds to the oblique camera and the fifth dot matrix image corresponds to the frontal camera; the fourth and fifth dot matrix images are images that have undergone geometric pose correction.

[0014] The coordinate sets of the fourth dot matrix image and the fifth dot matrix image are extracted respectively to obtain the fourth dot matrix coordinate set and the fifth dot matrix coordinate set;

[0015] Calculate the projection transformation matrix based on the fourth and fifth point matrix coordinate sets.

[0016] Optionally, calculating the projection transformation matrix based on the fourth and fifth point matrix coordinate sets includes:

[0017] The projection transformation matrix is ​​calculated based on the projection transformation formula, the fourth lattice coordinate set, and the fifth lattice coordinate set; wherein, the projection transformation formula is:

[0018]

[0019]

[0020] Wherein, H is the projection transformation matrix.

[0021] Optionally, calculating the mapping accuracy error between the second matrix coordinate set and the third matrix coordinate set includes:

[0022] The intrinsic parameters of the oblique-view camera and the orthographic camera are calculated based on the fourth and fifth point matrix coordinate sets, respectively.

[0023] Based on the intrinsic parameters of the oblique-viewing camera and the orthographic camera, calculate the epipolar error between the first matrix coordinate set and the second matrix coordinate set;

[0024] Calculate the first geometric distance between the first set of point matrix coordinates and the third set of point matrix coordinates;

[0025] Calculate the reprojection accuracy error based on the first geometric distance;

[0026] The mapping accuracy error is calculated based on the epipolar error and the reprojection accuracy error.

[0027] Optionally, calculating the epipolar error between the first and second point matrix coordinate sets based on the intrinsic parameters of the oblique-viewing camera and the orthographic camera includes:

[0028] Based on epipolar geometry constraints, the fundamental matrix is ​​calculated using the fourth lattice coordinate set and the fifth lattice coordinate set;

[0029] Calculate the first bipolar line of the first target point on the first dot matrix coordinate set in the second dot matrix image based on the fundamental matrix;

[0030] The second bipolar line of the second target point on the second dot matrix coordinate set is calculated in the first dot matrix image based on the fundamental matrix, wherein the first target point and the second target point are corresponding points to each other;

[0031] Calculate the first distance based on the first bipolar line and the second target point;

[0032] Calculate the second distance based on the second bipolar line and the first target point;

[0033] The epipolar error is calculated based on the first distance and the second distance.

[0034] Optionally, calculating the reprojection accuracy error based on the first geometric distance includes:

[0035] Calculate the full map reprojection accuracy error based on the first geometric distance;

[0036] Based on the same division, the second lattice coordinate set and the third lattice coordinate set are respectively divided into m*n non-overlapping regions;

[0037] Calculate the second geometric distance for each region separately;

[0038] Calculate the region reprojection accuracy error based on each of the second geometric distances;

[0039] The reprojection accuracy error is determined based on the full map reprojection accuracy error and the regional reprojection accuracy error.

[0040] Optionally, the step of performing geometric pose correction on the images captured by the front-view camera and the oblique-view camera respectively to obtain the first bitmap image and the second bitmap image includes:

[0041] The binocular vision system captures oblique and frontal images of the target display screen; wherein the oblique image corresponds to an oblique camera and the frontal image corresponds to a frontal camera.

[0042] Determine the oblique view boundary outline of the oblique view image and the front view boundary outline of the front view image respectively;

[0043] Obtain pre-stored oblique projection transformation relationships and orthographic projection transformation relationships;

[0044] Based on the strabismus projection transformation relationship, the region enclosed by the strabismus boundary contour line is geometrically pose-corrected to obtain a first bitmap image, and based on the orthographic projection transformation relationship, the region enclosed by the orthographic boundary contour line is geometrically pose-corrected to obtain a second bitmap image.

[0045] Optionally, determining the oblique view boundary outline of the oblique view image and the front view boundary outline of the front view image respectively includes:

[0046] An initial search box for squinting is generated based on the oblique view image, and an initial search box for normal viewing is generated based on the frontal view image;

[0047] From the initial search box for strabismus, determine candidate points for strabismus that satisfy the gradient threshold and conform to the polarity change, and from the initial search box for normal vision, determine candidate points for normal vision that satisfy the gradient threshold and conform to the polarity change.

[0048] The strabismus candidate points and the emmetropia candidate points are fitted using the least squares method to determine the strabismus boundary contour line and the emmetropia boundary contour line, respectively.

[0049] A second aspect of this application provides a dual-target accuracy optimization apparatus, the apparatus being used to perform the method of the first aspect and any possible implementation thereof, the apparatus comprising:

[0050] The acquisition unit is used to control the binocular vision system to capture images of the target display screen when preset monitoring conditions are met, and to perform geometric pose correction on the images captured by the front-view camera and the oblique-view camera respectively to obtain a first dot matrix image and a second dot matrix image, wherein the first dot matrix image corresponds to the oblique-view camera and the second dot matrix image corresponds to the front-view camera.

[0051] An extraction unit is used to extract coordinate sets from the first dot matrix image and the second dot matrix image respectively, so as to obtain a first dot matrix coordinate set and a second dot matrix coordinate set.

[0052] The transformation unit is used to transform the first set of point matrix coordinates using a pre-stored projection transformation matrix to obtain the third set of point matrix coordinates.

[0053] A calculation unit is used to calculate the mapping accuracy error between the second matrix coordinate set and the third matrix coordinate set;

[0054] The determining unit is used to determine whether to recalibrate the binocular vision system based on the mapping accuracy error.

[0055] A third aspect of this application provides a dual-target accuracy optimization device, comprising:

[0056] Processor, memory, input / output units, and bus;

[0057] The processor is connected to the memory, the input / output unit, and the bus;

[0058] The memory stores a program, and the processor calls the program to execute the method of the first aspect and any possible implementation of the first aspect.

