A reading recognition method and system for a square pointer watch

Abstract

The application relates to the technical field of pointer table reading identification, in particular to a reading identification method and system of a square pointer table, and the application proposes the following scheme: an image of a dial of a target square pointer table is acquired; a plurality of angle rotation processes are performed on the image to generate rotation images covering different scale sections; the rotation images are cropped to obtain a target image area including a scale frame and a digital frame; an image area to which a pointer belongs and a position of the image area relative to a scale line are determined through edge detection; and a scale value indicated by the pointer is calculated based on pixel distribution. Compared with the prior art, the application improves the reading accuracy and identification stability of the square pointer table under the conditions of image skew and pointer criticality, and is suitable for various industrial and equipment monitoring scenes.

Description

Technical Field

[0001] This application relates to the field of pointer meter reading recognition technology, and in particular to a method and system for recognizing the readings of a square pointer meter. Background Technology

[0002] Analog dial devices, such as voltmeters, ammeters, and thermometers, are widely used in various scenarios including industrial control, environmental monitoring, power systems, and equipment maintenance. With the advancement of automation and intelligence, how to perform image recognition and extract valid readings from traditional pointer meters has become a critical problem that urgently needs to be solved.

[0003] Currently, image recognition methods for round analog watches are relatively mature, typically employing techniques such as perspective correction, dial center positioning, and pointer angle analysis to achieve automatic reading. However, for square analog watches with their unique structural layout, the horizontal distribution of their scale intervals and the left-right swing of their pointers make conventional polar coordinate transformations and radial feature analysis difficult to apply directly. Furthermore, square analog watches are prone to issues such as angular misalignment, structural obstruction, or misjudgment of pointer critical positions during photography, further increasing the difficulty of image recognition.

[0004] For example, Chinese Patent CN113989513A discloses a method for recognizing the readings of a square pointer instrument, comprising the following steps: (1) extracting an image of the instrument's rectangular frame; (2) correcting the image; (3) performing binary thinning on the corrected image; (4) obtaining the center of the pointer's movement, the centroid of the upper left corner A, and the midpoint of the pointer's center line segment, connecting the centroid of the upper left corner A and the center of the pointer's movement as a baseline, and connecting the midpoint of the pointer's center line segment and the center of the pointer's movement as the pointer's needle line; (5) obtaining the angle between the baseline and the pointer's needle line, and fitting a linear equation reflecting the reading value based on the angle; (6) adjusting the fitted equation using the error distribution. This application solves the problem of square instruments lacking circular texture and the inaccuracy of finding the center of the circle where the scale lines are located, thus improving accuracy and precision.

[0005] The above patents have the following problems: they lack fine-grained analysis of image structure and the ability to dynamically adjust local areas, making it difficult to maintain high recognition accuracy under complex backgrounds and diverse angles. To solve the above problems, this application designs a reading recognition method and system for a square pointer meter. Summary of the Invention

[0006] The technical problem to be solved by this application is to address the shortcomings of the prior art by providing a method and system for recognizing the readings of a square pointer meter. The method involves acquiring an image of the dial of the target square pointer meter; rotating the image at multiple angles to generate a rotated image covering different scale segments; cropping the rotated image to obtain a target image region including scale frames and number frames; determining the image region to which the pointer belongs and its position relative to the scale lines through edge detection; and calculating the scale value indicated by the pointer based on pixel distribution.

[0007] To achieve the above objectives, this application provides the following technical solution:

[0008] A method for recognizing the readings of a square pointer meter, the method comprising:

[0009] Obtain the dial image of the target square indicator table;

[0010] The dial image is rotated at multiple angles to generate rotated images corresponding to different scale intervals;

[0011] The rotated image is cropped to obtain a target image region including a scale frame and a number frame;

[0012] The target image region is processed sequentially by an edge detection algorithm to determine the relative position of the target image region to which the pointer belongs and the pointer frame corresponding to the pointer relative to the tick line.

[0013] Based on the image region to which the pointer belongs and the relative position of the pointer frame with respect to the scale line, the pixel distribution of the pointer frame is calculated to obtain the scale value indicated by the pointer.

[0014] The dial image is rotated at multiple angles to generate rotated images corresponding to different scale intervals, including:

[0015] The dial image is rotated at a preset first angle to generate a first rotated image;

[0016] The dial image is rotated at a preset second angle to generate a second rotated image;

[0017] The dial image is rotated at a preset third angle to generate a third rotated image.

[0018] The cropping of the rotated image to obtain an image region including a scale frame and a number frame includes:

[0019] Traverse the rotated image to identify multiple scale lines whose length is greater than a preset length threshold;

[0020] The scale lines are filtered one by one according to the preset judgment conditions to obtain the scale line pairs that meet the judgment conditions. The image area between the scale line pairs is used as the target image area for cropping.

