A method of reading a pointer gauge

Abstract

This application discloses a method for reading pointer-type instruments. The method unifies the center, start and end angles, and the expansion ratio of the dynamic sampling band into the geometric parameters to be read. The dynamic sampling band is determined based on the distribution of scale pixel radii within the angle window. A scale sequence model is fitted in the expanded space to construct a closed-loop objective function including scale sequence residuals, boundary deviations, missing scale ratios, and smoothing constraints. After iteratively calibrating the geometric parameters, the reading is calculated and the reliability is output. This method establishes the dynamic sampling band through the angular radius distribution of the scale mask and uses the scale sequence residuals and related constraints in the expanded space to back-calibrate the original geometric parameters, making the scale distribution pattern a feedback constraint on the reading geometric parameters.

Description

Technical Field

[0001] This invention relates to the field of pointer instrument reading technology, and more specifically to a pointer instrument reading method. Background Technology

[0002] In existing reading chains, geometric unfolding parameters are typically determined once by the initial detection or calibration results. Subsequent scale sequences are mostly used for reading interpolation, and rarely participate in the calibration of center, start and end angles, and sampling band parameters. Therefore, when there is a deviation in the initial positioning or local scale is obstructed, the reading results are easily affected by amplification errors.

[0003] Existing technologies still have the following shortcomings. First, fixed ring unfolding typically uses preset inner and outer radii, which cannot adapt to changes in the effective scale radius caused by long or short scales, printing offsets, local stains, glass reflections, and occlusions at different angles and directions. This can easily lead to problems such as scale truncation or excessive background inclusion. Second, once the initial center, starting and ending scales are determined by detection, fitting, or manual calibration, they are usually directly used subsequently, lacking a closed-loop mechanism to reverse-correct the center, starting and ending angles, and unfolded sampling band parameters using the regularity of the unfolded scale sequence. Third, missing scale repair often stops at image line filling or simple point filling according to adjacent spacing, lacking constraints that bind the number of missing scales, scale sequence residuals, and reading reliability. This can easily lead to treating uncertain results in occluded areas as reliable identification results. Fourth, tilt parallax compensation often exists as an independent geometric correction step, without forming a unified reading link with dynamic sampling bands, scale sequence feedback, and occluded scale coordinate reconstruction.

[0004] Therefore, a reading method is needed that can combine the dynamic sampling band of the angle with the closed-loop feedback of the scale sequence residual, so that the scale distribution law is not only used for reading interpolation, but also participates in the calibration of the geometric unfolding parameters, thereby reducing the influence of initial geometric error, local occlusion, reflection and tilt parallax on the reading results. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems mentioned above and to propose a method for reading pointer-type instruments, comprising the following steps: Acquire an image containing pointer-type instruments and obtain the calibration dial image, scale mask, pointer mask, and initial geometric parameters; The geometric parameters to be read are uniformly represented as: ; in, The current center coordinates of the circle. and The starting and ending angles are those unfolded along the direction of the effective reading. and These are the safety extension ratio parameters for the inner and outer sides of the dynamic sampling band, respectively. Based on current geometric parameters Convert the tick marks in the tick mask to polar coordinates and divide the effective reading angle range into... A window at various angles; Within each angular window, the radial distance distribution of the scale pixels within that window is statistically analyzed. Low and high quantiles are extracted based on preset lower and upper probability thresholds, respectively. The dynamic inner and outer radii corresponding to that angular window are then determined based on these low and high quantiles. This is then combined with... and A dynamic sampling band that varies with angle is obtained; Based on dynamic sampling bands, the scale mask and pointer mask are mapped to the linear expansion space; Extract the set of scale centers in the linear expansion space, fit a scale sequence model to represent the correspondence between scale numbers and scale positions in the linear expansion space, and calculate the scale sequence residual; construct a closed-loop objective function that includes the scale sequence residual, the first and last scale boundary deviation penalty term, the missing scale ratio penalty term, and the dynamic radius smoothing constraint term. Within a defined neighborhood, based on the closed-loop objective function... The center, start and end angles, and dynamic sampling band expansion ratio parameters in the circle are iteratively calibrated until the change in the closed-loop objective function is less than the set threshold, or the number of iterations reaches the set upper limit, and the calibrated geometric parameters are obtained. Based on the calibrated geometric parameters and scale sequence model, the continuous scale number corresponding to the pointer position is determined, and the pointer instrument reading is calculated and output according to the instrument range.