[0059] A fourth aspect of this application provides a computer-readable storage medium storing a program that, when executed on a computer, causes the computer to perform the methods of the first aspect and any possible implementation thereof.

[0060] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0061] In this embodiment, the system can extract coordinate sets from two corrected images acquired by a binocular vision system and perform calculations and analyses using a projection transformation matrix to determine whether the calibration state of the current binocular system has shifted. Recalibration is triggered when the error exceeds a preset threshold, effectively ensuring the projection consistency and coordinate matching accuracy between the oblique and frontal view images, thus guaranteeing the accuracy and stability of the system in subsequent image recognition, position measurement, or spatial reconstruction processes. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating one embodiment of the dual-objective accuracy optimization method in this application.

[0063] Figure 2 This is a flowchart illustrating a sub-implementation of the dual-objective accuracy optimization method in this application.

[0064] Figure 3 This is a flowchart illustrating another sub-implementation of the dual-objective accuracy optimization method in this application.

[0065] Figure 4 This is a flowchart illustrating another sub-implementation of the dual-objective accuracy optimization method in this application.

[0066] Figure 5 This is a flowchart illustrating another sub-implementation of the dual-objective accuracy optimization method in this application.

[0067] Figure 6 This is a flowchart illustrating another sub-implementation of the dual-objective accuracy optimization method in this application.

[0068] Figure 7 This is a schematic diagram of one embodiment of the dual-target accuracy optimization device in this application.

[0069] Figure 8 This is a schematic diagram of the structure of an electronic device according to one embodiment of the present application. Detailed Implementation

[0070] This application provides a method, apparatus, and computer-readable storage medium for optimizing binocular positioning accuracy, thereby improving the stability of binocular vision systems.

[0071] The method described in this application can be applied to servers, terminals, or other devices with logical processing capabilities; therefore, this application does not limit its application. For ease of description, the following description uses a server as the executing entity.

[0072] The embodiments of this application will now be described with reference to the accompanying drawings.

[0073] Please see Figure 1 One embodiment of the dual-target accuracy optimization method in this application includes:

[0074] 101. When the preset monitoring conditions are met, control the binocular vision system to capture images of the target display screen, and perform geometric pose correction on the images captured by the front-view camera and the oblique-view camera respectively to obtain a first dot matrix image and a second dot matrix image, wherein the first dot matrix image corresponds to the oblique-view camera and the second dot matrix image corresponds to the front-view camera.

[0075] When preset monitoring conditions are met, the server acquires image information from two cameras in the binocular vision system. The image data from the oblique-view camera on the target display screen is corrected for geometric pose to obtain the first dot matrix image; similarly, the image data from the front-view camera on the same target display screen is corrected for geometric pose to obtain the second dot matrix image. Geometric pose correction is used to eliminate distortion and offset caused by differences in camera installation angle or position, making subsequent analysis more spatially consistent. It should be noted that the preset monitoring conditions can be set intervals, a decrease in the yield rate of the display screen detected by the binocular vision system, or other conditions; this application does not limit these. It should also be noted that before the server controls the binocular vision system to capture images, a given calibration image needs to be illuminated on the target display screen via PG dot-screen operation (PG dot-screen operation can be performed by the server or manually by the user). This involves using an image signal generator (PG device) to send the image signal corresponding to the calibration image to the target display screen under test, driving the target display screen to accurately present the calibration image, providing a standard and stable image source for subsequent binocular vision system capture.

[0076] Furthermore, to obtain high-precision calibration mark coordinates, the server performs image enhancement processing. First, the server uses bilateral filtering to remove background noise while preserving edge pixel information as much as possible. Then, it performs shadow correction or gamma correction to eliminate local gray-level inhomogeneity, ensuring uniform gray-level distribution across regions, which is beneficial for subsequent sub-pixel edge extraction. Next, the server extracts edge points from the enhanced image, filters out outliers using the RANSAC algorithm, and then uses a circle fitting method to accurately locate the position of each mark coordinate, thus providing high-quality point matrix data for subsequent stereo coordinate matching and accuracy analysis.

[0077] 102. Extract the coordinate sets of the first dot matrix image and the second dot matrix image respectively to obtain the first dot matrix coordinate set and the second dot matrix coordinate set;

[0078] The server performs image analysis on the first dot matrix image, identifying and extracting feature points or bright spots with regularity or calibrated significance, and stores the image coordinates of these points as the first dot matrix coordinate set. Simultaneously, the second dot matrix image is processed in the same way to obtain the second dot matrix coordinate set. These two coordinate sets reflect the positions of feature points on the same display screen as observed by two cameras from different perspectives, providing basic data for subsequent transformations and error calculations.

[0079] It should be noted that when extracting coordinate sets, the mark center coordinates are prone to shift during the extraction of the calibration mark coordinates. There are two main reasons for this: First, a circular mark will become an ellipse after perspective projection, and the edge pixels will be discretized; second, edge extraction algorithms (Canny, Sobel) are sensitive to lighting and noise and are prone to identifying false edges.

[0080] To achieve good accuracy in extracting the calibration mark coordinates, image quality is crucial. Therefore, in this embodiment, the server first performs enhancement processing on the image. The purpose is twofold: first, to remove some background noise and edge blurring interference; and second, to make the gray values ​​of each region evenly distributed to facilitate subsequent edge extraction. Specifically, this is done by first performing bilateral filtering and then preprocessing methods such as shadow correction.

[0081] For example, one processing flow in this application embodiment is as follows:

[0082] Image preprocessing enhancement employs bilateral filtering to minimize the loss of edge pixel information while removing interference noise.

