An image processing-based polar adaptive localization method and system

Abstract

This invention belongs to the field of image processing technology, specifically relating to an image processing-based polarimetric adaptive localization method and system. The method includes: acquiring edge pixels and their gradient directions in a polarimetric grayscale image; estimating the principal direction of illumination distortion based on the gradient direction, and dividing the edge pixels into a positive edge set and a negative edge set; based on the positive edge set, calculating weighted votes for the intersection points of the gradient directions of the edge pixels and accumulating them into a positive accumulator, then clustering the accumulated results to obtain initial positive candidate centers; finally, iteratively updating the positive center estimate by correcting the voting weights based on geometric consistency; obtaining a negative center estimate based on the negative edge set; and combining the positive and negative center estimates to determine the center location point of the polarimetric hole. This invention effectively overcomes the localization errors caused by uneven illumination and metal reflection, improving localization accuracy and robustness.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to an image processing-based polar adaptive localization method and system. Background Technology

[0002] In the assembly process of equipment such as vacuum circuit breakers, the precise positioning of the pole components is crucial to ensuring assembly quality. Machine vision technology is often used to guide robotic arms in grasping and assembling poles. Its core lies in accurately identifying the center position of key structures such as holes on the pole through image processing algorithms.

[0003] However, lighting conditions in industrial settings are often complex and variable. Especially when using a ring light source to illuminate a pole with a metallic surface, the non-uniformity of the light source and the anisotropic reflection characteristic of the metal surface cause even regular circular holes to exhibit uneven brightness and a noticeable systematic shift in the image. Typically, the illuminated edge shifts outward due to highlights, while the shaded edge shifts inward due to shadows.

[0004] Considering the typical size of the pole hole and the sub-pixel accuracy requirements for assembly, this systematic edge shift caused by illumination, if left unaddressed, will lead to positioning errors far exceeding the allowable range, severely impacting the accuracy of subsequent robotic arm operations. Furthermore, strong reflections on the illuminated side easily generate numerous specular false edges that do not match the true geometric edges, further interfering with the positioning algorithm. Traditional circle detection methods, such as Hough circle detection, rely on the consistency of gradient directions of edge pixels for voting. However, systematic shifts cause a systematic change in the gradient direction of the true edge points, resulting in peak values ​​in the accumulation space splitting into multiple scattered patches due to edge shifts, making it difficult to accurately determine the circle center. Even more seriously, specular false edges introduce a large number of erroneous votes, further blurring the already fragmented peak values, leading to unstable peak positions and failing to meet the stringent stability and repeatability requirements of industrial automation. Summary of the Invention

[0005] To address the technical problems in the prior art that lead to inaccurate pole positioning due to uneven illumination, metal reflection, and false edge interference, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides an image processing-based polar adaptive localization method, comprising:

[0007] The process involves: acquiring edge pixels and their gradient directions in a grayscale image of a polarity; estimating the principal direction of illumination distortion based on the gradient directions; dividing the edge pixels into positive and negative edge sets based on the principal direction of illumination distortion; obtaining a positive center estimate based on the positive edge sets, including: acquiring effective edge pixel pairs in the positive edge sets; calculating the intersection points of the gradient directions of the effective edge pixel pairs; determining the voting weights of the intersection points based on the quality scores of the edge pixels in the effective edge pixel pairs and the distance consistency weights of the effective edge pixel pairs; accumulating the voting weights into a positive accumulator; performing clustering processing on the accumulation results of the positive accumulator to obtain initial positive candidate centers; determining corrected voting weights based on the geometric consistency between the effective edge pixel pairs and the current positive candidate centers; updating the positive accumulator and extracting new positive candidate centers using the corrected voting weights; iteratively updating the positive candidate centers; and using the positive candidate centers at the end of the iteration as the positive center estimate; obtaining a negative center estimate based on the negative edge sets; and combining the positive and negative center estimates to determine the center location of the polarity hole.

