Multi-camera high-precision plane vision system calibration method and system

Abstract

The invention relates to the field of machine vision detection, in particular to a multi-camera high-precision plane vision system calibration method, which comprises the following steps of: S1, generating a plane calibration plate graph through a calibration plate graph generation method according to the field of view, the number of cameras and the single-pixel precision of a plane vision system; s2, manufacturing a plane calibration plate according to the plane calibration plate graph, and shooting and imaging the plane calibration plate by using each camera of a plane vision system to obtain a plane calibration plate image; s3, for the plane calibration plate image and the calibration plate graph, determining coordinate transformation from an image space of each camera of the plane vision system to a calibration plate graph model space by using a coarse alignment calibration method; and S4, carrying out coordinate transformation on the plane calibration plate image, the calibration plate graph and the image space of each camera of the plane vision system obtained by the coarse alignment calibration to the calibration plate graph model space, and determining the external parameters of each camera of the plane vision system and the coordinate transformation from the image space of each camera to the calibration plate graph model space by using a fine alignment calibration method.

Description

Technical Field

[0001] The present invention relates to the field of machine vision detection, and in particular to a multi-camera high-precision planar vision system calibration method and system. Background Art

[0002] A multi-camera high-precision plane vision system is a system that uses multiple cameras for image acquisition and processing, aiming to improve the accuracy and efficiency of plane measurement and detection. The system usually includes multiple cameras, calibration plates, data processing algorithms and other components, which work together to achieve high-precision plane vision measurement and detection. The multi-camera high-precision plane vision system can significantly improve the accuracy and efficiency of plane measurement and detection through the collaborative work of multiple cameras, efficient calibration methods and advanced data processing algorithms, and is widely used in industrial inspection, autonomous driving, scientific research and other fields.

[0003] In a multi-camera high-precision planar vision system, the shooting postures (including height and angle) of each camera need to be kept as consistent as possible when taking pictures, and the target coordinates in the image space of each camera need to be converted into a unified coordinate system in image data processing. All of these need to be achieved through calibration. The current multi-camera high-precision planar vision system calibration method has the problems of insufficient flexibility and robustness in application:

[0004] 1. There are significant differences in the number of cameras, overall field of view, and single-pixel accuracy of multi-camera high-precision planar vision systems in different application scenarios. The calibration plate designed for a specific multi-camera high-precision planar vision system often fails in other systems, and there is a lack of a simple and universal calibration plate design method.

[0005] 2. During the camera installation and debugging phase of the multi-camera high-precision planar vision system, there will be significant differences in the shooting postures of each camera, resulting in calibration failure or incorrect calibration results.

[0006] The above problems bring high labor costs and economic burdens to the production and use of equipment containing multi-camera high-precision planar vision systems. Therefore, it is necessary to propose a new multi-camera high-precision planar vision system calibration method to solve the above two outstanding problems. Summary of the invention

[0007] In view of the insufficient flexibility and robustness of the current multi-camera high-precision planar vision system calibration method, the present invention provides a multi-camera high-precision planar vision system calibration method and system to achieve accurate, easy-to-use and reliable multi-camera extrinsic calibration.

[0008] The first aspect of the present invention relates to a multi-camera high-precision planar vision system calibration method, which is used to determine the external parameters of each camera in the system and the coordinate transformation from its image space to the calibration board graphic model space, and specifically includes the following steps:

[0009] S1: Generate a plane calibration plate graphic by using a calibration plate graphic generation method according to the field of view, number of cameras and single pixel accuracy of the plane vision system;

[0010] S2: making a plane calibration plate according to the plane calibration plate pattern, and using each camera of the plane vision system to photograph the plane calibration plate to obtain an image of the plane calibration plate;

[0011] S3: for the plane calibration plate image and the calibration plate pattern, a coarse alignment calibration method is used to determine the coordinate transformation from the image space of each camera of the plane vision system to the calibration plate pattern model space;

[0012] S4: coordinate transformation from the image space of each camera of the planar vision system to the calibration plate graphic model space obtained by the planar calibration plate image, the calibration plate graphic and the rough alignment calibration, and use the fine alignment calibration method to determine the external parameters of each camera of the planar vision system and the coordinate transformation from its image space to the calibration plate graphic model space.