[0083] Then use shadow correction or gamma correction to correct local unevenness in the image;

[0084] Subpixel edges are extracted to find edge points, then ransac is used to filter out outliers, and then the mark coordinates are located by fitting a circle.

[0085] 103. Transform the first set of coordinates using a pre-stored projection transformation matrix to obtain the third set of coordinates.

[0086] The server calls a pre-saved projection transformation matrix in the system to transform the coordinates of each point in the first dot matrix coordinate set according to the matrix, resulting in a new coordinate set, namely the fourth dot matrix coordinate set. The projection transformation matrix is ​​used to map the dot matrix positions seen by the oblique camera onto the projection plane of the frontal view, simulating the point distribution that should be obtained when viewed from the frontal angle.

[0087] 104. Calculate the mapping accuracy error between the second and third point matrix coordinate sets;

[0088] The server calculates the coordinate differences between corresponding points in the first and fourth point matrix coordinate sets, evaluates the spatial offset distance between them point by point, and calculates the overall mapping accuracy error by combining all offset values. This error quantifies the accuracy of the projection transformation matrix stored in the current system in the real scene, that is, the degree of deviation between the actual projection and the ideal projection.

[0089] 105. Determine whether to recalibrate the binocular vision system based on the mapping accuracy error.

[0090] The server compares the calculated mapping accuracy error with a preset threshold in the system to determine whether the current error value is greater than or equal to the threshold. This threshold defines the maximum acceptable accuracy error range for the system. Once the actual error exceeds this range, the calibration state reflected by the current projection transformation matrix is ​​considered unreliable, and there may be inaccuracies due to installation offset, environmental changes, or changes in camera parameters.

[0091] When the server determines that the mapping accuracy error exceeds or reaches a preset threshold, that is, when it confirms that the existing projection transformation matrix cannot accurately reflect the geometric relationship between the oblique-view camera and the normal-view camera, the server determines that the binocular vision system needs to be recalibrated in order to regenerate an accurate projection transformation matrix, thereby restoring the spatial accuracy of image coordinate transformation and matching, and avoiding unreliable results in subsequent image processing due to error accumulation.

[0092] In this embodiment, the server can extract the coordinate set based on two corrected images acquired by the binocular vision system and perform calculations and analysis using the projection transformation matrix to determine whether the calibration state of the current binocular system has shifted. When the error exceeds a preset threshold, recalibration is triggered, effectively ensuring the projection consistency and coordinate matching accuracy between the oblique view image and the front view image, and guaranteeing the accuracy and stability of the system in subsequent image recognition, position measurement, or spatial reconstruction processes.

[0093] Furthermore, in some embodiments of this application, when calibrating the binocular vision system, the server first needs to independently calibrate the oblique-view camera and the main-view camera to obtain the corresponding intrinsic parameter matrices and distortion coefficients, and calculate the intrinsic parameters corresponding to each of the two cameras. Next, mark points are extracted from the pose correction images acquired by the oblique-view camera and the main-view camera after geometric pose correction, resulting in corresponding set of mark coordinate points (each camera corresponds to one set, and the mark points in either set can be found in the other set). Then, the fundamental matrix is ​​calculated using epipolar geometric constraints, and the relative pose (rotation and translation) from the oblique-view camera to the main-view camera is calculated using the fundamental matrix. The essential matrix is ​​then solved using SVD decomposition, and the projection transformation matrix is ​​calculated using the two sets. Finally, the server solves for the intrinsic parameters, relative pose, fundamental matrix, essential matrix, and projection transformation matrix corresponding to the oblique-view camera and the main-view camera, and persistently stores these parameters, thus completing the calibration process (this process is also known as offline calibration).

[0094] Please see Figure 2In some embodiments of this application, before step 101 of the above embodiments, which controls the binocular vision system to capture images of the target display screen, the binocular positioning accuracy optimization method may further include the following steps:

[0095] 201. Acquire the fourth and fifth dot matrix images obtained by the binocular vision system from the target display screen, wherein the fourth dot matrix image corresponds to the oblique view camera and the fifth dot matrix image corresponds to the front view camera; the fourth and fifth dot matrix images are images after geometric pose correction.

[0096] Before the server performs calibration accuracy monitoring, it acquires two data images for calibration from the target display screen captured by the binocular vision system. The image captured by the oblique-view camera is converted into a fourth dot matrix image after geometric pose correction, and the image captured by the front-view camera is also converted into a fifth dot matrix image after geometric pose correction. These fourth and fifth dot matrix images are used to calculate the projection relationship; therefore, the environment is usually kept stable and features are clear during image acquisition to improve subsequent calibration accuracy. It should be noted that, similar to step 101, the fourth and fifth dot matrix images acquired by the server also need to be illuminated on the target display screen using a PG dot matrix operation (this PG dot matrix operation can be performed by the server or manually by the user), details of which will not be elaborated here.

[0097] 202. Extract the coordinate sets of the fourth and fifth dot matrix images respectively to obtain the coordinate sets of the fourth and fifth dot matrix images;

[0098] The server performs image processing on the fourth dot matrix image, extracting points with geometric regularity or brightness characteristics to form the fourth dot matrix coordinate set; simultaneously, the same processing flow is performed on the fifth dot matrix image to obtain the fifth dot matrix coordinate set. Because these two images were captured under strict conditions for calibration, the coordinate extraction is highly stable, which helps in establishing accurate geometric mapping relationships subsequently.

[0099] 203. Calculate the projection transformation matrix based on the coordinate sets of the fourth and fifth lattice points.

[0100] Based on the positional relationships of corresponding points in the fourth and fifth point matrix coordinate sets, the server calculates a projection transformation matrix using a multi-point pair fitting method (e.g., least squares fitting) to describe the geometric mapping relationship between the oblique-view camera and the normal-view camera. This projection transformation matrix will serve as a spatial reference in subsequent image analysis processes, used to convert the coordinates of the oblique-view image into the position corresponding to the normal-view angle.