[0008] This invention estimates the principal direction of illumination distortion based on statistical analysis of gradient direction, and then divides edge pixels into positive and negative edge sets, effectively addressing the systematic offset problem caused by anisotropic reflection of metal surfaces. In the processing of the positive edge set, effective edge pixel pairs are constructed and their gradient direction intersections are calculated. Combined with edge quality assessment and distance consistency analysis, accurate voting on potential center positions is achieved. Initial candidate centers are obtained through cluster analysis of the accumulator, and the center estimates are continuously refined through an iterative optimization mechanism to gradually eliminate noise interference and improve positioning accuracy. Similarly, negative center estimates are obtained based on the negative edge set. Finally, the center estimates of the two directions are combined to effectively offset the systematic offset, achieving high-precision positioning of the center of the pole hole. This solves the problem of low positioning accuracy of traditional methods under strong reflection and multi-layer concentric circle structures, providing reliable visual guidance for automated assembly and meeting the stringent requirements of sub-pixel-level positioning accuracy for high-end power equipment assembly.

[0009] Preferably, estimating the principal direction of illumination distortion based on the gradient direction includes: constructing a gradient direction histogram of edge pixels, and determining the gradient direction corresponding to the peak value in the gradient direction histogram as the principal direction of illumination distortion.

[0010] Preferably, the step of dividing edge pixels into a positive edge set and a negative edge set based on the principal direction of illumination distortion includes: denoting the principal direction of illumination distortion as... In response to the gradient direction of the edge pixels, located at Within the range, edge pixels are assigned to the positive edge set in response to the gradient direction of the edge pixels. Within the range, edge pixels are assigned to the negative edge set.

[0011] This invention divides edge pixels based on the main direction of illumination, which can group edge pixels affected by similar illumination into one category, making subsequent processing of each subset more targeted and helping to compensate for edge offset under different illumination conditions.

[0012] Preferably, obtaining valid edge pixel pairs in the forward edge set includes: for any edge pixel in the forward edge set, finding other edge pixels in the forward edge set whose distance from the edge pixel is less than or equal to a preset distance threshold, and forming a valid edge pixel pair with the edge pixel and the other edge pixels found.

[0013] Preferably, the quality score of the edge pixels satisfies the expression: In the formula, Represents edge pixels The quality score edge pixels gradient magnitude, The maximum gradient magnitude among all edge pixels. This is a hyperparameter used to control the rate of decay of the quality score; It is a natural exponential function.

[0014] Preferably, the distance consistency weight of the effective edge pixel pairs satisfies the expression: In the formula, Indicates valid edge pixel pairs Distance consistency weight, edge pixels and The intersection of the gradient directions; edge pixels to the intersection The distance; edge pixels to the intersection The distance; This is a hyperparameter used to control the decay rate of the distance consistency weight; It is a natural exponential function.

[0015] This invention calculates distance consistency weights based on the distance difference between two edge pixels and their intersection point. When the distances between two edge pixels and their intersection point are similar, it indicates that they are more likely to belong to the same ring, and thus a higher voting weight is assigned. When the distance difference is large, a lower weight is assigned. This effectively identifies and strengthens the contribution of edge points of the real ring, while suppressing the influence of invalid point pairs caused by factors such as uneven lighting and noise. This makes the peak values ​​in the accumulator more concentrated and accurate, providing a high-quality voting basis for subsequent center estimation and improving the accuracy and robustness of the circle center localization.

[0016] Preferably, determining the corrected voting weight based on the geometric consistency between the effective edge pixel pair and the current positive candidate center includes: calculating the reliability weight of the effective edge pixel pair with respect to the current positive candidate center; and correcting the voting weight of the intersection point based on the reliability weight to obtain the corrected voting weight.

[0017] Preferably, the reliability weight satisfies the expression: ;in, For effective edge pixel pairs For the One positive candidate center Reliability weights; For the first One positive candidate center; For effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions; edge pixels Pointing to the One positive candidate center vector and The angle between the gradient directions; edge pixels Pointing to the One positive candidate center vector and The angle between the gradient directions; The attenuation parameter is the distance between the intersection centers. This is a hyperparameter used to control the rate of angular uniformity decay; It is a natural exponential function.

