Calibration plate feature point detection method, calibration method and detection system
By using edge detection and fitting processing in the panoramic image system combined with the neighbor point pair algorithm, the problem of insufficient feature point recognition in the fisheye camera image is solved, and the adaptability and accuracy of the detection are improved.
Patent Information
- Application Number
- CN202411970076.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-11
AI Technical Summary
The traditional feature point detection algorithm is not effective in the fisheye camera images dedicated to the panoramic image system, and cannot effectively identify the feature points of the calibration board, resulting in insufficient accuracy of camera calibration.
A calibration plate feature point detection method is adopted to obtain the vertices of the fitted quadrilateral through edge detection and fitting processing, and combine the neighbor point pair algorithm and preset feature parameters to filter and obtain the target feature points.
It improves the adaptability, recognition and accuracy of feature point detection, and can effectively identify feature points in complex visual systems such as fisheye cameras under different imaging conditions, enhancing the reliability of the calibration process.
Smart Images
Figure CN119810133A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vehicle-mounted panoramic image camera calibration, and in particular to a calibration plate feature point detection method, a calibration method and a detection system. Background Art
[0002] Currently, the panoramic imaging system is based on four surround-view fisheye cameras on the vehicle body and is implemented by executing a series of image processing algorithms. In actual application, there are many parts that can be optimized in the implementation of panoramic imaging, one of which is camera calibration.
[0003] The accuracy of camera calibration is related to the visibility and accuracy of the panoramic imaging system. The key step to whether the calibration result is accurate is to detect the feature points of the calibration plate (i.e., the corner points of the calibration plate). When calibrating the camera, for images captured by ordinary cameras, traditional feature point detection algorithms such as Harris and other algorithms have good recognition capabilities for the feature points of the calibration plate in the image. However, for the fisheye camera dedicated to the panoramic imaging system with a large imaging field of view and obvious distortion at the edge of the picture, the black and white grids of the calibration plate in the image are large near and small far away, straight near and curved far away. For such fisheye images, the effect of traditional feature point detection algorithms is greatly reduced, and even no feature points can be recognized. Therefore, for the fisheye camera calibration step of the panoramic imaging system, the traditional feature point detection algorithm has poor applicability and insufficient feature point recognition, which needs further improvement. Summary of the invention
[0004] The present invention aims to provide a calibration plate feature point detection method, a calibration method and a detection system to solve the above-mentioned technical problems and to improve the adaptability, recognition and accuracy of feature point detection.
[0005] In order to solve the above technical problems, the present invention provides a method for detecting characteristic points of a calibration plate, comprising the following steps:
[0006] Perform edge detection and fitting processing based on the detection image to obtain a fitting quadrilateral, thereby obtaining the vertices of the quadrilateral;
[0007] Obtain target neighbor point pairs based on neighbor point pair algorithm and quadrilateral vertices;
[0008] Filter neighbor point pairs based on target neighbor point pairs and preset feature parameters to obtain target feature points;
[0009] The target feature points are processed by averaging to obtain the target feature point coordinates and complete the feature point detection of the calibration plate.
[0010] The above scheme first performs edge detection and fitting processing on the detection image to obtain a fitted quadrilateral to obtain the quadrilateral vertices, and then uses the neighbor point pair algorithm to screen and extract the obtained quadrilateral vertices to obtain the required neighbor point pairs, and finally obtains the required target feature points. Based on the provided neighbor point pair algorithm, calculation and screening are performed in combination with actual conditions. It is not limited to the detection of standard chessboard corner points in the calibration plate taken by an ordinary camera, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small imaging characteristics, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of feature point detection and improving the recognition and accuracy of detection.
[0011] Furthermore, the step performs edge detection and fitting processing based on the detection image to obtain a fitted quadrilateral, thereby obtaining the vertices of the quadrilateral, including: obtaining the original image, performing region selection and denoising filtering processing based on the original image and image preprocessing, and obtaining the detection image; performing contour detection based on the detection image and the edge detection algorithm, and performing contour extraction based on the polygon fitting algorithm to obtain an approximate contour; obtaining a fitted quadrilateral based on the approximate contour and the contour approximation algorithm, and then obtaining the quadrilateral vertices based on the fitted quadrilateral.
[0012] The above scheme uses the edge detection algorithm and the contour approximation algorithm to fit the grid of the calibration plate in the detection image, thereby detecting the fitting quadrilaterals corresponding to all grids, and obtaining the vertices of the quadrilaterals, so as to better detect all relevant corner points in the image and provide a complete data range for the subsequent feature point extraction.