[0013] According to the above technical solution, the specific steps of the calibration plate graphics generation method in S1 are:

[0014] S1.1: According to the field of view of the planar vision system, the number of cameras and the single pixel accuracy, determine the calibration plate size, feature point radius, feature point array range size, and the minimum distance d of the nearest neighbor feature point. min , the maximum distance d of the nearest neighbor feature point max and calibration plate graphics generation parameters of the number of feature points contained in the feature point matrix;

[0015] S1.2: Randomly generate a feature point set within the range of the feature point matrix according to the number of feature points contained in the feature point matrix;

[0016] S1.3: Traverse each feature point in the feature point set, calculate the distance between it and the nearest neighbor feature point, and remove the nearest neighbor feature points in the feature point set whose distance is less than the minimum distance d of the nearest neighbor feature point. min The characteristic points of

[0017] S1.4: Check whether the distance between the nearest neighbor feature points of all feature points in the feature point set is greater than the maximum distance d between the nearest neighbor feature points max , if yes, go to S1.2, otherwise continue to S1.5;

[0018] S1.5: Calculate the convex hull of the feature point set to obtain a convex hull point set containing n feature points;

[0019] S1.6: Traverse each feature point in the convex hull point set, record the length of the n-segment side of the convex hull polygon starting from the point in a counterclockwise direction to obtain a feature vector set containing n n-dimensional feature vectors;

[0020] S1.7: Check the distance between the n-dimensional feature vectors in the feature vector set. If there are two n-dimensional feature vectors whose similarity is greater than the preset threshold, Then go to S1.2, otherwise continue to S1.8;

[0021] S1.8: Copy and translate the feature point array composed of the feature point set, and multiple feature point arrays are greater than d max The fixed spacing is arranged horizontally and vertically to form a feature point grid covering the entire plane calibration plate;

[0022] S1.9: Add a unique ID to each feature point in the feature point grid to form a plane calibration plate graphic.

[0023] According to the above technical solution, the specific steps of the rough alignment calibration method in S3 are:

[0024] S3.1: Select a calibration plate image taken by a camera, extract the centers of all feature points in the image to obtain a feature center point set;

[0025] S3.2: Take the points in the feature center as vertices, and the distance is less than d max Add edges to the point pairs to form a feature graph;

[0026] S3.3: extract the connected components in the feature map, each connected component corresponds to a feature point matrix, identify the ID of the feature point matrix from the image, discard the connected components whose feature point matrix ID identification fails, and discard the connected components whose vertex number is not equal to the number of feature points in the feature point matrix;

[0027] S3.4: Select a feature point array that has been successfully identified and not discarded, extract the convex hull point set of the feature center point set of the feature point array, traverse each feature point in the convex hull point set, and record the n-segment side lengths of the convex hull polygon starting from the point in a counterclockwise direction to obtain a calibration plate image feature vector set containing n n-dimensional feature vectors;

[0028] S3.5: extract the convex hull point set of the feature point set matching the feature point matrix ID in the calibration plate figure, traverse each feature point in the convex hull point set, and record the n-segment side lengths of the convex hull polygon starting from the point in a counterclockwise direction to obtain a calibration plate figure feature vector set containing n n-dimensional feature vectors;

[0029] S3.6: perform pairwise matching of the feature vector set of the feature calibration plate image and the feature vectors in the calibration plate graphic feature vector set to calculate similarity, and determine the maximum value of the feature vector similarity as the current feature dot matrix image-graphic similarity;

[0030] S3.7: Repeat S3.4 to S3.6 until all the feature points that have been successfully identified and not discarded are traversed, and the feature point ID with the largest feature point image-graphic similarity is determined;

[0031] S3.8: According to the feature dot matrix ID of the maximum feature dot matrix image-graphic similarity, determine the matching relationship between its feature center point set and the feature point set, and calculate the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate according to the matched point pairs;

[0032] S3.9: Repeat S3.1 to S3.8 until the calibration plate images taken by all cameras are traversed to obtain the coordinate transformation from the image space of each camera of the planar vision system to the calibration plate graphic model space.