[0101] Specifically, the server can calculate the projection transformation matrix based on the projection transformation formula, the fourth lattice coordinate set, and the fifth lattice coordinate set; whereby the projection transformation formula is:

[0102] Formula 1

[0103] Formula 2

[0104] Where H is the projection transformation matrix; Corresponding to the fifth point matrix coordinate set; This corresponds to the fourth point matrix coordinate set.

[0105] In this embodiment, the server can first calculate an accurate projection transformation matrix based on specially acquired dot matrix images used for calibration, and then use it for subsequent coordinate transformation and error analysis of the working image. This not only improves the initial accuracy of the projection matrix, but also provides a reliable reference for subsequent accuracy monitoring using this matrix, enhancing the overall stability and practicality of the binocular vision system calibration accuracy optimization method.

[0106] Please see Figure 3 In some embodiments of this application, step 104 in the above embodiments, calculating the mapping accuracy error between the second and third point matrix coordinate sets, may include the following steps:

[0107] 301. Calculate the intrinsic parameters of the oblique-view camera and the orthographic camera based on the fourth and fifth point matrix coordinate sets, respectively;

[0108] The server calculates the intrinsic parameter matrices for the oblique-view and front-view cameras based on the fourth and fifth point matrix coordinate sets, respectively. First, it extracts the coordinate sets of mark points from the two calibration maps, ensuring a one-to-one correspondence between the points in the fourth and fifth point matrix coordinate sets. Then, the server uses these corresponding points to solve for the intrinsic parameter matrix of each camera through geometric pose relationships, characterizing the camera's scale and principal point position on the image plane, among other imaging geometric relationships. To improve the stability of the intrinsic parameter matrices, the server also considers the influence of imaging distortion during calculation, preprocessing and correcting the input coordinates to reduce distortion interference and improve calculation accuracy. Specifically, in practical applications, the front-view and oblique-view cameras need to be calibrated separately. For any camera, during calibration, the server can first establish a camera pinhole model:

[0109] Formula 3

[0110] Then, calculate the intrinsic parameter matrix according to the following formula:

[0111] Formula 4

[0112] In practical applications, the distortion model can be solved using the Brown-Conrady model to obtain the intrinsic parameter matrix. Specifically, the radial distortion is:

[0113] Formula 5

[0114] Tangential distortion becomes:

[0115] Formula 6

[0116] The final internal parameters obtained by the server are ,in .

[0117] 302. Based on the intrinsic parameters of the oblique-viewing camera and the normal-viewing camera, calculate the epipolar error between the first and second point matrix coordinate sets.

[0118] After obtaining the intrinsic parameter matrices of the two cameras, the server calculates the epipolar error between the first and second point coordinate sets. Based on the intrinsic parameter matrices, the server constructs the geometric constraints between the two cameras, derives the fundamental and essential matrices describing the relative pose, and thus obtains the epipolar representation. For each set of matched points, the server calculates its corresponding epipolar line in another image and measures the geometric distance from the point to the epipolar line, summarizing the results to obtain the epipolar error set and calculating indicators such as mean, standard deviation, maximum, and minimum values. The server also analyzes the error distribution by region to identify areas with abnormal local geometric relationships.

[0119] 303. Calculate the first geometric distance between the first and third point matrix coordinate sets;

[0120] The server calculates the first geometric distance between the first and third point matrix coordinate sets. Using the first set as a reference and the third set as the mapping result, the server pairs points one by one and calculates the lateral and longitudinal deviations of corresponding points. The geometric distance for each point pair is then obtained by combining these deviations. Subsequently, the server aggregates all geometric distances into a distance set and analyzes its distribution characteristics (including mean, variance, maximum, and minimum values). To further identify local offsets, the server divides the image into several regions and calculates the distance distribution within each region to identify areas of concentrated mapping deviations.

[0121] 304. Calculate the reprojection accuracy error based on the first geometric distance;

[0122] The server calculates the reprojection accuracy error based on the first geometric distance. The server performs overall and regional statistics on the set of geometric distances, calculates the mean, standard deviation, maximum and minimum values ​​of the reprojection error, and determines the error concentration area through error distribution analysis. Based on this, the server compares the difference between the mean error of each region and the global mean. When the local deviation exceeds the threshold, it is marked as a reprojection anomaly and its location and characteristics are recorded to form the basic data for subsequent calibration and adjustment.

[0123] 305. Calculate the mapping accuracy error based on the epipolar error and the reprojection accuracy error.

[0124] The server calculates the mapping accuracy error based on epipolar error and reprojection accuracy error. To ensure comparability between different error types, the server first normalizes the epipolar error and reprojection error separately, then performs weighted fusion according to preset weights to obtain a comprehensive mapping accuracy error index. Simultaneously, it outputs the weight ratio of each error component and the distribution of abnormal regions to identify the source of the error. Finally, the server determines whether the mapping accuracy meets the requirements based on the comprehensive index and threshold rules, and provides the results to subsequent calibration or diagnostic modules as a decision-making basis.

[0125] Specifically, the server can calculate the mapping accuracy error using the following formula:

[0126] Formula 7

[0127] Here, α and β are weighting factors, which can be set according to the data distribution or an adaptive strategy. This represents the reprojection error calculation index value. This represents the calculated value of the epipolar matching error. It should be noted that, because the reprojection error and epipolar error have different evaluation dimensions and numerical distributions, they need to be normalized before calculating the mapping accuracy error. Normalization can be performed using the following formula:

[0128] , Formula 8

[0129] in, These are the statistical standard deviations or reference scales for the two types of errors, respectively.