[0013] Furthermore, the steps of obtaining an original image, performing region selection and denoising filtering based on the original image and image preprocessing, and obtaining a detection image include: obtaining an original image, performing region selection based on the original image and target region selection operation, and obtaining a calibration image; obtaining a grayscale image based on the calibration image and grayscale processing; obtaining a binarized image based on the grayscale image and adaptive threshold binarization processing; and performing several morphological operations and filtering processing based on the binarized image to obtain a detection image.
[0014] The above scheme is the process of image preprocessing of the original image. The calibration plate area to be detected is locked through the target area selection, and denoising and filtering are performed based on a series of image data processing operations to make the final detection image clearer and easier to identify, reducing interference from other factors.
[0015] Furthermore, the step of obtaining neighbor point pairs based on the neighbor point pair algorithm and quadrilateral vertices includes: obtaining target vertices based on quadrilateral vertices and preset boundary culling conditions; processing the target vertices according to a preset grouping method to obtain multiple vertex groups; executing a point pair acquisition step based on multiple vertex groups to obtain all neighbor point pairs; the point pair acquisition step includes: selecting one of the multiple vertex groups as the current vertex group to obtain minimum distance information based on the target vertex in the current vertex group and the target vertex in the remaining vertex groups, thereby determining the corresponding vertex group based on the minimum distance information, and forming an adjacent group based on the current vertex group and the corresponding vertex group, and then obtaining neighbor point pairs based on the adjacent groups and the preset corresponding method, and repeating the above steps until all neighbor point pairs are obtained; filtering and obtaining target neighbor point pairs based on all neighbor point pairs and the preset culling operation.
[0016] The above scheme obtains neighbor point pairs by grouping quadrilateral vertices, calculating distances, and corresponding adjacent groups based on the minimum distance, and uses a preset elimination operation to filter out target neighbor point pairs containing required feature points. Through vertex grouping and distance calculation, adjacent groups can still be identified and detected in the face of different actual situations, thereby being able to more effectively identify the required target neighbor point pairs, and can achieve detection under different imaging conditions, with higher adaptability and recognition.
[0017] A method for detecting feature points of a calibration plate provided by the present invention first performs edge detection and fitting processing on a detection image to obtain a fitted quadrilateral to obtain quadrilateral vertices, then groups the obtained quadrilateral vertices, and determines adjacent groups based on distance calculation between vertices in different groups, thereby screening and extracting required neighbor point pairs, and finally obtaining required target feature points. The method adopted is based on the neighbor point pair principle, and determines the neighbor point pairs containing feature points by calculating and confirming the minimum distance to obtain the adjacent group relationship. It is not limited to the detection of standard chessboard corner points in a calibration plate photographed by an ordinary camera, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of feature point detection and improving the recognition and accuracy of detection.
[0018] The present invention provides a calibration method for characteristic points of a calibration plate, which is implemented by adopting the calibration plate characteristic point detection method, and specifically comprises: performing calibration plate detection based on the calibration plate characteristic point detection method, acquiring target characteristic point coordinates, and thereby implementing characteristic point calibration of the calibration plate based on the target characteristic point coordinates.
[0019] The present invention provides a calibration method for characteristic points of a calibration plate, which is implemented by adopting the above-mentioned detection method for characteristic points of a calibration plate. The characteristic point detection has higher adaptability, recognition and accuracy, provides more accurate calibration coordinates for calibration, and improves the calibration efficiency and success rate.
[0020] The present invention also provides a calibration plate feature point detection system, which is used to implement the calibration plate feature point detection method, comprising:
[0021] A quadrilateral vertex acquisition module is used to perform edge detection and fitting processing based on the detection image to obtain a fitted quadrilateral, thereby obtaining quadrilateral vertices;
[0022] A neighbor point pair acquisition module is used to acquire target neighbor point pairs based on a neighbor point pair algorithm and quadrilateral vertices;
[0023] A feature point pair screening module is used to screen neighbor point pairs based on target neighbor point pairs and preset feature parameters to obtain target feature points;
[0024] The feature point processing module is used to perform mean processing based on the target feature points to obtain the target feature point coordinates and complete the feature point detection of the calibration plate.
[0025] The above scheme realizes the detection of feature points through the interaction between various modules. First, the detection image is edge detected and fitted by the quadrilateral vertex acquisition module to obtain the fitted quadrilateral to obtain the quadrilateral vertices. Then, the neighbor point pair acquisition module filters and extracts the obtained quadrilateral vertices based on the neighbor point pair algorithm to obtain the required neighbor point pairs, which are then filtered and processed by the feature point pair filtering module and the feature point processing module to finally obtain the required target feature points. Based on the provided neighbor point pair algorithm, calculation and screening are performed in combination with actual conditions, so that the monitoring system is not limited to the detection of standard checkerboard corner points in the calibration plate taken by ordinary cameras, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of the system for feature point detection and improving the recognition and accuracy of system detection.
