Visual positioning algorithm applied to industrial production line

Abstract

The invention discloses a visual positioning algorithm applied to an industrial production line, and the algorithm is characterized in that the algorithm comprises the steps: setting an industrial camera and a software system; the method comprises the following steps: acquiring an image in real time by using an industrial camera, transmitting the acquired image to an image processing system, and extracting angular point features in the image by using a detection algorithm in the image processing system, namely angular point features of a target image; matching the extracted angular point features of the target image with angular point features in a known matching template, and screening and rejecting obtained matching results to obtain a final matching mode; according to the method, an actual production line is tested, parameters of a visual positioning algorithm are adjusted according to a test result, angular point features in an image are extracted by using an angular point detection algorithm, and matching is performed in combination with a geometric distribution rule between template feature points, so that the positioning precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of machine vision technology, and in particular to a visual positioning algorithm applied to industrial production lines. Background Art

[0002] Visual positioning refers to a technology that uses cameras or other visual sensors to acquire environmental information and determine the position and posture of objects in space. On industrial production lines, visual positioning technology enables precise positioning and identification of objects. For example, on assembly lines, robots use visual positioning technology to accurately grasp and place parts. This not only improves production efficiency but also reduces the error rate of manual operations.

[0003] Product positioning is the prerequisite and foundation for target recognition, defect detection, dimensional measurement, and grasping and placement. Accurately identifying a product's position and orientation is the core technology of a visual positioning inspection system. Visual positioning technology involves extensive image processing and feature matching operations, placing high demands on computing resources. Improving computing efficiency while maintaining positioning accuracy is a pressing challenge in practical applications.

[0004] Visual positioning is based on the image information captured by the camera, and determines the position and posture of the object by processing and analyzing the image. In visual positioning, the core technologies include feature extraction and feature matching. Feature extraction refers to the extraction of feature points or feature descriptors related to the position and posture of the object from the image. Commonly used feature extraction algorithms include Harris corner detection, SIFT, SURF, etc. These algorithms can help identify key feature points in the image. Feature matching refers to matching the feature points extracted from the image with a known template or map to determine the position of the object in space. Matching algorithms usually include KNN matching, RANSAC, etc., which achieve matching by calculating the similarity between feature points. Summary of the Invention

[0005] This application provides a visual positioning algorithm for industrial production lines. It uses a corner detection algorithm to extract corner features in an image, and combines them with the geometric distribution rules between template feature points for matching, effectively improving the positioning accuracy. By calculating the positional relationship between corner points, feature information with rotational invariance is obtained, allowing the algorithm to maintain high-precision positioning capabilities when responding to product rotation.

[0006] This application provides a visual positioning algorithm for industrial production lines, including:

[0007] S101, setting up industrial cameras and software systems;

[0008] S102, using an industrial camera to collect images in real time, transmitting the collected images to an image processing system, and using a detection algorithm in the image processing system to extract corner features in the images, i.e., corner features of the target image;

[0009] S103, matching the extracted corner features of the target image with the corner features in the known matching template, filtering and eliminating the obtained matching results to obtain a final matching method;

[0010] S104: Test the actual production line and adjust the parameters of the visual positioning algorithm based on the test results.

[0011] Preferably, the geometric distribution law between the corner feature points of the matching template is used as the matching basis, and a matching strategy of grouping the corner points in pairs is adopted. If the template contains n corner points, then There are three combinations. The information contained in each corner point is expressed as (x, y, θ) representing the coordinates and angles of the corner point respectively. Then each corner point combination contains two corner point positions (x1, y1, θ1) and (x2, y2, θ2). The positional relationship between the two corner points is obtained by calculating: Δθ1=θ1-θ, Δθ2=θ2-θ, where d represents the distance between the two corner points, Δθ1 and Δθ2 represent the angular difference between the two corner points relative to a reference direction, and d, Δθ1, and Δθ2 are information with rotation invariance.