[0018] This invention intelligently adjusts the original voting weights by comprehensively considering the distance relationship between the intersection point and the candidate center, as well as the consistency between the gradient direction of the edge pixels and the geometric direction. When the intersection point is close to the candidate center and the gradient direction of the edge pixels is consistent with the geometric direction pointing to the center, a higher reliability weight is assigned. When there is a significant distance deviation or angle mismatch, the weight is reduced. This effectively filters out the influence of noise point pairs and systematic offsets, making the voting process more focused on the true center position. Through iterative optimization, the peak value in the accumulator gradually converges to the true center, improving the accuracy and stability of the center estimation.

[0019] Preferably, the step of combining positive center estimation and negative center estimation to determine the center positioning point of the pole hole includes: for any positive center estimation among all positive center estimations, finding the negative center estimation closest to any positive center estimation among all negative center estimations; and taking the midpoint between any positive center estimation and the found nearest negative center estimation as a center positioning point of the pole hole.

[0020] This invention finds the nearest negative center estimate for each positive center estimate and takes the midpoint as the center positioning point. It utilizes the symmetry of the positive and negative edge sets to offset the systematic offset caused by uneven illumination, making the positioning more accurate.

[0021] Secondly, the present invention provides an image processing-based polar adaptive positioning system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned image processing-based polar adaptive positioning method is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned image processing-based polar adaptive positioning method and stored in the memory so that it can be loaded and executed by the processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: Based on gradient direction, this invention uses statistical analysis to estimate the main direction of illumination distortion, and then divides edge pixels into positive and negative edge sets, effectively addressing the systematic offset problem caused by anisotropic reflection of metal surfaces. In the processing of the positive edge set, by constructing effective edge pixel pairs and calculating their gradient direction intersections, combined with edge quality assessment and distance consistency analysis, accurate voting on potential center positions is achieved. Initial candidate centers are obtained through cluster analysis of the accumulator, and the center estimation is continuously refined through an iterative optimization mechanism, gradually eliminating noise interference and improving positioning accuracy. Similarly, negative center estimates are obtained based on the negative edge set. Finally, by combining the center estimates from both directions, systematic offset is effectively offset, achieving high-precision positioning of the center of the pole hole. This solves the problem of low positioning accuracy of traditional methods under strong reflection and multi-layer concentric circle structures, providing reliable visual guidance for automated assembly and meeting the stringent requirements of sub-pixel-level positioning accuracy for high-end power equipment assembly. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating an image processing-based polar adaptive localization method according to the present invention;

[0025] Figure 2 This is an RGB image of the pole;

[0026] Figure 3 To use the method of the present invention to Figure 2 A schematic diagram showing the results of the positioning process;

[0027] Figure 4 To use the traditional Hough circle detection method Figure 2 A schematic diagram showing the results of the positioning process. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] This invention discloses an image processing-based polar adaptive localization method, referring to... Figure 1 This includes steps S1-S4:

[0031] S1. Acquire the polarimetric grayscale image and obtain the edge pixels in the polarimetric grayscale image.

[0032] Specifically, in the electrode assembly process, a white LED ring light source is set up, and an industrial CMOS camera is used to capture RGB images of the electrode. For example, Figure 2 This is an RGB image of the polarity. To facilitate subsequent processing, the RGB image of the polarity is converted to a grayscale image.

[0033] Edge detection is performed on the polarimetric grayscale image to obtain the edge pixels and their gradient directions. This embodiment uses the Canny edge detection algorithm for edge detection of the grayscale image. In other embodiments, implementers can choose edge detection algorithms such as Sobel, Prewitt, or Laplacian edge detection algorithms according to the actual implementation situation.