[0006] In the preferred scheme, the dynamic sampling band is calculated as follows: In the Within each angle window, the dynamic inner radius is calculated using the following formula. and dynamic outer radius : ; ; in, For the first The center angle of the window. This is the set of scale pixel radii within the window. To follow the lower probability threshold Quantile values ​​extracted from the radius distribution To follow the upper probability threshold Quantile values ​​extracted from the radius distribution You can use 0.05 or 0.08. 0.95 or 0.97 can be used; , This is the preset minimum bandwidth threshold.

[0007] In a preferred embodiment, when the number of scale pixels within a certain angle window is lower than the pixel number threshold, interpolation is performed using the dynamic inner radius and dynamic outer radius of adjacent angle windows, or quantile values ​​are extracted from the global scale pixel radius distribution according to preset lower and upper probability thresholds as backoff values.

[0008] In the preferred embodiment, the dynamic inner radius and dynamic outer radius are smoothed according to the following formula to obtain the smoothed dynamic inner radius function. and dynamic outer radius function To form the final dynamic sampling band of angles: in, Let be the normalized smoothing weights in the finite smoothing kernel, and .

[0009] In the preferred embodiment, for a uniformly calibrated instrument, the calibration sequence model is represented as:

[0010] in, For the first The theoretical abscissa of each scale in the linear expansion space The distance between adjacent scale marks. For sequence offset; and with and Given the parameters to be fitted, robust fitting is performed using the following formula to calculate the scale sequence residuals. : ; ; in, For the first The detected x-coordinate of the scale center, For the corresponding tick mark, This can be Huber loss, RANSAC in-point loss, or median absolute deviation loss.

[0011] In the preferred embodiment, for non-uniform scale instruments, the scale sequence model is a piecewise linear model based on the main scale position, and the reading is obtained by piecewise interpolation based on the adjacent main scale positions and corresponding range values ​​in the interval where the pointer position is located.

[0012] In the preferred scheme, the closed-loop objective function The specific construction formula is as follows: ; in, For the current geometric parameters The scale sequence residuals obtained from the calculation are as follows: This is a penalty term for the deviation between the theoretical first and last scales and the left and right boundaries of the linear expansion space. Penalty for missing scale ratios due to mismatch. For dynamic radius function and Second-order difference smoothing constraint term on adjacent angle windows , , These are the weighting coefficients.

[0013] In the preferred embodiment, iterative calibration includes: [the calibration process is described in the original text, but the translation is incomplete and requires further context.] Within the defined neighborhood, iteratively search according to the following formula to make the closed-loop objective function Minimum candidate parameter: The neighborhood range is defined by pixel-level offset of the center coordinates, angular offset of the start and end angles, and... , The expansion ratio changes; when the change in the closed-loop objective function between two consecutive iterations is less than a set threshold, or when the number of iterations reaches a set upper limit, the iteration stops and the calibrated geometric parameters are output. .

[0014] In the preferred scheme, when missing scale values ​​exist, the parameters of the calibrated scale sequence model are used. and and missing scale numbers The theoretical coordinates of the missing scale in the linear expansion space and the theoretical coordinates in the disk coordinate system are reconstructed using the following formula. : ; ; ; in, To expand the width of the space, The radius is obtained by interpolation of the center radius of adjacent visible scales or the center radius of the dynamic sampling band; the theoretical coordinates of missing scales are used for reading mapping and credibility evaluation, and are not used to forge the original image texture.

[0015] In the preferred embodiment, the step of calculating and outputting the pointer instrument reading based on the instrument range includes the following steps: Unfold the pointer mask according to the calibrated geometric parameters and extract the x-coordinate position of the pointer in the unfolded space. And calculate the consecutive tick numbers corresponding to the pointer. : ; For uniform scale instruments, based on the continuous scale sequence number Total number of effective graduations and measuring range Calculate the physical output reading using the following formula. : ; Simultaneously, the reading confidence level is output. The reading confidence level is determined by at least two of the following factors: visible scale ratio, scale sequence residual, number of missing scales, and pointer detection quality. When the confidence level is less than a preset threshold, a manual review prompt is triggered.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Compared with the fixed ring band, the dynamic sampling band of the present invention can adaptively adjust the sampling range according to the distribution of the scale pixel radius in different angular directions. Since the dynamic inner radius and dynamic outer radius are determined by the radial distance distribution of the scale pixels in each angular window and the quantile value corresponding to the preset probability threshold, the scale truncation or background introduction caused by long and short scales, stains, reflections or printing offsets can be reduced.