[0012] Preferably, the matching template contains n corner points, namely pS[0], pS[1]...pS[n-1], and the current target image to be positioned detects m corner points, namely pD[0], pD[1]...pD[m-1]. Since the positioning target product is a rigid body, the relative position relationship between any two corner points pS[i] and pS[j] is fixed and has translation and rotation invariance, so that they meet the same relative position relationship within the preset tolerance range. The formula is used. Calculate the distance between pS[i] and pS[j] in the matching template, where d ji represents the distance between pS[i] and pS[j], (x i ,y i ) represents the coordinates of the corner point pS[i], (x j ,y j ) represents the coordinates of pS[j], find two corner points pD[I] and pD[J] from pD, and similarly, use the formula Calculate the distance between pD[I] and pD[J] in the target image, where d JI represents the distance between pD[I] and pD[J], (x I ,yI ) represents the coordinates of the corner point pD[I], (x J ,y J ) represents the coordinates of pD[J].

[0013] Preferably, m corner points are detected in the target image, and then grouped to form Combination of target images Each combination and template Perform feature matching, a total of matching times are required Second-rate.

[0014] Preferably, the difference between the corner point distances in the target image and the template image is calculated as Δd=|d JI -d ji |, compare the calculated distance difference with the preset tolerance range. When the distance difference between the corner point pair (pS[i], pS[j]) and the corner point pair (pD[I], pD[J]) is less than the preset tolerance range, it is considered that the corner point pair (pS[i], pS[j]) and the corner point pair (pD[I], pD[J]) meet the matching relationship, otherwise it is not met.

[0015] Preferably, when the matching relationship is satisfied, the template corner point pS[i] corresponds to the corner point pD[I] of the target image, and the template corner point pS[j] corresponds to the corner point pD[J] of the target image. The algorithm uses voting to add 1 to each of the two matching methods, and so on. Voting statistics are performed on all matching methods. After filtering and eliminating the statistical results, the one with the most votes is the best matching method, and the number of votes for each possible matching method is initialized to 0.

[0016] Preferably, the screening and elimination method is as follows:

[0017] 1) The votes received by the candidate with the fewest votes will be deemed invalid, cleared, and eliminated;

[0018] 2) For those who are eliminated, their votes will be invalidated, and the number of votes received by those who accepted their votes will be reduced by 1;

[0019] 3) Repeat steps 1 and 2 until the person with the least votes has the same number of votes as the person with the most votes;

[0020] 4) At this point, each contestant has the same number of votes, and their vote count equals the number of remaining contestants minus 1, ending the cycle.

[0021] Through the above operations, any two of the remaining players vote for each other, which is equivalent to any two matching methods (pS[i], pD[I]) and (pS[j], pD[J]) in the remaining matching methods corroborating each other, where i, j∈[0,n), I, J∈[0,m), and the corner point group pD[I] in the target image, where I∈[0,m), is the final matching result.

[0022] One or more technical solutions provided in this application have at least the following technical effects or advantages: the Harris corner detection algorithm is used to extract corner features in the image, and the geometric distribution law between the template feature points is combined for matching, which effectively improves the positioning accuracy. By calculating the positional relationship between the corner points, feature information with rotational invariance is obtained, so that the algorithm can still maintain high-precision positioning capabilities when responding to product rotation. The algorithm design takes computational efficiency into consideration. By optimizing parameters such as the corner detection threshold and matching tolerance, unnecessary calculation amount is reduced and processing speed is improved. Testing is carried out on different hardware platforms to ensure the performance stability of the algorithm in various environments and meet the real-time requirements of industrial production lines. The algorithm parameters (such as corner detection threshold, matching tolerance, etc.) can be adjusted according to actual test results to optimize positioning performance, making the algorithm more flexible and practical. Through performance evaluation and testing, problems and deficiencies in the algorithm can be discovered in a timely manner, providing direction for subsequent improvement and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a characteristic schematic diagram of a visual positioning algorithm applied to an industrial production line according to the present invention;

[0024] Figure 2 This is a diagram showing the experimental results of corner feature extraction according to an embodiment of the present invention;

[0025] Figure 3 This is a rendering of the corner feature matching experiment results of an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To facilitate understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0027] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are for illustrative purposes only and do not represent the only implementation method.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains; the terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0029] Example 1: Figure 1 This is a flow chart of a visual positioning algorithm applied to an industrial production line according to an embodiment of the present invention, including:

[0030] S101, setting up industrial cameras and software systems;

[0031] Specifically, choose the Baslerace 2500 series industrial camera with high resolution, high frame rate and good color reproduction capability. According to the type and material of the product, choose LED as the lighting source. LED can provide uniform and adjustable lighting and reduce reflections. Design the light source layout according to the shape, size and shooting angle of the product. The designed light source layout can ensure that the light can evenly cover the entire product surface. At the same time, avoid direct light source exposure to the camera lens to avoid glare or shadows that affect image quality. Use different types of light source layouts such as ring light source, strip light source or coaxial light source, and choose according to the specific needs of the product. By adjusting the position, angle and brightness of the light source, you can optimize the lighting effect and reduce shadows and reflections. Impact on image quality; Select camera brackets and fixtures based on the layout of the production line and the camera's shooting angle, ensure that the camera brackets and fixtures are installed firmly and reliably to withstand vibration and shaking on the production line, use fasteners and connectors to ensure that the connection between the camera and the bracket is tight and stable, adjust the camera's angle and focal length based on the product's positioning requirements and shooting range, use the camera's built-in adjustment mechanism or external adjustment tools to ensure that the camera can accurately align with the product and capture clear images. During the adjustment process, pay attention to the camera's field of view and depth of field to ensure that the captured image meets the requirements of subsequent processing.

[0032] According to the requirements of the production line and visual positioning algorithm, a software system containing the visual positioning algorithm is integrated. The software system includes the functions of image reception, processing, analysis and positioning result output. According to the actual situation of the production line and the performance requirements of the algorithm, the algorithm parameters are adjusted and the performance is optimized. The stability and accuracy of the algorithm in the software system are verified to ensure the reliability of the positioning results.

[0033] S102, using an industrial camera to collect images in real time, transmitting the collected images to an image processing system, and using a detection algorithm in the image processing system to extract corner features in the images, i.e., corner features of the target image;

[0034] Furthermore, industrial cameras are used to ensure that they can work properly and capture images in real time. The resolution and frame rate of the industrial cameras are set to meet the needs of subsequent image processing and positioning. The industrial cameras capture product images on the production line in real time and transmit these images to the image processing system for subsequent processing. The captured images should be clear, without blur, and contain sufficient product details to extract effective feature information.

[0035] In the image processing system, the Harris corner detection algorithm is used to extract corner features from the captured image. The Harris corner detection algorithm is a mature and stable algorithm that can quickly and effectively extract corner features in the image. However, the Harris corner detection algorithm obtains a small amount of information. Each corner point only contains position and angle information. In the scene where the product may rotate, it is impossible to determine the one-to-one correspondence between the template corner points and the image corner points. Based on the matching problem of the Harris corner detection algorithm in the rotation scene, this embodiment uses the geometric distribution law between the template feature points as the matching basis, searches for corner points with similar distribution laws in the image, and adopts a matching strategy of grouping the corner points in pairs. Assuming that the template contains n corner points, it can be composed Assuming that the information contained in each corner point is represented by (x, y, θ) to represent the coordinates and angle of the corner point respectively, each corner point combination contains two corner point positions (x1, y1, θ1) and (x2, y2, θ2). By calculating the positional relationship between the two corner points, the following feature information with rotation invariance can be obtained: Δθ1=θ1-θ, Δθ2=θ2-θ, where d represents the distance between the two corner points, Δθ1 and Δθ2 represent the angular difference between the two corner points relative to a reference direction, and d, Δθ1, and Δθ2 are information with rotation invariance.

[0036] A specific example is as follows: images of parts on the production line are captured in real time by an industrial camera and transmitted to an image processing system for subsequent processing. The industrial camera is set to a resolution of 1920x1080 and a frame rate of 30fps to ensure that the captured images are clear, blur-free, and contain sufficient part details. In a part image, the Harris corner detection algorithm extracts five corner points, labeled A, B, C, D, and E. The corner point combination (A, B) is selected for detailed description. The information of corner point A is (x1, y1, θ1), and the specific value is (100, 150, 30°). The information of corner point B is (x2, y2, θ2), and the specific value is (200, 250, 60°). The positional relationship between the two corner points is calculated to obtain feature information with rotation invariance:

[0037] Δθ1=θ1-θ A , where θ A Let θ be the angle between the line from A to B and the x-axis. A =45°, then Δθ1=60°-45°=15°; Δθ2=θ2-θ B , similarly, let θ B Let θ be the angle of point B relative to a global reference direction. B =75°, then Δθ2=30°-75°=-45°.