[0034] S2. Estimate the principal direction of illumination distortion based on the gradient direction of edge pixels. Divide the edge pixels into positive edge sets and negative edge sets based on the principal direction of illumination distortion. Obtain valid edge pixel pairs. Determine the voting weight based on quality score and distance consistency weight. Accumulate the weights to the positive accumulator and the negative accumulator. Cluster the accumulated results to obtain the initial positive candidate center and negative candidate center.

[0035] It should be noted that during the electrode assembly process, due to the non-uniformity of the ring light source and the anisotropic reflection of the metal surface, the edge of the hole experiences a positive offset on the illuminated side and a negative offset on the shaded side. Since the diameter of the vacuum circuit breaker electrode hole is typically in the range of 10-20mm, corresponding to approximately 30-60 pixels in the image, and the assembly accuracy requirement is at the sub-pixel level, this systematic offset directly leads to a center positioning error exceeding 2 pixels, severely affecting the subsequent grasping accuracy of the robotic arm. If the illumination direction is not considered and a Hough circle transform is directly applied to all edge pixels, the accumulator peak in the Hough circle transform will split into two separate patches, making it impossible to obtain an accurate center position through conventional peak search. Therefore, this invention first accurately estimates the actual illumination direction using the gradient direction histogram of edge pixels, and then decomposes the edge set into two independent subsets based on the diameter attribute, ensuring that the edge pixels in each subset are affected by similar illumination conditions, thereby effectively reducing the impact of systematic offset on positioning accuracy.

[0036] Specifically, a gradient direction histogram is constructed based on the gradient directions of all edge pixels. The horizontal axis of the gradient direction histogram represents the gradient direction, and the vertical axis represents the number of edge pixels. The gradient direction corresponding to the peak value in the gradient direction histogram is taken as the main direction of illumination distortion. In this embodiment, the gradient direction histogram is statistically analyzed using 360 equally spaced intervals, with each interval being 1 degree. To improve the robustness of illumination direction estimation, the gradient direction histogram is Gaussian smoothed to eliminate local fluctuations caused by noise. In other embodiments, the implementer can set the number of histogram intervals according to the actual implementation situation.

[0037] Let the principal direction of the illumination distortion be denoted as In response to the gradient direction of the edge pixels, located at Within the range, edge pixels are assigned to the forward edge set. Response to the gradient direction of edge pixels Within the range, edge pixels are assigned to the negative edge set. .

[0038] Set a positive accumulator with the same size as the polar grayscale image. and negative accumulator Initialize the positive accumulator and negative accumulator All elements in the array are 0.

[0039] For positive edge sets any edge pixel in Obtain the positive edge set The distance between the pixel and the edge pixel is less than or equal to a preset distance threshold. Other edge pixels ,Will and Construct a valid edge pixel pair Obtain valid edge pixel pairs Mid-edge pixels and The intersection of the gradient directions Calculate the intersection point Voting weight Positive accumulator midpoint A vote is held, and the vote count is... Among them, the distance threshold This is used to limit the consideration to edge pixel pairs within a local area, avoiding noise intersections caused by distant pixel pairs, thus improving computational efficiency and positioning accuracy. In this embodiment, the distance threshold... The distance threshold is set to 30 pixels. In other embodiments, the implementer can set the distance threshold according to the actual implementation situation. .

[0040] Preferably, edge pixels and The intersection of the gradient directions The voting weights satisfy the expression:

[0041]

[0042] in, For effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions Voting weight; Represents edge pixels Quality score; Represents edge pixels Quality score; Indicates valid edge pixel pairs Distance consistency weights.

[0043] The quality scores of edge pixels satisfy the expression:

[0044]

[0045] In the formula, Represents edge pixels The quality score edge pixels gradient magnitude, The maximum gradient magnitude among all edge pixels. This is a hyperparameter used to control the decay rate of the quality score. This invention is based on a Gaussian distribution model, mapping the gradient magnitude to a quality score, when edge pixels... The largest gradient magnitude among all edge pixels At that time, edge pixels Quality score Close to 1; when much smaller Quality score Approaching 0. Hyperparameters The larger the value, the greater the allowable difference in gradient magnitude, and the slower the quality score decays; The smaller the value, the faster the quality score decays; edge pixels with high gradient magnitudes have higher quality scores. In this embodiment, Set as In other embodiments, implementers may set the appropriate parameters according to the actual implementation situation. .