[0033] According to the above technical solution, the specific steps of the precise alignment calibration method in S4 are:

[0034] S4.1: For each feature point in the calibration plate graphic feature point set, find its k nearest neighbor feature points in the calibration plate graphic feature point set, calculate the unit direction vector of the plane vector formed by the feature point and each corresponding k nearest neighbor mapping feature center point, sort the unit direction vectors by direction and splice them into a 2k-dimensional k nearest neighbor feature vector;

[0035] S4.2: Select a calibration plate image taken by a camera, and according to the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate obtained by the coarse alignment calibration, map each feature center point in the extracted feature center point set to the graphic model space of the calibration plate to obtain a mapped feature center point set;

[0036] S4.3: for each mapping feature center point in the mapping feature center point set, find its k nearest neighbor mapping feature center points in the mapping feature center point set, respectively calculate the unit direction vector of the plane vector formed by the mapping feature center point and each of its corresponding k nearest neighbor mapping feature center points, sort the unit direction vectors by direction and concatenate them into a 2k-dimensional k nearest neighbor feature vector;

[0037] S4.4: For each mapping feature center point in the mapping feature center point set, find the calibration plate graphic feature point set whose distance from it is less than the preset threshold d c The calibration plate graphic feature points are formed into a candidate matching feature point set of the mapping feature center point;

[0038] S4.5: For each mapping feature center point in the mapping feature center point set, respectively calculate the similarity between its k nearest neighbor feature vector and the k nearest neighbor feature vector of each candidate matching feature point in the corresponding candidate matching feature point set, determine the candidate matching feature point with the highest similarity as the best matching feature point, and discard the candidate matching feature point with a similarity lower than a preset threshold. The mapping feature center point matching result is obtained to obtain the matching result of the mapping feature center point set and the calibration plate graphic feature point set;

[0039] S4.6: Calculate the matching result of the calibration plate image feature center point set and the calibration plate graphic feature point set according to the matching result of the mapping feature center point set and the calibration plate graphic feature point set;

[0040] S4.7: Calculate the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate according to the matching result of the feature center point set of the calibration plate image and the feature point set of the calibration plate graphic;

[0041] S4.8: Repeat S4.2 to S4.7 until the calibration plate images taken by all cameras are traversed, and the external parameters of each camera of the planar vision system and the coordinate transformation from its image space to the calibration plate graphic model space are obtained.

[0042] A second aspect of the present invention relates to a multi-camera high-precision plane vision system, comprising the above-mentioned multi-camera high-precision plane vision system calibration method, a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code.

[0043] The present invention automatically generates a plane calibration plate graphic according to the design parameters of the camera high-precision plane vision system through a plane calibration plate graphic generation method, and realizes accurate, easy-to-use and reliable multi-camera extrinsic parameter calibration by sequentially performing coarse alignment calibration and fine alignment calibration processes. The present invention discloses a multi-camera high-precision plane vision system calibration method, and its beneficial effects include three aspects:

[0044] The present invention uses a calibration plate graphic generation method to automatically generate a plane calibration plate graphic according to the design parameters of a multi-camera high-precision plane vision system, and can automatically generate applicable plane calibration plate graphics for multi-camera high-precision plane vision systems in different application scenarios, thereby improving the flexibility and ease of use of the calibration method;

[0045] The present invention performs coarse-grained alignment matching and calibration through a coarse alignment calibration method, constructs a feature vector of a convex hull point set for feature matching, and can avoid calibration failure or error caused by posture differences of multiple cameras, thereby improving the accuracy and reliability of calibration.

[0046] The present invention performs fine alignment matching and calibration through a precise alignment calibration method, constructs a k-nearest neighbor feature vector for feature matching, compares and screens candidate matches, and utilizes all successfully matched feature points within the camera field of view for calibration calculation, thereby further improving the precision and accuracy of the calibration.

[0047] Based on the above three aspects, the present invention realizes accurate, easy-to-use and reliable multi-camera extrinsic calibration of a multi-camera high-precision planar vision system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the plane calibration plate of the present invention.

[0049] Figure 2 It is a schematic diagram of the construction of the feature vector of the convex hull point set of a feature point of the present invention.

[0050] Figure 3 The figure is a schematic diagram of the k nearest neighbor feature vectors of a feature point of the present invention.

[0051] Figure 4 It is a schematic diagram of the process of the present invention.

[0052] Figure 5 Schematic diagram of the flow of the calibration plate graphics generation method.