[0017] (2) Compared with one-time geometric correction or one-time polar coordinate expansion, the present invention uses the scale sequence residual in the expansion space to back-calibrate the center, start and end angles and sampling band expansion ratio parameters, so that the geometric expansion parameters are constrained by the scale periodicity, thereby reducing the influence of initial center positioning error, start and end scale detection error and sampling band selection error on the final reading.

[0018] (3) Compared with simply filling in missing scales, this invention only reconstructs the theoretical coordinates of missing scales, does not forge the original texture of the occluded area, and incorporates the number of missing scales, scale sequence residuals and pointer detection quality into the credibility evaluation, avoiding taking the uncertain or false texture of the occluded area as a reliable identification result.

[0019] (4) In the parallax compensation implementation, the pointer height compensation result can be embedded into the reading link after closed-loop calibration; when the pointer height or camera intrinsic parameters are missing, this step can degenerate into a no-compensation process, so that the method can still be degenerated when relevant parameters are missing, and the applicability is improved.

[0020] (5) In the implementation of image acquisition by drones, inspection robots, fixed inspection cameras or industrial cameras, the present invention combines the image feature processing results with the parallax compensation of physical cameras and the mapping of instrument range readings, so that when the acquired image has strong reflection, low illumination, tilt or partial occlusion, it can reduce the pointer position offset caused by hardware parallax and the resulting reading amplification error. Attached Figure Description

[0021] Figure 1 This is the overall process for reading pointer-type instruments.

[0022] Figure 2 This is a schematic diagram of instrument area detection and initial perspective correction.

[0023] Figure 3 This is a schematic diagram of angle dynamic sampling band estimation.

[0024] Figure 4 This is a diagram showing the dynamic unfolding.

[0025] Figure 5 This is a schematic diagram of closed-loop calibration of the scale sequence.

[0026] Figure 6 This is a schematic diagram of the reconstruction of the occluded scale coordinates.

[0027] Figure 7 This is a schematic diagram of a dual-plane pointer parallax compensation model.

[0028] Figure 8 This is a schematic diagram of readings and reliability output. Detailed Implementation

[0029] Example 1 This embodiment proposes a pointer instrument reading method based on dynamic angle sampling bands and closed-loop calibration of scale sequences. This method establishes a dynamic sampling band through the angular radius distribution of a scale mask, and uses the scale sequence residuals and related constraints in the unfolded space to back-calibrate the original geometric parameters, making the scale distribution pattern a feedback constraint on the reading geometric parameters. The specific scheme is as follows: The overall flow of this embodiment is as follows: Figure 1 As shown, the process includes image input, instrument area detection, perspective correction, dynamic sampling band estimation, dynamic expansion, scale sequence closed-loop calibration, missing scale processing, pointer position extraction, reading calculation, and confidence output.

[0030] Input includes instrument images Calibrate the dial image and scale mask. Pointer mask Initial center Initial starting angle Initial and termination angles and instrument range parameters , Perspective correction or equivalent geometric correction can be performed on the instrument area to obtain a correction dial image; the scale mask and pointer mask Obtained from the calibration disk image, or synchronously mapped to the calibration disk coordinate system.

[0031] The instrument area detection and initial perspective correction process is as follows: Figure 2 As shown, the instrument area is determined by the detection frame, and the initial geometric parameters required for subsequent unfolding are established based on the corrected center, radius and start and end angles.

[0032] The geometric parameters to be read are uniformly represented as: Formula (1): .

[0033] in, , The coordinates of the current center of the circle; , The starting and ending angles are those that are expanded along the direction of the effective reading; , These are the safety extension ratio parameters for the inner and outer sides of the dynamic sampling band, respectively. The initial parameters can be obtained from... , , It is obtained from the preset expansion ratio parameter.

[0034] Based on the current geometric parameters, the scale mask Scale pixels in Convert to polar coordinates: Formula (2): , .