[0026] Furthermore, the quadrilateral vertex acquisition module includes: an image preprocessing submodule, which is used to obtain the original image, perform region selection and denoising filtering based on the original image and image preprocessing, and obtain a detection image; an edge detection submodule, which is used to perform contour detection based on the detection image and the edge detection algorithm, and perform contour extraction based on the polygon fitting algorithm to obtain an approximate contour; a contour approximation submodule, which is used to obtain a fitted quadrilateral based on the approximate contour and the contour approximation algorithm, and then obtain the quadrilateral vertices based on the fitted quadrilateral.
[0027] The above scheme fits the grid of the calibration plate in the detection image through the edge detection submodule and the contour approximation submodule, thereby detecting the fitted quadrilaterals corresponding to all grids, and obtaining the vertices of the quadrilaterals, so as to better detect all relevant corner points in the image and provide a complete data range for the subsequent feature point extraction.
[0028] Furthermore, the image preprocessing submodule includes: an image extraction unit, which is used to obtain the original image, perform region selection based on the original image and the target region selection operation, and obtain a calibration image; a grayscale processing unit, which is used to obtain a grayscale image based on the calibration image and grayscale processing; a binarization processing unit, which is used to obtain a binarized image based on the grayscale image and adaptive threshold binarization processing; a denoising filtering unit, which is used to perform several morphological operations and filtering processing based on the binarized image to obtain a detection image.
[0029] The above scheme is the process of image preprocessing by the image preprocessing submodule on the original image. The calibration plate area to be detected is selected and locked through the image extraction unit, and a series of image data processing operations such as denoising and filtering are performed based on the grayscale processing unit, the binarization processing unit and the denoising filtering unit, so that the final detection image is clearer and easier to identify, reducing the interference of other factors.
[0030] Furthermore, the neighbor point pair acquisition module includes: a vertex screening submodule, which is used to obtain target vertices based on quadrilateral vertices and preset boundary culling conditions; a grouping submodule, which is used to process the target vertices according to a preset grouping method to obtain multiple vertex groups; a point pair submodule, which is used to execute a point pair acquisition step based on multiple vertex groups to obtain all neighbor point pairs; the point pair acquisition step includes: selecting one of the multiple vertex groups as the current vertex group, so as to obtain the minimum distance information based on the target vertex in the current vertex group and the target vertex in the remaining vertex groups, thereby determining the corresponding vertex group based on the minimum distance information, and forming an adjacent group based on the current vertex group and the corresponding vertex group, and then obtaining neighbor point pairs based on the adjacent groups and the preset corresponding method, and repeating the above steps until all neighbor point pairs are obtained; a point pair screening submodule, which is used to screen and obtain target neighbor point pairs based on all neighbor point pairs and the preset culling operation.
[0031] The above scheme cooperates with each other in the neighbor point pair acquisition module, obtains neighbor point pairs through grouping of quadrilateral vertices, distance calculation and corresponding neighbor groups based on minimum distance, and uses preset elimination operations to filter and obtain target neighbor point pairs containing required feature points. Through vertex grouping and distance calculation, neighbor groups can still be identified and detected in the face of different actual situations, so that the required target neighbor point pairs can be identified more effectively, and detection under different imaging conditions can be achieved, with higher adaptability and recognition.
[0032] Furthermore, the point pair submodule includes: a distance calculation unit, which is used to select one of the multiple vertex groups as the current vertex group, and obtain the minimum distance information based on the target vertex in the current vertex group and the target vertex in the remaining vertex groups; an adjacent group determination unit, which is used to determine the corresponding vertex group based on the minimum distance information, and form an adjacent group based on the current vertex group and the corresponding vertex group; a point pair acquisition unit, which is used to obtain neighbor point pairs based on the adjacent groups and a preset corresponding method; and a repeated execution unit, which is used to enable the distance calculation unit, the adjacent group determination unit and the point pair acquisition unit to repeatedly execute the above steps until all neighbor point pairs are obtained.