[0130] In this embodiment, the server calculates the intrinsic parameter matrix, epipolar error, and reprojection error in sequence and then fuses them to generate a mapping accuracy error. This allows the accuracy evaluation of the calibration results to simultaneously cover both geometric constraints and projection consistency dimensions, thereby distinguishing error sources, locating abnormal areas, improving the reliability of binocular mapping accuracy judgment, and ultimately reducing the difficulty of parameter debugging.

[0131] Please see Figure 4 In some embodiments of this application, step 302 in the above embodiments, which calculates the epipolar error between the first and second point matrix coordinate sets based on the intrinsic parameters of the oblique-viewing camera and the front-viewing camera, may include the following steps:

[0132] 401. Based on epipolar geometry constraints, calculate the fundamental matrix using the coordinate sets of the fourth and fifth lattice points;

[0133] The server first calculates the fundamental matrix based on epipolar geometry constraints, using the fourth and fifth point set coordinates. By analyzing the correspondence between the two point sets, the server determines the relative spatial mapping relationship between the two cameras in the stereo system and generates the fundamental matrix, which describes the geometric correspondence of matching points in the two images. During the calculation process, the server also considers the accuracy and distribution of the point sets to ensure that the fundamental matrix stably reflects the geometric relationship between the cameras, providing a reliable basis for subsequent bipolar line generation and epipolar error calculation.

[0134] Specifically, the epipolar geometric constraints are:

[0135] Formula 9

[0136] Among them, Based on the matrix, and These represent the corresponding mark coordinates of the oblique-view camera and the main-view camera, respectively, and their relationship with the extrinsic parameters is as follows:

[0137] Formula 10

[0138] here and These represent the intrinsic parameter matrices obtained from the calibration of the two cameras, respectively. , This indicates the relative orientation (rotation and translation) of camera 1 to camera 2. The antisymmetric matrix representing the translation vector.

[0139] 402. Calculate the first bipolar line of the first target point on the first lattice coordinate set in the second lattice image based on the fundamental matrix;

[0140] After obtaining the fundamental matrix, the server calculates the first bipolar line of the first target point in the first matrix coordinate set within the second matrix image. Using the first target point as a reference, the server derives its corresponding epipolar line in the second matrix image using the fundamental matrix, and represents the direction and position of this bipolar line as a straight line. This bipolar line provides a basis for subsequent measurements of the distance from the target point to the corresponding geometric constraints. Simultaneously, the server can record the coverage area of ​​each epipolar line in the image, preparing for multi-point matching and local accuracy analysis. Specifically, the server can calculate the first bipolar line according to the following formula:

[0141] Formula 11

[0142] It can be described using a linear equation:

[0143] Formula 12

[0144] Therefore, the server can extract:

[0145] Formula 13

[0146] 403. Calculate the second bipolar line of the second target point on the second matrix coordinate set in the first matrix image based on the fundamental matrix. The first target point and the second target point are corresponding points to each other.

[0147] Simultaneously, the server calculates the second bipolar line of the second target point in the first dot matrix image based on the fundamental matrix, where the first and second target points are corresponding points. Through this operation, the server establishes a bipolar line structure with corresponding point pairs, providing complete geometric constraint information for epipolar error calculation. The server repeats this process for all matching points to ensure that each point has a corresponding bipolar line, thus covering the entire dot matrix image during error statistics. It should be noted that the specific calculation process for the second bipolar line is similar to that of the first bipolar line in step 402, and will not be repeated here.

[0148] 404. Calculate the first distance based on the first bipolar line and the second target point;

[0149] The server calculates the first distance using a first bipolar line and a second target point. Specifically, the server measures the perpendicular distance from the second target point to its corresponding first bipolar line; this distance reflects the deviation of the matching point under the epipolar geometric constraint. The server performs preliminary statistics on all calculated first distances, including the deviation value of each point and the deviation distribution in local areas, providing a quantitative basis for judging the geometric consistency of the stereo system. Specifically, the server can calculate the first distance according to the following formula:

[0150] Formula 14

[0151] in, Indicates the first distance.

[0152] 405. Calculate the second distance based on the second bipolar line and the first target point;

[0153] Similarly, the server calculates the second distance using the second bipolar line and the first target point. The server measures the perpendicular distance from the first target point to its corresponding second bipolar line to obtain the deviation amount that complements the direction of step 404. By combining the first and second distances, the server can more comprehensively reflect the geometric consistency of the matching point pairs and mark points or regions with large deviations, providing a basis for subsequent mapping accuracy analysis and possible local adjustments.

[0154] It should be noted that the specific calculation process of the second distance is similar to that of the first distance in step 402, and will not be repeated here.

[0155] 406. Calculate the epipolar error based on the first distance and the second distance.

[0156] Finally, the server calculates the epipolar error based on the first and second distances. The server combines the distances in both directions to obtain the total epipolar error of the matching point, and performs statistical analysis on the error set of all matching points, including indicators such as mean, standard deviation, maximum, and minimum values. Simultaneously, it can perform regional analysis by image area to identify areas with large local epipolar deviations. Through this quantitative information, the server can provide comprehensive data support for evaluating the accuracy of binocular mapping and, when necessary, prompt recalibration or adjustment, thereby ensuring the stability and reliability of the binocular system under attitude changes or local offsets. Specifically, the server can calculate the epipolar error using the following formula:

[0157] Formula 15

[0158] in, Indicates corresponding points The epipolar error is expressed as the sum of the squares of the first and second distances.

[0159] The fundamental matrix Substituting, we get:

[0160] Formula 16

[0161] It should be noted that the above calculation is for the epipolar error of a pair of target points. In practical applications, to improve accuracy, it is generally necessary to calculate the epipolar error of all target points. The server can calculate the final epipolar error using the following formula:

[0162] Formula 17

[0163] In this embodiment, the server achieves a comprehensive quantitative evaluation of the geometric consistency of the binocular system by progressively generating the basic matrix, deriving the bipolar lines, measuring the matching point deviation, and statistically analyzing the errors. This facilitates accurate reflection of the spatial mapping deviation between the two cameras, timely identification of abnormal regions and local errors, thereby maintaining the stability of the mapping results under slight attitude changes or local offsets, while reducing the workload of manual parameter adjustment.