[0035] Effective reading angle range As a one-dimensional directed angular interval expanded along the reading direction, and divided into A window at an angle .when and When the determined effective reading arc segment crosses the zero angle point, for The obtained angle is used for phase expansion, so that , and Located in the same increasing or decreasing directed angular coordinate system. In the... Within each angular window, the set of statistical scale pixel radii. Based on a preset lower probability threshold Extracting low quantiles based on a preset upper probability threshold Extract the high quantiles and use them to determine the dynamic inner and outer radii of the window: Formula (3): , .

[0036] in, For the first The center angle of the window; To follow the lower probability threshold from The quantile values ​​extracted from the radial distance distribution. To follow the upper probability threshold from The quantile values ​​extracted from the radial distance distribution. You can use 0.05 or 0.08. 0.95 or 0.97 can be used; , This is the minimum bandwidth threshold used to prevent the sampling band from being too narrow. This yields the dynamic inner radius function that varies with the angle. and dynamic outer radius function .

[0037] When the number of tick marks within a certain angle window is less than a threshold When using adjacent window interpolation, global radius statistics, or a preset safety radius as a backoff value, the dynamic radius function is smoothed to reduce abrupt radius changes caused by local noise. Formula (4): , ,and .

[0038] in, These are the normalized smoothing weights in the finite smoothing kernel. After smoothing... and The dynamic sampling band of the angle can be denoted as: .

[0039] The angle-related dynamic radius estimation process is as follows: Figure 3 As shown, by statistically analyzing the distribution of scale pixel radii within each angular window, the dynamic inner radius and dynamic outer radius that change with the angle are obtained, thereby forming a sampling band that adapts to scale lengths and local interference.

[0040] Construction width is Height is linear expansion space Expand the horizontal coordinate of the space. Corresponding angle, ordinate Corresponding sampling radius: Formula (5): , .

[0041] Formula (6): , .

[0042] Formula (7): , .

[0043] in, , To correct the sampling coordinates in the disk coordinate system of the disk image or mask image, formulas (5) to (7) are used to map the scale mask and pointer mask to the linear expansion space. This expansion adopts a dynamic expansion method with different sampling radius ranges corresponding to different angle directions to adapt to the situation where the scale radius changes with the angle.

[0044] The dynamic unfolding process is as follows Figure 4 As shown, the dynamic angular sampling band in the calibration disk is mapped to the linear expansion space, so that the scale mask and pointer mask can be subsequently detected and fitted in a unified one-dimensional angular coordinate system.

[0045] The scale mask after unfolding Cumulative scale response along the radial direction: Formula (8): .

[0046] according to Extract the set of scale centers from peak values, connected component centers, or weighted centroids. For a uniformly graduated instrument, the fitted graduation sequence model is as follows: Formula (9): , .

[0047] in, For the first The theoretical horizontal coordinate of each scale point in the unfolded space. The distance between adjacent scale marks. For sequence offset, The effective number of scale points. Robust loss fitting is used. and : Formula (10): .

[0048] in, To match the test scale Matching tick numbers, This can be Huber loss, RANSAC in-point loss, or median absolute bias loss. The scaled sequence residual is defined as: Formula (11): .

[0049] The closed-loop objective function is constructed by combining the scale sequence residuals, the first and last scale boundary deviation penalty term, the missing scale proportion penalty term, and the dynamic radius smoothing constraint term. Formula (12): .

[0050] in, For geometric parameters The scale sequence residuals obtained by expansion, detection, and fitting; This is the normalized sum of the deviations between the theoretical first and last scales and the left and right boundaries of the unfolded space; The number of unmatched tick marks and the number of valid tick marks. The ratio; for and The absolute value or sum of squares of the second-order difference over adjacent angular windows; , , These are the weighting coefficients. All the above terms can be normalized to dimensionless quantities, or scaled using the weighting coefficients.

[0051] The closed-loop update process is as follows: using the initial parameters Starting from this point, dynamic sampling band estimation, dynamic unrolling, scale center detection, and scale sequence fitting are performed; within a defined neighborhood... Internal search enables Minimum parameter: Formula (13): .

[0052] The defined neighborhood range includes pixel-level offset of the center coordinates, angular offset of the start and end angles, and... , The expansion ratio changes. For Perform iterative calibration until... Less than the set threshold Or the number of iterations reaches the set limit. The calibrated geometric parameters are obtained. .