[0033] The system provided by the present invention is simple and easy to implement. First, edge detection and fitting processing are performed on the detection image to obtain a fitted quadrilateral to obtain quadrilateral vertices. Then, the obtained quadrilateral vertices are grouped, and the adjacent groups are determined based on the distance calculation between the vertices of different groups, so as to screen and extract the required neighbor point pairs, and finally obtain the required target feature points. The method adopted is based on the neighbor point pair principle. The minimum distance is calculated to confirm the relationship between adjacent groups to determine the neighbor point pairs containing feature points. The method is not limited to the detection of standard chessboard corner points in the calibration plate taken by an ordinary camera, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of feature point detection and improving the recognition and accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 A schematic diagram of a method for detecting characteristic points of a calibration plate provided by an embodiment of the present invention;
[0035] Figure 2 A schematic flow chart of a method for detecting characteristic points of a calibration plate provided by an embodiment of the present invention;
[0036] Figure 3 A schematic diagram of an original image during feature point detection of a calibration plate provided by an embodiment of the present invention;
[0037] Figure 4 A schematic diagram of a ROI image during feature point detection of a calibration plate provided by an embodiment of the present invention;
[0038] Figure 5 A schematic diagram of a detection image when detecting feature points of a calibration plate provided by an embodiment of the present invention;
[0039] Figure 6 A schematic diagram of a fitting process for detecting feature points of a calibration plate provided by an embodiment of the present invention;
[0040] Figure 7A schematic diagram of feature point depiction during feature point detection of a calibration plate provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0042] Embodiment 1:
[0043] This embodiment provides a method for detecting characteristic points of a calibration plate. Figure 1 As shown, the following steps are included:
[0044] S1: performing edge detection and fitting processing based on the detection image to obtain a fitting quadrilateral, thereby obtaining the vertices of the quadrilateral;
[0045] S2: Obtain target neighbor point pairs based on the neighbor point pair algorithm and quadrilateral vertices;
[0046] S3: Filter neighbor point pairs based on target neighbor point pairs and preset feature parameters to obtain target feature points;
[0047] S4: Perform mean processing based on the target feature points to obtain the target feature point coordinates and complete the feature point detection of the calibration plate.
[0048] The above scheme first performs edge detection and fitting processing on the detection image to obtain a fitted quadrilateral to obtain the quadrilateral vertices, and then uses the neighbor point pair algorithm to screen and extract the obtained quadrilateral vertices to obtain the required neighbor point pairs, and finally obtains the required target feature points. Based on the provided neighbor point pair algorithm, calculation and screening are performed in combination with actual conditions. It is not limited to the detection of standard chessboard corner points in the calibration plate taken by an ordinary camera, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small imaging characteristics, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of feature point detection and improving the recognition and accuracy of detection.
[0049] Optionally, step S1 includes: obtaining an original image, performing region selection and denoising filtering based on the original image and image preprocessing to obtain a detection image; performing contour detection based on the detection image and an edge detection algorithm, and performing contour extraction based on a polygon fitting algorithm to obtain an approximate contour; obtaining a fitted quadrilateral based on the approximate contour and the contour approximation algorithm, and then obtaining the quadrilateral vertices based on the fitted quadrilateral.
[0050] In the specific implementation process, the edge detection algorithm is used to further process the detection image after denoising and filtering to detect the outline of the black grid of the calibration plate in the detection image, and then the image is contour detected and extracted through the polygon fitting algorithm to obtain the approximate outline of the black grid in the calibration plate, and the obtained outline is stored; then the stored black grid outline of the calibration plate is polylined through the contour approximation algorithm, and the approximated fitting quadrilateral is obtained to obtain the vertices of the quadrilateral, and the vertices of each quadrilateral are stored, recorded as Vertex, as a basis for extracting feature points (i.e., the intersection points of black and white grids, called corner points).
[0051] Optionally, the steps of obtaining an original image, performing region selection and denoising filtering based on the original image and image preprocessing, and obtaining a detection image include: obtaining the original image, performing region selection based on the original image and target region selection operation, and obtaining a calibration image; obtaining a grayscale image based on the calibration image and grayscale processing; obtaining a binarized image based on the grayscale image and adaptive threshold binarization processing; and performing several morphological operations and filtering processing based on the binarized image to obtain a detection image.
[0052] In the specific implementation process, the wide-angle image (i.e., the original image) taken by the surround-view fisheye camera after the dedistortion processing is subjected to the target area selection operation, such as using ROI area selection, retaining the ROI local image where the calibration plate is located in the picture, and setting the remaining areas to black to obtain the calibration image; the calibration image is subjected to image grayscale processing, and after obtaining the grayscale image, the grayscale image is subjected to adaptive threshold binarization processing to obtain the required binary image; the binary image is subjected to unlimited morphological operations according to actual conditions, such as expansion and corrosion, opening and closing operations, etc., and the morphologically processed image is subjected to filtering processing such as not limited to Sobel, etc., to finally obtain a denoising filtered detection image.
[0053] Optionally, step S2 includes: obtaining target vertices based on quadrilateral vertices and preset boundary culling conditions; processing the target vertices according to a preset grouping method to obtain multiple vertex groups; executing a point pair acquisition step based on the multiple vertex groups to obtain all neighbor point pairs; the point pair acquisition step includes: selecting one of the multiple vertex groups as the current vertex group to obtain minimum distance information based on the target vertex in the current vertex group and the target vertices in the remaining vertex groups, thereby determining the corresponding vertex group based on the minimum distance information, and forming an adjacent group based on the current vertex group and the corresponding vertex group, and then obtaining neighbor point pairs based on the adjacent groups and the preset corresponding method, repeating the above steps until all neighbor point pairs are obtained; filtering and obtaining target neighbor point pairs based on all neighbor point pairs and the preset culling operation.