[0164] Please see Figure 5 In some embodiments of this application, step 304 in the above embodiments, which calculates the reprojection accuracy error based on the first geometric distance, may include the following steps:

[0165] 501. Calculate the full map reprojection accuracy error based on the first geometric distance;

[0166] The server uses all the first geometric distances obtained in step 303 to calculate the full-image mapping accuracy error across the entire image range using statistical methods such as average, standard deviation, or maximum value. This error is used to reflect the error level between the original coordinate sets that have not undergone projection transformation, thereby providing a baseline for judging the effectiveness of the transformation.

[0167] 502. Based on the same division, the second and third point matrix coordinate sets are each divided into m*n non-overlapping regions;

[0168] The server divides the image regions corresponding to the first and fourth point matrix coordinate sets into m×n non-overlapping sub-regions according to fixed rules. For example, it can divide the image into grid blocks based on the pixel coordinate axes, and ensures that the two coordinate sets are consistent in spatial division, so as to facilitate subsequent region-by-region error analysis.

[0169] 503. Calculate the second geometric distance for each region separately;

[0170] In each region divided in step 502, the server extracts the corresponding points of the first and third lattice coordinates within that region and calculates the second geometric distance between them. This distance represents the degree of local coordinate offset between points within the corresponding region after the projection transformation, and is used to finely evaluate the local projection effect.

[0171] 504. Calculate the reprojection accuracy error of the region based on each second geometric distance;

[0172] In each region divided in step 502, the server extracts the corresponding points of the first and third lattice coordinates within that region and calculates the second geometric distance between them. This distance represents the degree of local coordinate offset between points within the corresponding region after the projection transformation, and is used to finely evaluate the local projection effect.

[0173] 505. Determine the reprojection accuracy error based on the overall map reprojection accuracy error and the regional reprojection accuracy error.

[0174] The server combines the full-map mapping accuracy error obtained in step 501 with the regional mapping accuracy errors obtained in step 504 to form the final mapping accuracy error index used for calibration and judgment. For example, by setting a weighted strategy, the full-map error can be used as a global benchmark, and the regional errors can be used as local supplements, thereby obtaining a more representative and judgmental overall mapping error value.

[0175] In this embodiment, by introducing a multi-level error analysis method using first and second geometric distances, the server can not only obtain the overall image mapping accuracy error across the entire image range, but also refine it to m×n sub-regions for local error evaluation, and comprehensively determine the mapping accuracy error by combining the error values ​​of both global and local dimensions. This layered accuracy calculation method makes the error evaluation more comprehensive and accurate, helps to more reliably determine whether the current projection transformation matrix is ​​invalid, and improves the accuracy and stability of calibration status judgment.

[0176] Please see Figure 6 In some embodiments of this application, step 101 in the above embodiments, which performs geometric pose correction on the images captured by the front-view camera and the oblique-view camera to obtain a first bitmap image and a second bitmap image, may include the following steps:

[0177] 601. Acquire oblique and frontal view images of the target display screen captured by a binocular vision system; wherein the oblique view image corresponds to the oblique camera and the frontal view image corresponds to the frontal camera;

[0178] The server controls a binocular vision system that uses both a slanted-view camera and a front-view camera to acquire image data from the target display screen, resulting in slanted and front-view images. The slanted-view image is captured from a side view at a certain angle, while the front-view image is captured directly facing the display screen. Both images are acquired simultaneously by the two cameras, providing temporal synchronization and spatial complementarity, thus providing a source of images for subsequent extraction of precise dot matrix regions. Since the display screen boundary exhibits significant geometric differences at different angles, combining these two images allows for a better capture of the true boundary contours of the display area.

[0179] 602. Generate an initial search box for oblique view based on the oblique view image, and generate an initial search box for frontal view based on the frontal view image;

[0180] Based on prior information (such as an estimate of the approximate location of the display screen in the image) or manually provided rectangular bounding boxes, the server constructs initial search regions in the oblique and frontal views, respectively, called the oblique initial search boxes and the frontal initial search boxes. These boxes are essentially rectangular regions (i.e., Region of Interest) that cover the locations in the image where the display screen boundary might exist. Because there is interference and noise between the target boundary and the background in real images, performing boundary extraction on the entire image would result in high computational cost and low accuracy. Therefore, using the ROI method to limit the candidate range can significantly improve the efficiency and accuracy of boundary detection. This search box constitutes the limited range for subsequent candidate point extraction.

[0181] 603. Determine candidate points for strabismus that satisfy the gradient threshold and conform to the polarity change from the initial search box for strabismus, and determine candidate points for normal vision that satisfy the gradient threshold and conform to the polarity change from the initial search box for normal vision.

[0182] The server performs image gradient calculation and polarity change analysis within the ROI region of each image to identify potential boundary candidate points. Image gradient is an indicator that measures the intensity of pixel gray-level changes in an image; it is defined as the degree of change in gray-level values ​​between adjacent pixels and is commonly used for edge detection. To improve the robustness of gradient calculation, a morphological gradient calculation method is employed, namely:

[0183] Formula 18

[0184] Where InImg represents the preprocessed input image, S is the structuring element used to define the neighborhood range, ⊕ represents the grayscale morphological dilation operation, and ⊖ represents the grayscale morphological erosion operation. Dilation and erosion are fundamental operations in image morphology: dilation emphasizes bright areas, and erosion emphasizes dark areas. Their combination forms a gradient, which can effectively enhance edge contrast and suppress high-frequency noise.