[0053] The closed-loop calibration process of the scale sequence is as follows Figure 5 As shown, the detection scale position in the unfolded space is matched with the theoretical scale sequence, and the center, start and end angles and sampling band expansion ratio parameters are corrected in reverse using sequence residuals, boundary deviations, missing scale ratios and smoothing constraints.

[0054] When partial occlusion, stains, or reflections cause some scale markings to go undetected, the theoretical coordinates of the missing scale markings are reconstructed in the calibrated unfolded space based on the scale sequence model. For the missing scale marking number... : Formula (14): .

[0055] Formula (15): .

[0056] Formula (16): .

[0057] in, It can be obtained by interpolation from the center radius of adjacent visible scales, or by taking the center radius of the dynamic sampling band. This step only reconstructs the theoretical coordinates of the missing scale, which are used for reading mapping and confidence evaluation, without falsifying the original image texture.

[0058] The process of reconstructing occluded scale coordinates is as follows Figure 6 As shown, for undetected scales, their theoretical positions are reconstructed only based on the calibrated scale sequence model, and these theoretical positions are used for reading mapping, start and end angle verification, and reliability evaluation.

[0059] pointer mask According to the calibrated geometric parameters Expand and extract pointer position The results can be obtained by using the median of the pointer pixel's horizontal coordinate, the intersection of the pointer's center line and the scale band, or robust straight-line fitting. This is based on the scale sequence parameters. , Calculate the consecutive tick numbers corresponding to the pointer: Formula (17): .

[0060] For uniformly graduated instruments, the reading is calculated based on the measuring range: Formula (18): .

[0061] For instruments with non-uniform scales, segmented interpolation can be performed based on adjacent major scale values. Let the adjacent major scale numbers be... and Their positions in the unfolded space are respectively and The corresponding range values ​​are respectively and When the pointer position When the value is between the two, the reading is calculated using the following formula: Formula (19): .

[0062] in, The output reading of a non-uniform scale instrument; To and The next adjacent major scale number; and This represents the position of adjacent main scale marks in the unfolded space. and This represents the range value corresponding to the adjacent main scale graduation.

[0063] Output reading reliability This is used to determine whether the reading is reliable and whether manual verification is necessary. Reliability Visible scale ratio Normalized scale sequence residuals Number or proportion of missing scale marks Pointer detection quality and parallax compensation displacement It is determined by one or more of the following factors. When Less than the preset confidence threshold When prompted, a prompt for manual review will be displayed.

[0064] The process of reading and confidence output is as follows: Figure 8 As shown, the system obtains the physical reading based on the pointer position, scale sequence model, and range mapping, and generates a reliability score by combining the visible scale ratio, scale sequence residual, number of missing scales, and pointer detection quality.

[0065] Example 2 In one implementation, an image of the pointer-type instrument to be identified is acquired. The image can be captured by a drone, inspection robot, fixed inspection camera, or industrial camera; the instrument area is obtained through target detection, template matching, edge localization, or manual configuration; perspective correction is performed on the instrument area to obtain the correction panel image. The scale mask is obtained through semantic segmentation, threshold segmentation, edge detection, or other equivalent methods. and pointer mask .

[0066] The initial geometric parameters are obtained based on the initial center and the start and end scales. The system converts the scale mask pixels into polar coordinates and divides the effective reading angle range into multiple angle windows after expanding it according to the reading direction; within each angle window, it statistically analyzes the radial distance distribution of the scale pixels, based on a preset lower probability threshold. Extract low quantiles based on a preset upper probability threshold. High quantiles are extracted to obtain dynamic inner and outer radii, which are then interpolated and smoothed to form dynamic angular sampling bands.

[0067] The system unfolds the scale mask into a linear space according to the dynamic sampling band and accumulates the scale response along the vertical direction. Extract the set of scale centers And fit a scale sequence model to represent the correspondence between scale numbers and scale positions in the linear expansion space. For a uniform scale instrument, this model can be represented as: For non-uniform scale instruments, the model can be represented as a piecewise linear model based on the main scale position. Then, according to formula (12), a model containing the scale sequence residuals is constructed. Penalty item for deviation between first and last scale points Missing scale ratio penalty item and dynamic radius smoothing constraint term Closed-loop objective function Within a defined neighborhood, the center, start and end angles, and sampling band expansion ratio parameters are iteratively calibrated until... If the change is less than the set threshold, or the number of iterations reaches the set upper limit, the calibrated geometric parameters and scale sequence model parameters are obtained.