[0053] Figure 6 Schematic diagram of the process of rough alignment calibration method.

[0054] Figure 7 Schematic diagram of the process of precise alignment calibration method. DETAILED DESCRIPTION

[0055] The present invention discloses a multi-camera high-precision planar vision system calibration method. The specific implementation of the present invention is further described below in conjunction with preferred embodiments and with reference to the accompanying drawings.

[0056] Example 1

[0057] Reference Figure 1-Figure 7 The chip bracket appearance defect detection template matching and positioning method of the present invention comprises the following steps:

[0058] S1: Generate a plane calibration plate graphic by using a calibration plate graphic generation method according to the field of view, number of cameras and single pixel accuracy of the plane vision system;

[0059] S2: making a plane calibration plate according to the plane calibration plate pattern, and using each camera of the plane vision system to photograph the plane calibration plate to obtain an image of the plane calibration plate;

[0060] S3: for the plane calibration plate image and the calibration plate pattern, a coarse alignment calibration method is used to determine the coordinate transformation from the image space of each camera of the plane vision system to the calibration plate pattern model space;

[0061] S4: coordinate transformation from the image space of each camera of the planar vision system to the calibration plate graphic model space obtained by the planar calibration plate image, the calibration plate graphic and the rough alignment calibration, and use the fine alignment calibration method to determine the external parameters of each camera of the planar vision system and the coordinate transformation from its image space to the calibration plate graphic model space.

[0062] like Figure 1 As shown, the specific steps of the calibration plate graphics generation method in the above S1 are:

[0063] S1.1: According to the field of view of the planar vision system, the number of cameras and the single pixel accuracy, determine the calibration plate size, feature point radius, feature point array range size, and the minimum distance d of the nearest neighbor feature point. min , the maximum distance d of the nearest neighbor feature point max and calibration plate graphics generation parameters of the number of feature points contained in the feature point matrix;

[0064] S1.2: Randomly generate a feature point set within the range of the feature point matrix according to the number of feature points contained in the feature point matrix;

[0065] S1.3: Traverse each feature point in the feature point set, calculate the distance between it and the nearest neighbor feature point, and remove the nearest neighbor feature points in the feature point set whose distance is less than the minimum distance d of the nearest neighbor feature point. min The characteristic points of

[0066] S1.4: Check whether the distance between the nearest neighbor feature points of all feature points in the feature point set is greater than the maximum distance d between the nearest neighbor feature points max , if yes, go to S1.2, otherwise continue to S1.5;

[0067] S1.5: Calculate the convex hull of the feature point set to obtain a convex hull point set containing n feature points;

[0068] S1.6: If Figure 2 As shown, traverse each feature point in the convex hull point set, record the length of the n-segment side of the convex hull polygon starting from the point in a counterclockwise direction to obtain a feature vector set containing n n-dimensional feature vectors;

[0069] S1.7: Check the distance between the n-dimensional feature vectors in the feature vector set. If there are two n-dimensional feature vectors whose similarity is greater than the preset threshold, Then go to S1.2, otherwise continue to S1.8;

[0070] S1.8: Copy and translate the feature point array composed of the feature point set, and multiple feature point arrays are greater than d max The fixed spacing is arranged horizontally and vertically to form a feature point grid covering the entire plane calibration plate;

[0071] S1.9: Add a unique ID to each feature point in the feature point grid to form a plane calibration plate graphic.

[0072] Preferably, the specific method of determining the calibration plate graphics generation parameters according to the field of view of the planar vision system, the number of cameras and the single-pixel accuracy of S1.1 is that the plate size is set to a size slightly larger than the field of view of the planar vision system, the feature point radius is 1 / 100 of the width or height of the single camera target plane field of view, the feature point array range size is set to 1 / 3 of the width and height of the single camera target plane field of view to ensure that there is at least one complete feature point array within the camera field of view, and the number of feature points contained in the feature point array is set to the maximum distance d of the nearest neighbor feature point. max The number of feature points embedded most densely in the feature point matrix range is the minimum distance d of the nearest neighbor feature point. min The value between the number of the most densely embedded feature points in the feature point array.

[0073] Preferably, the feature points generated in S1.2 are solid circles.

[0074] Preferably, the specific method of calculating the convex hull of the feature point set in S1.5 to obtain a convex hull point set containing n feature points is a Graham scanning algorithm.