[0185] Within the ROI region, gradient values ​​are searched pixel by pixel. If the gradient value of a pixel is greater than a set threshold, the pixel is considered a "candidate boundary point". Then, it is further judged whether there is a polarity change (i.e., a significant grayscale jump from white to black or black to white) to filter out background interference and thus retain edge points with structural significance.

[0186] 604. Fit the candidate points for strabismus and the candidate points for emmetropia using the least squares method respectively to determine the boundary contour lines for strabismus and emmetropia.

[0187] The server performs least-squares linear fitting on the candidate points for both oblique and frontal views selected in the previous step to fit the boundary contour line. Least-squares is a commonly used data fitting method whose goal is to find a straight line. Make all candidate points The goal is to minimize the sum of the squares of the perpendicular distances to this line. The corresponding optimization objective is:

[0188] Formula 19

[0189] Here, δ(α) is the objective function, measuring the overall deviation between the fitted line and the candidate points. The least squares method obtains the optimal slope α1 and intercept α0 by differentiating and solving the equation, thus fitting the boundary line of the display screen. This fitting process can further resist the interference of isolated noise points and enhance the stability of boundary recognition.

[0190] 605. Obtain the pre-stored oblique projection transformation relationship and orthographic projection transformation relationship;

[0191] The server loads the projection transformation relationships corresponding to the camera configuration from pre-stored system calibration data, including the projection transformation matrices for both oblique-view and orthographic cameras. These transformation relationships are established by the system during the initialization phase using a calibration board or calibration pattern, and are typically represented as a homography matrix. This matrix describes the transformation between the image coordinate system (i.e., pixel coordinates) and the world coordinate system or ideal projection plane. Specifically, it includes geometric transformations such as rotation, translation, scaling, and perspective changes, used for subsequent image correction.

[0192] 606. Perform geometric pose correction on the region enclosed by the strabismus boundary contour line according to the strabismus projection transformation relationship to obtain a first dot matrix image, and perform geometric pose correction on the region enclosed by the orthographic boundary contour line according to the orthographic projection transformation relationship to obtain a second dot matrix image.

[0193] The server uses the region enclosed by the boundary contour line fitted in step 604 as the correction target region, and applies both oblique projection transformation and orthographic projection transformation relationships to perform geometric pose correction on this region in the original image. This process eliminates distortions and tilting effects caused by the shooting angle in both oblique and orthographic images, maps the target display area to a standard reference coordinate system, and finally obtains a geometrically aligned first and second dot matrix image. These two images are derived from the oblique and orthographic images respectively, but they are spatially consistent, facilitating subsequent precise registration, dot matrix extraction, and error calculation.

[0194] By combining image processing techniques such as ROI region generation, morphological gradient extraction, polarity change screening, and least squares fitting with the existing binocular image analysis workflow, the server can not only accurately extract target boundaries from complex images but also effectively suppress image noise interference and avoid false boundaries affecting judgment, thereby improving the accuracy and stability of boundary detection. After performing geometric correction on candidate regions using preset projection transformation relationships, the generated first and second dot matrix images have high geometric consistency and boundary clarity, providing a reliable basis for subsequent calibration point extraction and error analysis. The overall workflow improves the system's robustness in dot matrix region recognition and the usability of the corrected images, making it suitable for high-precision visual calibration and image registration scenarios.

[0195] Please see Figure 7 One embodiment of the dual-target accuracy optimization device in this application includes:

[0196] The acquisition unit 701 is used to control the binocular vision system to capture the target display screen when the preset monitoring conditions are met, and to perform geometric pose correction on the images captured by the front-view camera and the oblique-view camera respectively, so as to obtain the first dot matrix image and the second dot matrix image.

[0197] Extraction unit 702 is used to extract coordinate sets from the first dot matrix image and the second dot matrix image respectively, so as to obtain the first dot matrix coordinate set and the second dot matrix coordinate set;

[0198] Transformation unit 703 is used to transform the first matrix coordinate set using a pre-stored projection transformation matrix to obtain the third matrix coordinate set;

[0199] The calculation unit 704 is used to calculate the mapping accuracy error between the second matrix coordinate set and the third matrix coordinate set;

[0200] The determination unit 705 is used to determine whether to recalibrate the binocular vision system based on the mapping accuracy error.

[0201] In this embodiment, the function of the dual-target accuracy optimization device is the same as described above. Figures 1 to 6 The steps in the illustrated embodiments are the same and will not be repeated here.

[0202] Please see Figure 8 Another embodiment of the dual-target accuracy optimization device in this application includes:

[0203] Processor 801, memory 802, input / output unit 803, and bus 804;

[0204] The processor 801 is connected to the memory 802, the input / output unit 803, and the bus 804;

[0205] The memory 802 stores a program, which the processor 801 calls to execute. Figures 1 to 6 The steps in the illustrated embodiment.

[0206] In this embodiment, the function of processor 801 is the same as described above. Figures 1 to 6 The steps in the illustrated embodiments are the same and will not be repeated here.

[0207] This application also provides a computer-readable storage medium on which a program is stored. When the program is executed on a computer, it causes the computer to perform the aforementioned actions. Figures 1 to 6 The method in any possible implementation.

[0208] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0210] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0211] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0212] If the integrated unit is implemented as 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 technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A dual-objective precision optimization method, characterized in that, The method comprises the following steps: controlling the binocular vision system to capture the target display screen when a preset monitoring condition is met, and performing geometric pose correction on images captured by the orthovision camera and the skew vision camera respectively to obtain a first dot array image and a second dot array image, wherein the first dot array image corresponds to the skew vision camera, and the second dot array image corresponds to the orthovision camera; extracting coordinate sets of the first dot array image and the second dot array image respectively to obtain a first dot array coordinate set and a second dot array coordinate set; transforming the first dot array coordinate set through a pre-stored projection transformation matrix to obtain a third dot array coordinate set; calculating a mapping precision error between the second dot array coordinate set and the third dot array coordinate set; determining whether to recalibrate the binocular vision system according to the mapping precision error.