[0068] The system will press the pointer mask according to To obtain the pointer height and camera intrinsic parameters, dual-plane parallax compensation can be performed on the pointer mask or pointer centerline before extracting the pointer unfolded position. And determine based on the calibrated scale sequence model The corresponding continuous scale number is used to calculate the pointer instrument reading based on the instrument's range and then output it. In an industrial field implementation, this output value can be used as the actual reading of the physical instrument for inspection records. For uniformly scaled instruments, it can be calculated according to... Calculate the serial number of continuous scales; for non-uniform scale instruments, segmented interpolation can be performed between adjacent main scales.

[0069] In one embodiment, when the number of scale pixels in a certain angle window is insufficient, the interpolation result of the inner and outer radii of adjacent angle windows can be used as the dynamic inner and outer radii of the window; if multiple consecutive windows are missing scale pixels, the global radial distance distribution of scale pixels is used, and quantile values ​​are extracted as backoff values ​​based on preset lower probability thresholds and upper probability thresholds; if the overall quality of the scale mask is low, low confidence is output and manual review is triggered.

[0070] In one embodiment, when highlights, stains, or occlusions cause certain scale peaks to go undetected, the system does not perform texture spoofing on the original image, but instead performs texture spoofing in the unfolded space based on... Calculate the theoretical coordinates of the missing scale marks and map them back to the disk surface coordinates using formulas (15) and (16). These coordinates are only used for reading mapping, start and end angle verification, and reliability evaluation. The more missing scale marks there are, the lower the reliability.

[0071] In one embodiment, after obtaining the pointer height and camera internal reference In the implementation method, the dial plane is represented as Representing the pointer plane as Let the rotation matrix in the camera's extrinsic parameters be... The translation vector is Then the dial plane homography matrix and the pointer plane homography matrix are respectively , For the homogeneous image coordinates of the pointer plane projection points. You can press Homogeneous image coordinates compensated to the dial plane .when or If the parallax is unknown, this compensation step is not performed. The above parallax compensation step can be embedded in the calibrated readout link.

[0072] Dual-plane pointer parallax compensation model, such as Figure 7 As shown, when the pointer height and camera intrinsic parameters are known, the projection point on the pointer plane is compensated to the dial plane to reduce the reading deviation caused by the tilt angle or pointer height.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reading pointer-type instruments, characterized in that, Includes the following steps: Acquire an image containing pointer-type instruments and obtain the calibration dial image, scale mask, pointer mask, and initial geometric parameters; The geometric parameters to be read are uniformly represented as: ； in, The current center coordinates of the circle. and The starting and ending angles are those unfolded along the direction of the effective reading. and These are the safety extension ratio parameters for the inner and outer sides of the dynamic sampling band, respectively. Based on current geometric parameters Convert the tick marks in the tick mask to polar coordinates and divide the effective reading angle range into... A window at various angles; Within each angular window, the radial distance distribution of the scale pixels within that window is statistically analyzed. Low and high quantiles are extracted based on preset lower and upper probability thresholds, respectively. The dynamic inner and outer radii corresponding to that angular window are then determined based on these low and high quantiles. This is then combined with... and A dynamic sampling band that varies with angle is obtained; Based on dynamic sampling bands, the scale mask and pointer mask are mapped to the linear expansion space; Extract the set of scale centers in the linear expansion space, fit a scale sequence model to represent the correspondence between scale numbers and scale positions in the linear expansion space, and calculate the scale sequence residual; construct a closed-loop objective function that includes the scale sequence residual, the first and last scale boundary deviation penalty term, the missing scale ratio penalty term, and the dynamic radius smoothing constraint term. Within a defined neighborhood, based on the closed-loop objective function... The center, start and end angles, and dynamic sampling band expansion ratio parameters in the circle are iteratively calibrated until the change in the closed-loop objective function is less than the set threshold, or the number of iterations reaches the set upper limit, and the calibrated geometric parameters are obtained. Based on the calibrated geometric parameters and scale sequence model, the continuous scale number corresponding to the pointer position is determined, and the pointer instrument reading is calculated and output according to the instrument range.