[0054] In the specific implementation process, the "neighbor point pair" algorithm is used to screen the coordinates of each quadrilateral vertex stored in Vertex: (1) The quadrilateral boundary vertices of Vertex are eliminated to obtain the target vertex, and the target vertices in Vertex after elimination are divided into groups, and the vertices that do not meet the characteristics of the feature points on the ROI boundary are also eliminated, and the remaining vertices in the current group are retained and re-stored; (2) For the target vertices representing each black quadrilateral grid stored in Vertex, which contain the corresponding vertex coordinate elements, taking the first group of coordinate elements as an example, the distance between each vertex coordinate element in the group and each vertex coordinate element in the remaining groups is calculated, and the minimum value, that is, the minimum distance information, is screened out, which is recorded as distance. ce; (3) After obtaining the distance, the two groups corresponding to the distance can be regarded as adjacent black grid groups, and the corresponding vertex coordinates in the adjacent groups can be regarded as "neighbor point pairs". At this time, the two corresponding vertex coordinates can be stored in a new container and added to the result; (4) Repeat steps (2) and (3) until all neighbor point pairs are obtained; (5) For a pair of "neighbor point pairs", duplicate storage will occur in the process of calculating whether they are "neighbor point pairs", such as point 1 to point 2 and point 2 to point 1. After obtaining the result container, it is necessary to remove duplicate element pairs based on the preset removal operation, and update the result container to obtain all the final "neighbor point pairs" without duplication, that is, the target neighbor point pairs.
[0055] A method for detecting feature points of a calibration plate provided in the present embodiment first performs edge detection and fitting processing on the detection image to obtain a fitted quadrilateral to obtain quadrilateral vertices, then groups the obtained quadrilateral vertices, and determines adjacent groups based on the distance calculation between the vertices of different groups, thereby screening and extracting the required neighbor point pairs, and finally obtaining the required target feature points. The method adopted is based on the neighbor point pair principle, and determines the neighbor point pairs containing feature points by calculating and confirming the minimum distance to obtain the adjacent group relationship. It is not limited to the detection of standard checkerboard corner points in the calibration plate captured by an ordinary camera, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of feature point detection and improving the recognition and accuracy of detection.
[0056] Embodiment 2:
[0057] The present embodiment provides a calibration method for feature points of a calibration plate, which is implemented using the aforementioned calibration plate feature point detection method, and specifically includes: performing calibration plate detection based on the aforementioned calibration plate feature point detection method, acquiring target feature point coordinates, and thereby implementing feature point calibration of the calibration plate based on the target feature point coordinates.
[0058] In the specific implementation process, the key to the calibration process is to detect and extract the calibration plate feature points in the view under the camera lens angle of view. When the calibration method of this embodiment is used, the calibration plate feature points in the view under the camera lens angle of view and the corresponding feature points in the virtual top view are matched using the target feature point coordinates, and then the relevant parameters for perspective transformation can be calculated through the matched data information to restore and adjust the image, and finally a bird's-eye view image around the vehicle body is obtained.
[0059] The calibration method for the feature points of a calibration plate provided in this embodiment is implemented by adopting the aforementioned method for detecting the feature points of a calibration plate, and its feature point detection has higher adaptability, recognition and accuracy, provides more accurate calibration coordinates for calibration, and improves the calibration efficiency and success rate.
[0060] Embodiment three:
[0061] This embodiment provides a calibration plate feature point detection system, which is used to implement the calibration plate feature point detection method, including:
[0062] A quadrilateral vertex acquisition module is used to perform edge detection and fitting processing based on the detection image to obtain a fitted quadrilateral, thereby obtaining quadrilateral vertices;
[0063] A neighbor point pair acquisition module is used to acquire target neighbor point pairs based on a neighbor point pair algorithm and quadrilateral vertices;
[0064] A feature point pair screening module is used to screen neighbor point pairs based on target neighbor point pairs and preset feature parameters to obtain target feature points;
[0065] The feature point processing module is used to perform mean processing based on the target feature points to obtain the target feature point coordinates and complete the feature point detection of the calibration plate.