[0046] Distance consistency weights satisfy the expression:

[0047]

[0048] In the formula, Indicates valid edge pixel pairs Distance consistency weight, edge pixels and The intersection of the gradient directions; edge pixels to the intersection The distance; edge pixels to the intersection The distance; This is a hyperparameter used to control the decay rate of the distance consistency weight. On the same circle, all edge pixels should be equidistant from the center; therefore, when... arrive distance and arrive When the distances are close, Smaller, edge pixels With edge pixels More likely, they are edge pixels on the same circle. Approaching 1; when arrive distance and arrive When the distance difference is large, Larger, edge pixels With edge pixels The less likely they are to be on the same circle, the more likely they are to be on the same circle. Approaching 0. In this embodiment, The resolution is set to 3 pixels. In other embodiments, the implementer may set the resolution according to the actual implementation situation. .

[0049] Similarly, traversing the negative edge set Obtain the negative edge set from the edge pixels in the image. The distance between the pixel and the edge pixel is less than or equal to a preset distance threshold. Other edge pixels form valid edge pixel pairs. The intersection points of the gradient directions of the edge pixels in each valid edge pixel pair and their voting weights are calculated. These voting weights are then used to construct a negative accumulator. To conduct a vote.

[0050] The final positive accumulator Perform clustering and select all clusters whose peak intensities are greater than a threshold. The clusters are designated as positive candidate peak regions, and the peak intensity is the sum of the votes from all points in the cluster. The centroid of each positive candidate peak region is used as the initial positive candidate center. ( ),in, Indicates the first One positive candidate center, This indicates the number of positive candidate peak regions. In this embodiment, Set as positive accumulator The threshold is 0.7 times the mean peak intensity of all clusters in the clustering results. This threshold can effectively distinguish between true peak values ​​and noise peak values. In other embodiments, implementers can set it according to the actual implementation situation. It should be noted that the present invention uses a density-based spatial clustering algorithm (DBSCAN) for clustering. In other embodiments, implementers may choose a clustering algorithm according to the actual implementation situation.

[0051] Similarly, for the final negative accumulator Perform clustering and select all clusters whose peak intensities are greater than a threshold. The clusters are used as negative candidate peak regions, and the centroids of each negative candidate peak region are used as initial negative candidate centers. In this embodiment, Set as negative accumulator The threshold is 0.7 times the mean peak intensity of all clusters in the clustering results. This threshold can effectively distinguish between true peak values ​​and noise peak values. In other embodiments, implementers can set it according to the actual implementation situation. .

[0052] S3. Determine the corrected voting weight based on the geometric consistency between the effective edge pixel pairs in the positive edge set and the current positive candidate center. Use the corrected voting weight to iteratively update the positive accumulator and extract new positive candidate centers until the iteration stops, and obtain the positive center estimate based on the effective edge pixel pairs in the negative edge set.

[0053] It should be noted that on the light-receiving side of the electrode hole edge, due to the specular reflection characteristics of the metal surface, a large number of specular false edges are generated. The gradient direction of these false edges is inconsistent with the true geometric direction, resulting in false peaks in Hough space. During the assembly of the vacuum circuit breaker, if the center estimation error on the light-receiving side exceeds 1 pixel, it will cause the robotic arm to collide with the workpiece or fail to grasp it. Since the gradient direction of the true edge pixel should be consistent with the geometric direction pointing to the center, while the gradient direction of the noise point is inconsistent with the geometric direction pointing to the current estimated center, this invention provides appropriate tolerance for small angle deviations and gradually optimizes the accumulator through weighted voting to obtain a more accurate center estimate.