2. A method for reading a pointer-type instrument according to claim 1, characterized in that, The method for calculating the dynamic sampling band is as follows: In the Within each angle window, the dynamic inner radius is calculated using the following formula. and dynamic outer radius : ； ； in, For the first The center angle of the window. This is the set of scale pixel radii within the window. To follow the lower probability threshold Quantile values ​​extracted from the radius distribution To follow the upper probability threshold Quantile values ​​extracted from the radius distribution You can use 0.05 or 0.

08. 0.95 or 0.97 can be used; , This is the preset minimum bandwidth threshold.

3. A method for reading a pointer-type instrument according to claim 1, characterized in that, When the number of scale pixels within a certain angle window is lower than the pixel number threshold, interpolation is performed using the dynamic inner radius and dynamic outer radius of adjacent angle windows, or quantile values ​​are extracted from the global scale pixel radius distribution according to preset lower and upper probability thresholds as backoff values.

4. A method for reading a pointer-type instrument according to claim 1, characterized in that, The dynamic inner radius and dynamic outer radius are smoothed according to the following formula to obtain the smoothed dynamic inner radius function. and dynamic outer radius function To form the final dynamic sampling band of angles: in, Let be the normalized smoothing weights in the finite smoothing kernel, and .

5. A method for reading a pointer-type instrument according to claim 1, characterized in that, For a uniformly graduated instrument, the graduation sequence model is represented as: in, For the first The theoretical abscissa of each scale in the linear expansion space The distance between adjacent scale marks. For sequence offset; and with and Given the parameters to be fitted, robust fitting is performed using the following formula to calculate the scale sequence residuals. : ； ； in, For the first The detected x-coordinate of the scale center, For the corresponding tick mark, This can be Huber loss, RANSAC in-point loss, or median absolute deviation loss.

6. A method for reading a pointer-type instrument according to claim 1, characterized in that, For non-uniform scale instruments, the scale sequence model is a piecewise linear model based on the position of the main scale. The reading is obtained by piecewise interpolation based on the adjacent main scale positions and corresponding range values ​​in the interval where the pointer position is located.

7. A method for reading a pointer-type instrument according to claim 1, characterized in that, Closed-loop objective function The specific construction formula is as follows: ； in, For the current geometric parameters The scale sequence residuals obtained from the calculation are as follows: This is a penalty term for the deviation between the theoretical first and last scales and the left and right boundaries of the linear expansion space. Penalty for missing scale ratios due to mismatch. For dynamic radius function and Second-order difference smoothing constraint term on adjacent angle windows , , These are the weighting coefficients.

8. A method for reading a pointer-type instrument according to claim 7, characterized in that, Iterative calibration includes: under the current geometric parameters Within the defined neighborhood, iteratively search according to the following formula to make the closed-loop objective function Minimum candidate parameter: The neighborhood range is defined by pixel-level offset of the center coordinates, angular offset of the start and end angles, and... , The expansion ratio changes; when the change in the closed-loop objective function between two consecutive iterations is less than a set threshold, or when the number of iterations reaches a set upper limit, the iteration stops and the calibrated geometric parameters are output. .

9. A method for reading a pointer-type instrument according to claim 1, characterized in that, When missing scale values ​​exist, the parameters of the calibrated scale sequence model should be used. and and missing scale numbers The theoretical coordinates of the missing scale in the linear expansion space and the theoretical coordinates in the disk coordinate system are reconstructed using the following formula. : ； ； ； in, To expand the width of the space, The radius is obtained by interpolation of the center radius of adjacent visible scales or the center radius of the dynamic sampling band; the theoretical coordinates of missing scales are used for reading mapping and credibility evaluation, and are not used to forge the original image texture.

10. A method for reading a pointer-type instrument according to claim 1, characterized in that, The steps of calculating and outputting the pointer instrument reading based on the instrument range include the following: Unfold the pointer mask according to the calibrated geometric parameters and extract the x-coordinate position of the pointer in the unfolded space. And calculate the consecutive tick numbers corresponding to the pointer. : ； For uniform scale instruments, based on the continuous scale sequence number Total number of effective graduations and measuring range Calculate the physical output reading using the following formula. : ； Simultaneously, the reading confidence level is output. The reading confidence level is determined by at least two of the following factors: visible scale ratio, scale sequence residual, number of missing scales, and pointer detection quality. When the confidence level is less than a preset threshold, a manual review prompt is triggered.

Images