[0066] This embodiment realizes the detection of feature points through the interaction between various modules. First, the detection image is edge detected and fitted by the quadrilateral vertex acquisition module to obtain a fitted quadrilateral to obtain quadrilateral vertices. Then, the neighbor point pair acquisition module filters and extracts the obtained quadrilateral vertices based on the neighbor point pair algorithm to obtain the required neighbor point pairs, and the feature point pair screening module and the feature point processing module perform screening and processing to finally obtain the required target feature points. Based on the provided neighbor point pair algorithm, calculation and screening are performed in combination with actual conditions, so that the monitoring system is not limited to the detection of standard chessboard corner points in the calibration plate photographed by ordinary cameras, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with near-large and far-small, near-straight and far-curved imaging characteristics, thereby enhancing the adaptability of the system in feature point detection, improving the recognition and accuracy of system detection, and can be applied to the calibration of surround-view fisheye cameras of different vehicle models, different installation methods and different specifications and parameters.
[0067] Optionally, the quadrilateral vertex acquisition module includes: an image preprocessing submodule, used to acquire the original image, perform region selection and denoising filtering based on the original image and image preprocessing, and acquire a detection image; an edge detection submodule, used to perform contour detection based on the detection image and the edge detection algorithm, and perform contour extraction based on the polygon fitting algorithm to acquire an approximate contour; a contour approximation submodule, used to obtain a fitted quadrilateral based on the approximate contour and the contour approximation algorithm, and then acquire quadrilateral vertices based on the fitted quadrilateral.
[0068] In the specific implementation process, the grid of the calibration plate in the detection image is fitted through the edge detection submodule and the contour approximation submodule, so as to detect the fitted quadrilaterals corresponding to all grids, and obtain the vertices of the quadrilaterals, so as to better detect all relevant corner points in the image, providing a complete data range for the subsequent feature point extraction.
[0069] Optionally, the image preprocessing submodule includes: an image extraction unit, used to obtain the original image, perform region selection based on the original image and the target region selection operation, and obtain a calibrated image; a grayscale processing unit, used to obtain a grayscale image based on the calibrated image and grayscale processing; a binarization processing unit, used to obtain a binarized image based on the grayscale image and adaptive threshold binarization processing; a denoising filtering unit, used to perform several morphological operations and filtering processing based on the binarized image to obtain a detection image.
[0070] In the specific implementation process, the image preprocessing submodule performs image preprocessing on the original image. The image extraction unit selects and locks the calibration plate area to be detected, and performs a series of image data processing based on the grayscale processing unit, the binarization processing unit and the denoising filter unit to perform denoising filtering and other operations, so that the final detection image is clearer and easier to identify, reducing the interference of other factors.
[0071] Optionally, a neighbor point pair acquisition module includes: a vertex screening submodule, which is used to obtain target vertices based on quadrilateral vertices and preset boundary culling conditions; a grouping submodule, which is used to process the target vertices in a preset grouping manner to obtain multiple vertex groups; a point pair submodule, which is used to execute a point pair acquisition step based on multiple vertex groups to obtain all neighbor point pairs; the point pair acquisition step includes: selecting one of the multiple vertex groups as the current vertex group, so as to obtain the minimum distance information based on the target vertex in the current vertex group and the target vertex in the remaining vertex groups, thereby determining the corresponding vertex group based on the minimum distance information, and forming an adjacent group based on the current vertex group and the corresponding vertex group, and then obtaining neighbor point pairs based on the adjacent groups and the preset corresponding method, and repeating the above steps until all neighbor point pairs are obtained; a point pair screening submodule, which is used to screen and obtain target neighbor point pairs based on all neighbor point pairs and the preset culling operation.
[0072] In the specific implementation process, the sub-modules in the neighbor point pair acquisition module cooperate with each other, and the neighbor point pairs are obtained through the grouping of quadrilateral vertices, distance calculation, and corresponding neighbor groups based on the minimum distance. The preset elimination operation is used to filter and obtain the target neighbor point pairs containing the required feature points. Through vertex grouping and distance calculation, the neighbor group can still be identified and detected in the face of different actual situations, so that the required target neighbor point pairs can be identified more effectively, and detection under different imaging conditions can be achieved, with higher adaptability and recognition.
[0073] Optionally, the point pair submodule includes: a distance calculation unit, used to select one of multiple vertex groups as the current vertex group, and obtain the minimum distance information based on the target vertex in the current vertex group and the target vertex in the remaining vertex groups; an adjacent group determination unit, used to determine the corresponding vertex group based on the minimum distance information, and form an adjacent group based on the current vertex group and the corresponding vertex group; a point pair acquisition unit, used to obtain neighbor point pairs based on the adjacent groups and a preset corresponding method; a repeated execution unit, used to enable the distance calculation unit, the adjacent group determination unit and the point pair acquisition unit to repeatedly execute the above steps until all neighbor point pairs are obtained.