[0038] S103, matching the extracted corner point features of the target image with the corner point features in the known matching template, filtering and eliminating the obtained matching results to obtain a final matching method.

[0039] Specifically, such as Figure 3 As shown, m corner points are detected in the target image, and then grouped to form Combination of target images Each combination and template Perform feature matching, a total of matching times are required Suppose the matching template contains n corner points, namely pS[0], pS[1]...pS[n-1], and the current target image to be positioned has m corner points detected, namely pD[0], pD[1]...pD[m-1]. Since the positioning target product is a rigid body, the relative position relationship between any two corner points pS[i] and pS[j] is fixed and has translation and rotation invariance, so that they meet the same relative position relationship within the preset tolerance range, such as Figure 2 , using the formula Calculate the distance between pS[i] and pS[j] in the matching template, where d ji represents the distance between pS[i] and pS[j], (x i ,y i ) represents the coordinates of the corner point pS[i], (x j ,y j ) represents the coordinates of pS[j], find two corner points pD[I] and pD[J] from pD, and similarly, use the formula Calculate the distance between pD[I] and pD[J] in the target image, where d JI represents the distance between pD[I] and pD[J], (x I ,y I ) represents the coordinates of the corner point pD[I], (x J ,y J ) represents the coordinates of pD[J], and the difference between the corner point distances in the target image and the template image is calculated as Δd = |d JI -dji |, compare the calculated distance difference with the preset tolerance range. When the distance difference between the corner point pair (pS[i], pS[j]) and the corner point pair (pD[I], pD[J]) is less than the preset tolerance range, it is considered that the corner point pair (pS[i], pS[j]) and the corner point pair (pD[I], pD[J]) meet the matching relationship, otherwise they do not meet the matching relationship. When the matching relationship is satisfied, the template corner point pS[i] corresponds to the corner point pD[I] of the target image, and the template corner point pS[j] corresponds to the corner point pD[I] of the target image. The corner point pD[J] of the target image, that is, the matching method (pS[i], pD[I]) and the matching method (pS[j], pD[J]) support each other. The algorithm uses a voting method to add 1 to each of the two matching methods. Similarly, all matching methods are voted and counted. After filtering and eliminating the statistical results, the one with the most votes is the best matching method. The number of votes for each possible matching method (that is, the correspondence between the template corner point and the target image corner point) is initialized to 0. The specific filtering and elimination methods are as follows:

[0040] 1. The votes received by the candidate with the fewest votes will be deemed invalid, cleared, and eliminated;

[0041] After the voting is over, the number of votes for each matching method is counted, and the matching method with the least votes is found. The matching method with the least votes means that it has the lowest support among all possible matching methods, so it is considered unreliable, its votes are reset to zero, and it is removed from the candidate matching methods.

[0042] 2. For those who are eliminated, their votes will be invalidated, and the number of votes received by those who accepted their votes will be reduced by 1;

[0043] Since the eliminated matching method has a mutually corroborating relationship with other matching methods (i.e., it voted for other matching methods), its votes must also be considered invalid. At the same time, the number of votes for those matching methods that accepted the votes of the eliminated matching method must be reduced by 1 to reflect this change.

[0044] 3. Repeat steps 1 and 2 until the candidate with the fewest votes has the same number of votes as the candidate with the most votes.

[0045] Repeat the above screening and elimination process until the number of votes for the match with the least votes is the same as the number of votes for the match with the most votes, which means that the remaining matches have reached a certain balance in support and no further obviously unreliable matches can be eliminated.