[0054] Specifically, the centroid of each positive candidate peak region is used as the positive candidate center. ( ),in, Indicates the first One positive candidate center, This indicates the number of positive candidate peak regions. The positive accumulator... Reset, making the positive accumulator All elements in the array are 0.

[0055] Response to effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions With positive candidate centers The distance is less than the distance threshold Calculate effective edge pixel pairs For positive candidate centers Reliability weights:

[0056]

[0057] in, For effective edge pixel pairs For the One positive candidate center Reliability weights; For the first One positive candidate center; For effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions; edge pixels Pointing to the One positive candidate center vector and The angle between the gradient directions; edge pixels Pointing to the One positive candidate center vector and The angle between the gradient directions; This is the intersection center distance decay parameter, used to control the weight decay rate of the distance from the intersection point to the candidate center; This is a hyperparameter used to control the rate of angular uniformity decay; It is a natural exponential function. In this embodiment, Set to 5 pixels. Set as In other embodiments, implementers may set the appropriate parameters according to the actual implementation situation. as well as .

[0058] Furthermore, the voting weights are adjusted based on reliability weights, specifically:

[0059]

[0060] in, For effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions In the One positive candidate center The adjusted weights below; For effective edge pixel pairs For the One positive candidate center Reliability weights; For effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions The voting weight.

[0061] The positive accumulator is adjusted according to the weight. midpoint A vote is held, and the vote count is... .

[0062] A forward accumulator can be implemented by traversing all valid edge pixel pairs and forward candidate centers. The update is performed using the method in step S2 to obtain the updated positive accumulator. All positive candidate peak regions in the middle will be updated in the positive accumulator. The centroids of each positive candidate peak region are used as new positive candidate centers. In response to a new positive candidate center being greater than a preset convergence threshold, step S3 is repeated until the distance between the new and previous positive candidate centers is no greater than the preset convergence threshold or the maximum number of iterations is reached. The latest positive candidate center is then used as the positive center estimate. In this embodiment, the convergence threshold is 0 pixels and the maximum number of iterations is 5. In other embodiments, the implementer can set the convergence threshold and the maximum number of iterations according to the actual implementation situation; for example, the convergence threshold could be 1 pixel and the maximum number of iterations could be 5.

[0063] Similarly, the negative center estimate is obtained. It should be noted that the negative center estimate is based on the negative edge set. The method for obtaining negative candidate peak regions is the same as that for obtaining positive center estimation, and will not be described in detail here.

[0064] S4. Combining the positive center estimation and the negative center estimation, determine the center positioning point of the pole hole.

[0065] Specifically, all center points are located based on the positive and negative center estimates, including:

[0066] For any positive center estimate, obtain the nearest negative center estimate and use the midpoint between the positive and negative center estimates as a center location point for the pole hole. By traversing all positive and negative center estimates, all center location points are obtained.

[0067] This completes the positioning of the center of the pole hole.

[0068] For example, Figure 3 To use the method of the present invention to Figure 2 A schematic diagram showing the results of the positioning process. Figure 4 To use the traditional Hough circle detection method Figure 2 The diagram showing the positioning results illustrates that... Figure 4 The previous method only detected one center point, and the detected center point was offset. However, the present invention can detect all the center points, and the positions of the center points are accurate.

[0069] This invention also discloses an image processing-based polar adaptive positioning system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an image processing-based polar adaptive positioning method according to the present invention.