[0074] The system provided by the present embodiment is simple and easy to implement. First, edge detection and fitting processing are performed on the detection image to obtain a fitted quadrilateral to obtain quadrilateral vertices. Then, the obtained quadrilateral vertices are grouped, and the adjacent groups are determined based on the distance calculation between the vertices of different groups, so as to screen and extract the required neighbor point pairs, and finally obtain the required target feature points. The method adopted is based on the neighbor point pair principle. The minimum distance is calculated to confirm the relationship between adjacent groups to determine the neighbor point pairs containing feature points. It is not limited to the detection of standard checkerboard corner points in the calibration plate taken by an ordinary camera, so that it can realize detection under different imaging conditions, and can more effectively identify and detect feature points in complex visual systems such as fisheye cameras with imaging characteristics of near large and far small, near straight and far curved, thereby enhancing the adaptability of feature point detection and improving the recognition and accuracy of detection.
[0075] Embodiment 4:
[0076] This embodiment provides a specific application process of a calibration plate feature point detection method. Figure 2 As shown, the following steps are included:
[0077] Obtain the original image after dedistortion and select the ROI area to obtain the ROI image;
[0078] The ROI image is subjected to grayscale processing, adaptive binarization processing and denoising filtering processing to obtain a detection image;
[0079] Perform edge detection on the detection image to generate an initial contour;
[0080] Perform quadrilateral fitting on the initial contour to obtain an approximate contour;
[0081] Perform contour approximation on the approximate contour to obtain a fitting quadrilateral;
[0082] Use the neighbor point pair algorithm to screen the fitted quadrilateral to obtain all neighbor point pairs;
[0083] Filter all neighbor point pairs to obtain the required neighbor point pairs;
[0084] The target neighbor point pairs are averaged to obtain the target feature points.
[0085] In the specific implementation process, when this detection method is used for calibration plate detection, polygon fitting is performed to obtain the vertices of all the black quadrilateral grids of the calibration plate in the image, and then all vertices are processed by the neighbor point pair algorithm to obtain all neighbor point pairs, and then the point pairs where the required feature points are located are screened out from the neighbor point pairs, and the point pairs are averaged to finally obtain the required feature points for subsequent camera calibration. If the original image is as follows Figure 3 As shown, Figure 3Select the ROI area and get Figure 4 The ROI image is shown; the ROI image is processed by image grayscale, binarization, denoising filtering, etc. to obtain a smooth image, that is, a detection image, such as Figure 5 shown; based on Figure 5 Perform contour detection and fitting processing to obtain a fitting quadrilateral, and fit it in the original image, such as Figure 6 As shown; based on Figure 6 The fitting quadrilateral in the figure obtains the quadrilateral vertices, and performs vertex screening, neighbor point pair extraction, and average processing of neighbor point pair operations to obtain all target feature points and their coordinates in the calibration plate; finally, Figure 6 Describe the feature points in Figure 7 As shown, the feature detection of the calibration plate is completed. This process can also be used to verify the effectiveness of the detection method provided in this embodiment for detecting feature points. In actual applications, the selection of specific feature points is not limited to a form.
[0086] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for detecting characteristic points of a calibration plate, characterized in that: The following steps are involved: Perform edge detection and fitting processing based on the detection image to obtain a fitting quadrilateral, thereby obtaining the vertices of the quadrilateral; Obtain target neighbor point pairs based on neighbor point pair algorithm and quadrilateral vertices; Filter neighbor point pairs based on target neighbor point pairs and preset feature parameters to obtain target feature points; The target feature points are processed by averaging to obtain the target feature point coordinates and complete the feature point detection of the calibration plate.
2. The method for detecting characteristic points of a calibration plate according to claim 1, characterized in that: The edge detection and fitting processing is performed based on the detection image to obtain a fitted quadrilateral, thereby obtaining the vertices of the quadrilateral, including: Obtain the original image, perform region selection and denoising filtering based on the original image and image preprocessing, and obtain the detection image; Perform contour detection based on the detected image and edge detection algorithm, and extract contour based on the polygon fitting algorithm to obtain approximate contour; A fitting quadrilateral is obtained based on the approximate contour and the contour approximation algorithm, and then the quadrilateral vertices are obtained based on the fitting quadrilateral.
3. The method for detecting characteristic points of a calibration plate according to claim 2, characterized in that: The method of obtaining an original image, performing region selection and denoising filtering based on the original image and image preprocessing, and obtaining a detection image includes: Obtain an original image, perform region selection based on the original image and the target region selection operation, and obtain a calibrated image; Obtaining a grayscale image based on the calibration image and grayscale processing; Obtain a binary image based on a grayscale image and an adaptive threshold binarization process; Several morphological operations and filtering processes are performed on the binary image to obtain the detection image.