[0021] Perform iterative operations until the first rotated image, the second rotated image, and the third rotated image are all cropped, resulting in the first target image region, the second target image region, and the third target image region.

[0022] The judgment criteria include that the two scale lines are perpendicular in direction and aligned in the horizontal position in the rotated image.

[0023] The step of processing the target image region sequentially using an edge detection algorithm to determine the relative position of the target image region to which the pointer belongs and the pointer frame corresponding to the pointer relative to the tick marks includes:

[0024] Edge detection is performed on the first target image region, the second target image region, and the third target image region respectively to obtain edge features;

[0025] Based on the extracted edge features, identify the edge features that form closed pointer boxes and determine the image region where the corresponding pointer boxes are located;

[0026] Extract the edge features of the tick marks in the image region to which the pointer bounding box belongs;

[0027] Calculate the pixel coordinates of the center position of the pointer frame in the image coordinate system, and determine the relative position of the pointer frame in the corresponding image region based on the pixel coordinates and the pixel spacing between the tick mark edge features.

[0028] Edge detection is performed on the first target image region, the second target image region, and the third target image region, respectively, including:

[0029] The target image region is processed by an edge detection algorithm to obtain a binary edge map;

[0030] Contour tracing is performed on the binary edge map to extract a set of edge segments composed of continuous pixels in the binary edge map. The set of edge segments corresponds to the structural boundary in the binary edge map.

[0031] For each edge segment in the set of edge segments, parameter features are extracted. The parameter features include the start and end coordinates, pixel length, direction, edge strength, and connectivity between adjacent edge segments.

[0032] Based on the parameter characteristics, calculate the edge features.

[0033] Calculate the pixel distribution of the pointer frame to obtain the scale value indicated by the pointer, including:

[0034] Within the image area to which the pointer frame belongs, determine the actual scale values ​​of the first and second scale lines adjacent to the pointer frame;

[0035] Based on the relative position and the actual scale value, the scale value corresponding to the pointer frame is calculated by linear interpolation.

[0036] If the image region where the pointer box is located is not unique, the method further includes:

[0037] Based on the rotation angle corresponding to the image region, the geometric structure of the identified pointer frame is inversely transformed from the coordinate system of the corresponding image region to the original image coordinate system;

[0038] In the original image coordinate system, the pointer frames are geometrically synthesized, and the pointer geometry is calculated.

[0039] The pointer geometry is projected onto the image region coordinate system respectively;

[0040] Calculate the degree of boundary overlap and region envelope relationship of the pointer geometry in the corresponding image region coordinate system to obtain the first overlap and the second overlap.

[0041] The first overlap and the second overlap are used to determine the image region corresponding to the mapping structure with the larger overlap, which is then used as the target image region to which the pointer belongs.

[0042] A reading recognition system for a square pointer meter, the system comprising:

[0043] The image acquisition module is used to acquire the dial image of the target square pointer watch;

[0044] The image processing module is used to rotate the dial image from multiple angles and crop the rotated image to obtain a target image area including a scale frame and a number frame.

[0045] The edge analysis module is used to perform edge detection on the target image region, identify the edge features that form a closed pointer box, and determine the relative position of the pointer box with respect to the scale line.

[0046] The scale calculation module is used to calculate the scale value indicated by the pointer frame based on the actual scale difference between the relative position and the scale line.

[0047] The edge analysis module includes:

[0048] The edge detection unit is used to perform edge detection on the target image region and generate the corresponding binary edge map.

[0049] The contour extraction unit is used to perform contour tracking operation on the binary edge map and extract a set of edge line segments composed of continuous pixels.

[0050] An edge parameter calculation unit is used to extract parameter features, including start and end coordinates, pixel length, direction, edge strength, and connectivity, from the edge segments in the edge segment set.

[0051] The pointer recognition unit is used to identify the pointer frame that constitutes a closed contour structure based on the parameter characteristics of the edge line segment, and to determine the image area where the pointer frame is located.

[0052] Compared with the prior art, the beneficial effects of this application are:

[0053] This application combines image rotation and cropping with region structure analysis, enabling it to adapt to image input under multi-angle shooting conditions and avoiding the recognition failure problem caused by pointer tilt or skew when extracting fixed regions using traditional methods. By accurately cropping the tick marks and number boxes in the rotated image, explicit alignment of local structures is achieved, improving the registration accuracy between the pointer and the tick marks. Through edge detection and closed pointer box recognition, the true pointer region can be stably located, effectively handling abnormal situations such as pointer crossing regions and critical swings. Attached Figure Description

[0054] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0055] Figure 1 This is a schematic diagram illustrating an exemplary application scenario of an embodiment of this application;

[0056] Figure 2 This is a flowchart illustrating a method for recognizing the readings of a square pointer meter according to an embodiment of this application.