[0075] Preferably, the specific method of adding a unique ID identification to each feature dot matrix in the feature dot matrix grid in S1.9 is to add a digital character identification feature array ID within the range of each feature dot matrix.

[0076] The specific steps of the rough alignment calibration method in S3 are as follows:

[0077] S3.1: Select a calibration plate image taken by a camera, extract the centers of all feature points in the image to obtain a feature center point set;

[0078] S3.2: Take the points in the feature center as vertices, and the distance is less than d max Add edges to the point pairs to form a feature graph;

[0079] S3.3: extract the connected components in the feature map, each connected component corresponds to a feature point matrix, identify the ID of the feature point matrix from the image, discard the connected components whose feature point matrix ID identification fails, and discard the connected components whose vertex number is not equal to the number of feature points in the feature point matrix;

[0080] S3.4: Select a feature point array that has been successfully identified and not discarded, extract the convex hull point set of the feature center point set of the feature point array, traverse each feature point in the convex hull point set, and record the n-segment side lengths of the convex hull polygon starting from the point in a counterclockwise direction to obtain a calibration plate image feature vector set containing n n-dimensional feature vectors;

[0081] S3.5: extract the convex hull point set of the feature point set matching the feature point matrix ID in the calibration plate figure, traverse each feature point in the convex hull point set, and record the n-segment side lengths of the convex hull polygon starting from the point in a counterclockwise direction to obtain a calibration plate figure feature vector set containing n n-dimensional feature vectors;

[0082] S3.6: perform pairwise matching of the feature vector set of the feature calibration plate image and the feature vectors in the calibration plate graphic feature vector set to calculate similarity, and determine the maximum value of the feature vector similarity as the current feature dot matrix image-graphic similarity;

[0083] S3.7: Repeat S3.4 to S3.6 until all the feature points that have been successfully identified and not discarded are traversed, and the feature point ID with the largest feature point image-graphic similarity is determined;

[0084] S3.8: According to the feature dot matrix ID of the maximum feature dot matrix image-graphic similarity, determine the matching relationship between its feature center point set and the feature point set, and calculate the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate according to the matched point pairs;

[0085] S3.9: Repeat S3.1 to S3.8 until the calibration plate images taken by all cameras are traversed to obtain the coordinate transformation from the image space of each camera of the planar vision system to the calibration plate graphic model space.

[0086] Preferably, the specific method of extracting the centers of all feature points in the image in S3.1 is to use Otsu's method to threshold to obtain a binary image, extract the connected components of the binary image to obtain the area where each feature point is located, and then use the centroid method to extract the sub-pixel feature point centers.

[0087] The specific steps of the precise alignment calibration method in S4 are as follows:

[0088] S4.1: If Figure 3 As shown, for each feature point in the calibration plate graphic feature point set, find its k nearest neighbor feature points in the calibration plate graphic feature point set, calculate the unit direction vector of the plane vector formed by the feature point and each corresponding k nearest neighbor mapping feature center point, sort the unit direction vectors by direction and splice them into a 2k-dimensional k nearest neighbor feature vector;

[0089] S4.2: Select a calibration plate image taken by a camera, and according to the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate obtained by the coarse alignment calibration, map each feature center point in the extracted feature center point set to the graphic model space of the calibration plate to obtain a mapped feature center point set;

[0090] S4.3: for each mapping feature center point in the mapping feature center point set, find its k nearest neighbor mapping feature center points in the mapping feature center point set, respectively calculate the unit direction vector of the plane vector formed by the mapping feature center point and each of its corresponding k nearest neighbor mapping feature center points, sort the unit direction vectors by direction and concatenate them into a 2k-dimensional k nearest neighbor feature vector;

[0091] S4.4: For each mapping feature center point in the mapping feature center point set, find the calibration plate graphic feature point set whose distance from it is less than the preset threshold d c The calibration plate graphic feature points are formed into a candidate matching feature point set of the mapping feature center point;

[0092] S4.5: For each mapping feature center point in the mapping feature center point set, respectively calculate the similarity between its k nearest neighbor feature vector and the k nearest neighbor feature vector of each candidate matching feature point in the corresponding candidate matching feature point set, determine the candidate matching feature point with the highest similarity as the best matching feature point, and discard the candidate matching feature point with a similarity lower than a preset threshold. The mapping feature center point matching result is obtained to obtain the matching result of the mapping feature center point set and the calibration plate graphic feature point set;