2. The method of claim 1, wherein, Before the step of controlling the binocular vision system to capture the target display screen, the method further comprises the following steps: obtaining a fourth dot array image and a fifth dot array image obtained by the binocular vision system capturing the target display screen, wherein the fourth dot array image corresponds to the skew vision camera, and the fifth dot array image corresponds to the orthovision camera; the fourth dot array image and the fifth dot array image are images that have been subjected to geometric pose correction; extracting coordinate sets of the fourth dot array image and the fifth dot array image respectively to obtain a fourth dot array coordinate set and a fifth dot array coordinate set; calculating a projection transformation matrix according to the fourth dot array coordinate set and the fifth dot array coordinate set.

3. The method of claim 2, wherein, The step of calculating a projection transformation matrix according to the fourth dot array coordinate set and the fifth dot array coordinate set comprises the following steps: calculating a projection transformation matrix according to a projection transformation formula, the fourth dot array coordinate set and the fifth dot array coordinate set; wherein the projection transformation formula is: wherein H is the projection transformation matrix.

4. The method of claim 2, wherein, The step of calculating a mapping precision error between the second dot array coordinate set and the third dot array coordinate set comprises the following steps: calculating intrinsic parameters of the skew vision camera and the orthovision camera according to the fourth dot array coordinate set and the fifth dot array coordinate set; calculating epipolar errors between the first dot array coordinate set and the second dot array coordinate set according to the intrinsic parameters of the skew vision camera and the orthovision camera; calculating a first geometric distance between the second dot array coordinate set and the third dot array coordinate set; calculating a re-projection precision error according to the first geometric distance; calculating a mapping precision error according to the epipolar errors and the re-projection precision error.

5. The method of claim 4, wherein, The step of calculating epipolar errors between the first dot array coordinate set and the second dot array coordinate set according to the intrinsic parameters of the skew vision camera and the orthovision camera comprises the following steps: calculating a fundamental matrix based on epipolar geometric constraints through the fourth dot array coordinate set and the fifth dot array coordinate set; calculating a first epipolar line of a first target point on the first dot array coordinate set in the second dot array image according to the fundamental matrix; calculating a second epipolar line of a second target point on the second dot array coordinate set in the first dot array image according to the fundamental matrix, the first target point and the second target point being corresponding points; calculating a first distance according to the first epipolar line and the second target point; calculating a second distance according to the second epipolar line and the first target point; Calculate epipolar error according to the first distance and the second distance.

6. The method of claim 4, wherein, The calculating re-projection accuracy error according to the first geometric distance comprises: Calculate full-image re-projection accuracy error according to the first geometric distance; According to the same division, the second point array coordinate set and the third point array coordinate set are divided into m*n non-overlapping regions respectively; Calculate second geometric distance in each region respectively; Calculate regional re-projection accuracy error according to each second geometric distance; Determine re-projection accuracy error according to the full-image re-projection accuracy error and the regional re-projection accuracy error.

7. The method according to any one of claims 1 to 6, characterized in that, The geometric pose correction of the images taken by the front-view camera and the oblique-view camera respectively to obtain the first point array image and the second point array image, comprising: Obtain the oblique-view image and the front-view image taken by the binocular vision system, wherein the oblique-view image corresponds to the oblique-view camera, and the front-view image corresponds to the front-view camera; Determine the oblique-view boundary contour line of the oblique-view image and the front-view boundary contour line of the front-view image respectively; Obtain the pre-stored oblique-view projection conversion relationship and the front-view projection conversion relationship; According to the oblique-view projection conversion relationship, the geometric pose correction is performed on the region surrounded by the oblique-view boundary contour line to obtain the first point array image, and according to the front-view projection conversion relationship, the geometric pose correction is performed on the region surrounded by the front-view boundary contour line to obtain the second point array image.

8. The method of claim 7, wherein, The determination of the oblique-view boundary contour line of the oblique-view image and the front-view boundary contour line of the front-view image respectively, comprising: Generate an oblique-view initial search box based on the oblique-view image, and generate a front-view initial search box based on the front-view image; Determine the oblique-view candidate points that meet the gradient threshold and conform to the polarity change from the oblique-view initial search box, and determine the front-view candidate points that meet the gradient threshold and conform to the polarity change from the front-view initial search box; Fit the oblique-view candidate points and the front-view candidate points respectively by the least square method to determine the oblique-view boundary contour line and the front-view boundary contour line.

9. A dual-objective precision optimization apparatus, characterized by comprising: The device is used to execute the method in any one of claims 1 to 8, and the device comprises: An acquisition unit is configured to control the binocular vision system to take the target display screen when a preset monitoring condition is met, and perform geometric pose correction on the images taken by the front-view camera and the oblique-view camera respectively to obtain a first point array image and a second point array image, wherein the first point array image corresponds to the oblique-view camera, and the second point array image corresponds to the front-view camera; An extraction unit is configured to extract a first point array coordinate set and a second point array coordinate set from the first point array image and the second point array image respectively; A transformation unit is configured to transform the first point array coordinate set by a pre-stored projection transformation matrix to obtain a third point array coordinate set; A calculation unit is configured to calculate the mapping accuracy error of the second point array coordinate set and the third point array coordinate set; A determination unit is configured to determine whether to re-calibrate the binocular vision system according to the mapping accuracy error.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium has saved thereon a program which, when executed on a computer, causes the computer to perform the method of any one of claims 1 to 8.

Images