[0046] 4. At this point, each contestant has the same number of votes, and their vote count equals the number of remaining contestants minus 1, ending the loop;

[0047] When the loop terminates, the number of votes for the remaining matches should be the same, and this number of votes is equal to the number of remaining matches minus 1. This is because, ideally, each remaining match should be corroborated by all other remaining matches (that is, each match votes for all other matches), so their votes should be equal.

[0048] Through the above screening and elimination process, we obtain the final set of matching methods. In this set, any two matching methods are mutually corroborated, that is, the feature differences between them are within the tolerance range, and their support (votes) are consistent. Therefore, these matching methods are reliable, and the corner point group pD[I] in the target image (where I∈[0,m)) is the final result of matching with the template corner point. Through the above operation, any two of the remaining players vote for each other, which is equivalent to any two matching methods (pS[i], pD[I]) and (pS[j], pD[J]) in the remaining matching methods corroborating each other, where i, j∈[0,n), I, J∈[0,m). Through the above method, it can be seen that the corner point group pD[I] in the target image, where I∈[0,m), is the final matching result.

[0049] S104, testing the actual production line and adjusting the parameters of the visual positioning algorithm based on the test results;

[0050] Furthermore, the positioning accuracy is evaluated by comparing the algorithm's positioning results with the actual position of the product. Statistical indicators such as mean square error (MSE) and absolute error (MAE) are used to quantify positioning accuracy. Tests are conducted under different lighting conditions, product postures, and background environments. The time required for the algorithm to process a single image is measured to evaluate its computational efficiency. Tests are also conducted on different hardware platforms (such as CPU and GPU) to compare the performance of the algorithm in different environments. By adjusting the threshold of corner detection, the number and quality of detected corners can be controlled. If too many or too few corners are detected, positioning accuracy will decrease. Therefore, it is necessary to adjust the corner detection threshold based on actual test results to achieve the best results. During the feature matching process, the size of the tolerance directly affects the matching accuracy. If the tolerance is set too large, it may result in mismatching; if the tolerance is set too small, it may result in matching failure. Therefore, it is necessary to adjust the matching tolerance based on actual test results to achieve the best matching effect.

[0051] In practical applications, abnormal conditions such as poor image quality and product deformation may cause the algorithm to fail in positioning or generate large errors. When the image quality is poor (such as blur or high noise), pre-processing techniques (such as filtering and enhancement) can be used to improve the image quality. If the positioning requirements are still not met after pre-processing, feature extraction and matching algorithms can be used. Product deformation may cause the position of feature points to change, thereby affecting positioning accuracy. The deformation of the product can be estimated by introducing a deformation model, and the position of the feature points can be corrected. An abnormality monitoring mechanism can be designed to monitor the operation status of the algorithm in real time, and abnormalities can be discovered and handled in a timely manner. When an abnormality is detected, an alarm mechanism can be triggered to notify the operator to intervene or take appropriate remedial measures.

[0052] The technical solutions in the above-mentioned embodiments of the present application have at least the following technical effects or advantages: the Harris corner detection algorithm is used to extract corner features in the image, and the geometric distribution law between the template feature points is combined for matching, which effectively improves the positioning accuracy. By calculating the positional relationship between the corner points, feature information with rotational invariance is obtained, so that the algorithm can still maintain high-precision positioning capabilities when responding to product rotation. The algorithm design takes computational efficiency into consideration. By optimizing parameters such as the corner detection threshold and matching tolerance, unnecessary calculation amount is reduced and the processing speed is improved. Testing is carried out on different hardware platforms to ensure the performance stability of the algorithm in various environments and meet the real-time requirements of industrial production lines. The algorithm parameters (such as corner detection threshold, matching tolerance, etc.) can be adjusted according to actual test results to optimize the positioning performance, making the algorithm more flexible and practical. Through performance evaluation and testing, problems and deficiencies in the algorithm can be discovered in a timely manner, providing direction for subsequent improvement and optimization.

[0053] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A visual positioning algorithm applied to industrial production lines, characterized by: include: S101, setting up industrial cameras and software systems; S102, using an industrial camera to collect images in real time, transmitting the collected images to an image processing system, and using a detection algorithm in the image processing system to extract corner features in the images, i.e., corner features of the target image; S103, matching the extracted corner features of the target image with the corner features in the known matching template, filtering and eliminating the obtained matching results to obtain a final matching method; S104: Test the actual production line and adjust the parameters of the visual positioning algorithm based on the test results.