[0093] S4.6: Calculate the matching result of the calibration plate image feature center point set and the calibration plate graphic feature point set according to the matching result of the mapping feature center point set and the calibration plate graphic feature point set;

[0094] S4.7: Calculate the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate according to the matching result of the feature center point set of the calibration plate image and the feature point set of the calibration plate graphic;

[0095] S4.8: Repeat S4.2 to S4.7 until the calibration plate images taken by all cameras are traversed, and the external parameters of each camera of the planar vision system and the coordinate transformation from its image space to the calibration plate graphic model space are obtained.

[0096] Embodiment 2:

[0097] The present embodiment relates to a multi-camera high-precision plane vision system calibration method and device, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement a multi-camera high-precision plane vision system calibration method of the present embodiment 1.

[0098] Embodiment 3:

[0099] The present embodiment relates to a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a multi-camera high-precision planar vision system calibration method of embodiment 1 is implemented.

[0100] The contents described in the embodiments of this specification are merely an enumeration of implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A multi-camera high-precision planar vision system calibration method, characterized by: S1: Generate a plane calibration plate graphic by using a calibration plate graphic generation method according to the field of view, number of cameras and single pixel accuracy of the plane vision system; S2: making a plane calibration plate according to the plane calibration plate pattern, and using each camera of the plane vision system to photograph the plane calibration plate to obtain an image of the plane calibration plate; S3: for the plane calibration plate image and the calibration plate pattern, a coarse alignment calibration method is used to determine the coordinate transformation from the image space of each camera of the plane vision system to the calibration plate pattern model space; S4: coordinate transformation from the image space of each camera of the planar vision system to the calibration plate graphic model space obtained by the planar calibration plate image, the calibration plate graphic and the rough alignment calibration, and use the fine alignment calibration method to determine the external parameters of each camera of the planar vision system and the coordinate transformation from its image space to the calibration plate graphic model space.

2. The method for calibrating a multi-camera high-precision planar vision system according to claim 1, characterized in that: The calibration plate graphics generation method of S1 specifically includes: S1.1: According to the field of view of the planar vision system, the number of cameras and the single pixel accuracy, determine the calibration plate size, feature point radius, feature point array range size, and the minimum distance d of the nearest neighbor feature point. min , the maximum distance d of the nearest neighbor feature point max and calibration plate graphics generation parameters of the number of feature points contained in the feature point matrix; S1.2: Randomly generate a feature point set within the range of the feature point matrix according to the number of feature points contained in the feature point matrix; S1.3: Traverse each feature point in the feature point set, calculate the distance between it and the nearest neighbor feature point, and remove the nearest neighbor feature points in the feature point set whose distance is less than the minimum distance d of the nearest neighbor feature point. min The characteristic points of S1.4: Check whether the distance between the nearest neighbor feature points of all feature points in the feature point set is greater than the maximum distance d between the nearest neighbor feature points max , if yes, go to S1.2, otherwise continue to S1.5; S1.5: Calculate the convex hull of the feature point set to obtain the The convex hull point set of feature points; S1.6: Traverse each feature point in the convex hull point set and record the convex hull polygon starting from this point in a counterclockwise direction. The segment length is included indivual The set of feature vectors of dimension feature vectors; S1.7: Check the feature vector set The distance between two eigenvectors is The similarity of the feature vectors is greater than the preset threshold Then go to S1.2, otherwise continue to S1.8; S1.8: Copy and translate the feature point array composed of the feature point set, and multiple feature point arrays are greater than d max The fixed spacing is arranged horizontally and vertically to form a feature point grid covering the entire plane calibration plate; S1.9: Add a unique ID to each feature point in the feature point grid to form a plane calibration plate graphic.