4. The method for detecting characteristic points of a calibration plate according to claim 1, characterized in that: The method of obtaining neighbor point pairs based on a neighbor point pair algorithm and quadrilateral vertices includes: Get the target vertex based on the quadrilateral vertex and the preset boundary culling condition; Process the target vertices according to a preset grouping method to obtain multiple vertex groups; Execute a point pair acquisition step based on multiple vertex groups to acquire all neighbor point pairs; The point pair acquisition step includes: selecting one of the multiple vertex groups as the current vertex group, acquiring minimum distance information based on a target vertex in the current vertex group and a target vertex in the remaining vertex groups, thereby determining a corresponding vertex group based on the minimum distance information, and forming an adjacent group based on the current vertex group and the corresponding vertex group, and then acquiring neighbor point pairs based on the adjacent group and a preset corresponding method, and repeating the above steps until all neighbor point pairs are acquired; Filter and obtain the target neighbor point pairs based on all neighbor point pairs and preset elimination operations.
5. A method for calibrating characteristic points of a calibration plate, characterized in that: The method is implemented by using a calibration plate feature point detection method as described in any one of claims 1 to 4, including: performing calibration plate detection based on the calibration plate feature point detection method, obtaining target feature point coordinates, and thereby realizing feature point calibration of the calibration plate based on the target feature point coordinates.
6. A calibration plate feature point detection system, characterized in that: A method for detecting characteristic points of a calibration plate according to any one of claims 1 to 4, comprising: A quadrilateral vertex acquisition module is used to perform edge detection and fitting processing based on the detection image to obtain a fitted quadrilateral, thereby obtaining quadrilateral vertices; A neighbor point pair acquisition module is used to acquire target neighbor point pairs based on a neighbor point pair algorithm and quadrilateral vertices; A feature point pair screening module is used to screen neighbor point pairs based on target neighbor point pairs and preset feature parameters to obtain target feature points; The feature point processing module is used to perform mean processing based on the target feature points to obtain the target feature point coordinates and complete the feature point detection of the calibration plate.
7. The calibration plate feature point detection system according to claim 6, characterized in that: The quadrilateral vertex acquisition module comprises: The image preprocessing submodule is used to obtain the original image, perform region selection and denoising filtering based on the original image and image preprocessing, and obtain the detection image; The edge detection submodule is used to perform contour detection based on the detection image and the edge detection algorithm, and to extract the contour based on the polygon fitting algorithm to obtain an approximate contour; The contour approximation submodule is used to obtain a fitting quadrilateral based on an approximate contour and a contour approximation algorithm, and then obtain quadrilateral vertices based on the fitting quadrilateral.
8. The calibration plate feature point detection system according to claim 7, characterized in that: The image preprocessing submodule comprises: An image extraction unit, used to obtain an original image, perform region selection based on the original image and a target region selection operation, and obtain a calibration image; A grayscale processing unit, used for obtaining a grayscale image based on a calibration image and grayscale processing; A binarization processing unit, used for obtaining a binarized image based on a grayscale image and an adaptive threshold binarization process; The denoising filter unit is used to perform several morphological operations and filtering processes based on the binary image to obtain a detection image.
9. The calibration plate feature point detection system according to claim 6, characterized in that: The neighbor point pair acquisition module includes: Vertex filtering submodule, used to obtain target vertices based on quadrilateral vertices and preset boundary culling conditions; A grouping submodule, used for processing target vertices according to a preset grouping method to obtain multiple vertex groups; A point pair submodule is used to perform a point pair acquisition step based on multiple vertex groups to acquire all neighbor point pairs; the point pair acquisition step includes: selecting one of the multiple vertex groups as the current vertex group, acquiring minimum distance information based on a target vertex in the current vertex group and a target vertex in the remaining vertex groups, thereby determining a corresponding vertex group based on the minimum distance information, and forming an adjacent group based on the current vertex group and the corresponding vertex group, and then acquiring neighbor point pairs based on the adjacent group and a preset corresponding method, and repeating the above steps until all neighbor point pairs are acquired; The point pair screening submodule is used to screen and obtain target neighbor point pairs based on all neighbor point pairs and preset elimination operations.
10. The method for detecting characteristic points of a calibration plate according to claim 9, characterized in that: The point-to-point submodule comprises: A distance calculation unit, used for selecting one of the multiple vertex groups as a current vertex group, and obtaining minimum distance information based on a target vertex in the current vertex group and target vertices in other vertex groups; An adjacent group determining unit, configured to determine a corresponding vertex group based on the minimum distance information, and to form an adjacent group based on the current vertex group and the corresponding vertex group; A point pair acquisition unit, used to acquire neighbor point pairs based on adjacent groups and a preset corresponding method; The repetitive execution unit is used to make the distance calculation unit, the adjacent group determination unit and the point pair acquisition unit repeatedly execute the above steps until all neighbor point pairs are acquired.