2. The visual positioning algorithm for industrial production lines according to claim 1, characterized in that: The geometric distribution law between the corner feature points of the matching template is used as the matching basis, and the matching strategy of grouping the corner points in pairs is adopted. If the template contains n corner points, it can form There are three combinations. The information contained in each corner point is expressed as (x, y, θ) representing the coordinates and angles of the corner point respectively. Then each corner point combination contains two corner point positions (x1, y1, θ1) and (x2, y2, θ2). The positional relationship between the two corner points is obtained by calculating: Δθ1=θ1-θ, Δθ2=θ2-θ, where d represents the distance between the two corner points, Δθ1 and Δθ2 represent the angular difference between the two corner points relative to a reference direction, and d, Δθ1, and Δθ2 are information with rotation invariance.

3. The visual positioning algorithm for industrial production lines according to claim 1, characterized in that: The matching template contains n corner points, namely pS[0], pS[1]...pS[n-1]. The current target image to be positioned has m corner points detected, namely pD[0], pD[1]...pD[m-1]. Since the positioning target product is a rigid body, the relative position relationship between any two corner points pS[i] and pS[j] is fixed and has translation and rotation invariance, so that they meet the same relative position relationship within the preset tolerance range. The formula is used. Calculate the distance between pS[i] and pS[j] in the matching template, where d ji represents the distance between pS[i] and pS[j], (x i ,y i ) represents the coordinates of the corner point pS[i], (x j ,y j ) represents the coordinates of pS[j], find two corner points pD[I] and pD[J] from pD, and similarly, use the formula Calculate the distance between pD[I] and pD[J] in the target image, where d JI represents the distance between pD[I] and pD[J], (x I ,y I ) represents the coordinates of the corner point pD[I], (x J ,y J ) represents the coordinates of pD[J].

4. The visual positioning algorithm for industrial production lines according to claim 1, characterized in that: If m corner points are detected in the target image, they are grouped to form Combination of target images Each combination and template Perform feature matching, a total of matching times are required Second-rate.

5. The visual positioning algorithm for industrial production lines according to claim 3, characterized in that: Calculate the difference between the corner point distances in the target image and the template image Δd = |d JI -d ji |, compare the calculated distance difference with the preset tolerance range. When the distance difference between the corner point pair (pS[i], pS[j]) and the corner point pair (pD[I], pD[J]) is less than the preset tolerance range, it is considered that the corner point pair (pS[i], pS[j]) and the corner point pair (pD[I], pD[J]) meet the matching relationship, otherwise it is not met.

6. The visual positioning algorithm for industrial production lines according to claim 5, characterized in that: When the matching relationship is satisfied, the template corner point pS[i] corresponds to the corner point pD[I] of the target image, and the template corner point pS[j] corresponds to the corner point pD[J] of the target image. The algorithm uses voting to add 1 to each of the two matching methods, and so on. All matching methods are voted and counted. After filtering and eliminating the statistical results, the one with the most votes is the best matching method. The number of votes for each possible matching method is initialized to 0.

7. The visual positioning algorithm for industrial production lines according to claim 6, characterized in that: The specific screening and elimination methods are as follows: 1) The votes received by the candidate with the fewest votes will be deemed invalid, cleared, and eliminated; 2) For those who are eliminated, their votes will be invalidated, and the number of votes received by those who accepted their votes will be reduced by 1; 3) Repeat steps 1 and 2 until the person with the least votes has the same number of votes as the person with the most votes; 4) At this point, each contestant has the same number of votes, and their vote count equals the number of remaining contestants minus 1, ending the cycle. Through the above operations, any two of the remaining players vote for each other, which is equivalent to any two matching methods (pS[i], pD[I]) and (pS[j], pD[J]) in the remaining matching methods corroborating each other, where i, j∈[0,n), I, J∈[0,m), and the corner point group pD[I] in the target image, where I∈[0,m), is the final matching result.

Images