3. The method for calibrating a multi-camera high-precision planar vision system according to claim 1, characterized in that: The rough alignment calibration method of S3 specifically includes: S3.1: Select a calibration plate image taken by a camera, extract the centers of all feature points in the image to obtain a feature center point set; S3.2: Take the points in the feature center as vertices, and the distance is less than d max Add edges to the point pairs to form a feature graph; S3.3: extract the connected components in the feature map, each connected component corresponds to a feature point matrix, identify the ID of the feature point matrix from the image, discard the connected components whose feature point matrix ID identification fails, and discard the connected components whose vertex number is not equal to the number of feature points in the feature point matrix; S3.4: Select a feature point matrix that has been successfully identified and not discarded, extract the convex hull point set of the feature center point set of the feature point matrix, traverse each feature point in the convex hull point set, and record the convex hull polygon with this point as the starting point in a counterclockwise direction. The segment length is included indivual The calibration plate image feature vector set of dimensional feature vector; S3.5: Extract the convex hull point set of the feature point set matching the feature point matrix ID in the calibration plate graphics, traverse each feature point in the convex hull point set, and record the convex hull polygon with the point as the starting point in a counterclockwise direction. The segment length is included indivual dimensional feature vector of the calibration plate graphics feature vector set; S3.6: perform pairwise matching of the feature vector set of the feature calibration plate image and the feature vectors in the calibration plate graphic feature vector set to calculate similarity, and determine the maximum value of the feature vector similarity as the current feature dot matrix image-graphic similarity; S3.7: Repeat S3.4 to S3.6 until all the feature points that have been successfully identified and not discarded are traversed, and the feature point ID with the largest feature point image-graphic similarity is determined; S3.8: According to the feature dot matrix ID of the maximum feature dot matrix image-graphic similarity, determine the matching relationship between its feature center point set and the feature point set, and calculate the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate according to the matched point pairs; S3.9: Repeat S3.1 to S3.8 until the calibration plate images taken by all cameras are traversed to obtain the coordinate transformation from the image space of each camera of the planar vision system to the calibration plate graphic model space.

4. The method for calibrating a multi-camera high-precision planar vision system according to claim 1, characterized in that: The precise alignment calibration method of S4 specifically includes: S4.1: For each feature point in the calibration plate graphic feature point set, find its k nearest neighbor feature points in the calibration plate graphic feature point set, calculate the unit direction vector of the plane vector formed by the feature point and each corresponding k nearest neighbor mapping feature center point, sort the unit direction vectors by direction and splice them into a 2k-dimensional k nearest neighbor feature vector; S4.2: Select a calibration plate image taken by a camera, and according to the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate obtained by the coarse alignment calibration, map each feature center point in the extracted feature center point set to the graphic model space of the calibration plate to obtain a mapped feature center point set; S4.3: for each mapping feature center point in the mapping feature center point set, find its k nearest neighbor mapping feature center points in the mapping feature center point set, respectively calculate the unit direction vector of the plane vector formed by the mapping feature center point and each of its corresponding k nearest neighbor mapping feature center points, sort the unit direction vectors by direction and concatenate them into a 2k-dimensional k nearest neighbor feature vector; S4.4: For each mapping feature center point in the mapping feature center point set, find the calibration plate graphic feature point set whose distance from it is less than the preset threshold d c The calibration plate graphic feature points are formed into a candidate matching feature point set of the mapping feature center point; S4.5: For each mapping feature center point in the mapping feature center point set, respectively calculate the similarity between its k nearest neighbor feature vector and the k nearest neighbor feature vector of each candidate matching feature point in the corresponding candidate matching feature point set, determine the candidate matching feature point with the highest similarity as the best matching feature point, and discard the candidate matching feature point with a similarity lower than a preset threshold. The mapping feature center point matching result is obtained to obtain the matching result of the mapping feature center point set and the calibration plate graphic feature point set; S4.6: Calculate the matching result of the calibration plate image feature center point set and the calibration plate graphic feature point set according to the matching result of the mapping feature center point set and the calibration plate graphic feature point set; S4.7: Calculate the coordinate transformation from the image space of the current camera to the graphic model space of the calibration plate according to the matching result of the feature center point set of the calibration plate image and the feature point set of the calibration plate graphic; S4.8: Repeat S4.2 to S4.7 until the calibration plate images taken by all cameras are traversed, and the external parameters of each camera of the planar vision system and the coordinate transformation from its image space to the calibration plate graphic model space are obtained.

5. A multi-camera high-precision planar vision system, characterized in that: It comprises a multi-camera high-precision plane vision system calibration method as described in claims 1-4, a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code.

Images