[0070] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A polar adaptive localization method based on image processing, characterized in that, include: Obtain the edge pixels and their gradient directions in a polar grayscale image; The principal direction of illumination distortion is estimated based on the gradient direction, and the edge pixels are divided into positive edge set and negative edge set based on the principal direction of illumination distortion. Obtaining a forward center estimate based on a forward edge set includes: acquiring valid edge pixel pairs from the forward edge set; calculating the intersection of the gradient directions of the valid edge pixel pairs; determining the voting weight of the intersection based on the quality score of the edge pixels in the valid edge pixel pairs and the distance consistency weight of the valid edge pixel pairs; accumulating the voting weight into a forward accumulator; performing clustering processing on the accumulation result of the forward accumulator to obtain initial forward candidate centers; determining corrected voting weights based on the geometric consistency between the valid edge pixel pairs and the current forward candidate centers; updating the forward accumulator using the corrected voting weights and extracting new forward candidate centers, realizing iterative updating of the forward candidate centers; and using the forward candidate centers at the end of the iteration as the forward center estimate. Negative center estimates are obtained based on the negative edge set; the negative center estimates are based on the negative edge set. The negative candidate peak region is obtained using the same method as the positive center estimation. The center location of the pole hole is determined by combining the positive and negative center estimations.

2. The polar adaptive localization method based on image processing according to claim 1, characterized in that, Estimating the principal direction of illumination distortion based on the gradient direction includes: Construct a gradient direction histogram for edge pixels, and determine the gradient direction corresponding to the peak value in the gradient direction histogram as the main direction of illumination distortion.

3. The polar adaptive localization method based on image processing according to claim 1, characterized in that, The method of dividing edge pixels into positive edge sets and negative edge sets based on the principal direction of illumination distortion includes: Let the principal direction of the illumination distortion be denoted as In response to the gradient direction of the edge pixels, located at Within the range, edge pixels are assigned to the positive edge set in response to the gradient direction of the edge pixels. Within the range, edge pixels are assigned to the negative edge set.

4. The polar adaptive localization method based on image processing according to claim 1, characterized in that, The process of obtaining valid edge pixel pairs in the positive edge set includes: For any edge pixel in the forward edge set, find other edge pixels in the forward edge set whose distance from the edge pixel is less than or equal to a preset distance threshold, and form a valid edge pixel pair with the other edge pixels found.

5. The polar adaptive localization method based on image processing according to claim 1, characterized in that, The quality score of the edge pixel satisfies the expression: ； In the formula, Represents edge pixels The quality score edge pixels gradient magnitude, The maximum gradient magnitude among all edge pixels. This is a hyperparameter used to control the rate of decay of the quality score; It is a natural exponential function.

6. The polar adaptive localization method based on image processing according to claim 1, characterized in that, The distance consistency weight of the effective edge pixel pairs satisfies the expression: ； In the formula, Indicates valid edge pixel pairs Distance consistency weight, edge pixels and The intersection of the gradient directions; edge pixels to the intersection The distance; edge pixels to the intersection The distance; This is a hyperparameter used to control the decay rate of the distance consistency weight; It is a natural exponential function.

7. The polar adaptive localization method based on image processing according to claim 1, characterized in that, The step of determining the corrected voting weight based on the geometric consistency between the effective edge pixel pairs and the current positive candidate center includes: Calculate the reliability weight of the effective edge pixel pair for the current positive candidate center; adjust the voting weight of the intersection point according to the reliability weight to obtain the adjusted voting weight.

8. The polar adaptive localization method based on image processing according to claim 7, characterized in that, The reliability weights satisfy the expression: ； in, For effective edge pixel pairs For the One positive candidate center Reliability weights; For the first One positive candidate center; For effective edge pixel pairs Mid-edge pixels and The intersection of the gradient directions; edge pixels Pointing to the One positive candidate center vector and The angle between the gradient directions; edge pixels Pointing to the One positive candidate center vector and The angle between the gradient directions; The attenuation parameter is the distance between the intersection centers. This is a hyperparameter used to control the rate of angular uniformity decay; It is a natural exponential function.

9. The polar adaptive localization method based on image processing according to claim 1, characterized in that, The method of determining the center positioning point of the pole hole by combining positive center estimation and negative center estimation includes: For any positive center estimate among all positive center estimates, find the nearest negative center estimate from all negative center estimates; use the midpoint between the positive center estimate and the nearest negative center estimate as the center point of the pole hole.

10. An image processing-based polar adaptive positioning system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement an image processing-based polar adaptive localization method according to any one of claims 1-9.

Images