[0057] Figure 3 This is a schematic diagram of a rotated image according to an embodiment of this application;

[0058] Figure 4 This is a schematic diagram of the pointer positioning process in an embodiment of this application. Detailed Implementation

[0059] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0060] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0061] This application applies to automatic reading scenarios involving the recognition of scale values ​​on square pointer meters using image recognition technology, and is particularly suitable for industrial instruments with a fan-shaped scale structure and finely subdivided pointer areas. Industrial instruments typically consist of arc-shaped scale lines and a single pointer. Application scenarios include, but are not limited to:

[0062] Industrial instruments such as square ammeters and voltmeters with a wide dial scale range and requiring partitioning during the identification process;

[0063] The instrument image has a fixed shooting angle and a symmetrical structure, making it suitable for segmenting the scale range image by rotating it.

[0064] The pointer is a slender, symmetrical structure, and its position relative to the pixels of the scale lines can be used to indicate the final scale value.

[0065] The recognition process applicable to this application typically includes rotating the original dial image from multiple angles, cropping it according to a preset method to obtain several target image regions, identifying the positional relationship between the scale lines and the pointer frame in each image region, and deriving the scale value indicated by the pointer based on the pixel distribution relationship.

[0066] It should be noted that when some pointers are close to the region boundary, the complete pointer bounding box may be identified in multiple image regions. This application does not rely on a single confidence level judgment or position threshold division, but instead uses structural synthesis and coordinate mapping to reconstruct a unified pointer geometry from multiple images. Based on this, the final target image region is selected, thereby ensuring the stability and reliability of the reading process.

[0067] It is understood that the reading method proposed in this application does not rely on the division of the dial structure template, nor does it require the preset of fixed pointer path features. It has good versatility and practicality and is suitable for pointer scale recognition tasks under complex lighting and shooting conditions in industrial sites.

[0068] Please see Figure 1 This figure is a schematic diagram of an exemplary application scenario provided by an embodiment of this application.

[0069] like Figure 1As shown, the method of this application is applicable to the reading recognition scenario of a square pointer meter with a fan-shaped scale distribution structure and a single pointer component. The pointer meter to which this application is applicable typically includes a rectangular dial frame, arc-shaped scale lines, numerical markers indicating scale values, and a rotatable pointer structure extending from the central axis to the scale range.

[0070] Figure 1 The square pointer meter shown is a typical analog display instrument, widely used in industrial measurement scenarios such as voltage, current, pressure, or flow. The scale lines are distributed in a ring shape within the dial, generally covering a sector angle range of approximately 0 to 150 degrees, with the starting and ending points corresponding to the minimum and maximum values, respectively. The pointer rotates along an arc path to indicate the scale position corresponding to the current measurement value.

[0071] In practical applications, an image acquisition device is used to capture an image of the dial of a square analog watch. The method of this application involves rotating the original image by an angle based on the captured dial image and cropping out several target image regions containing the scale frame and the number frame.

[0072] exist Figure 1 In the structure shown, the dial scale is divided into intervals of 0 to 150 degrees, with the numerical markings located on the main scale lines. The pointer, with the central axis as its base point, points to a specific interval at any given time. Due to the wide range of the pointer dial and the characteristics of dense scale lines and a slender pointer structure, conventional methods based on direct recognition of the entire image have limitations in terms of accuracy or computational efficiency.

[0073] Understandable Figure 1 The square pointer meter shown is only a typical application example. The method described in this application is also applicable to other analog pointer instruments with similar structures and scale layouts, including but not limited to multi-range pointer meters, embedded mechanical dials, and combined industrial display units.

[0074] Next, with reference to the accompanying drawings, a method for recognizing the readings of a square pointer meter provided in an embodiment of this application will be described. Figure 2 The method shown includes the following steps S1-S5:

[0075] S1: Obtain the dial image of the target square indicator table;

[0076] In this embodiment, the acquired dial image is a complete analog dial image captured by an industrial camera or other image acquisition device, including the complete scale area, the center pointer, and the numerical markers. After conventional preprocessing, this image can be used as the input source image for subsequent recognition processes. The image format can be JPG, PNG, or BMP, and the image resolution can be flexibly set according to the device configuration. Those skilled in the art will understand that as long as the image completely contains the dial area and the structural information between the pointer and the scale lines can be distinguished, the image input requirements of this application are met, and this application does not impose any limitations on this.

[0077] S2: Rotate the dial image from multiple angles to generate rotated images corresponding to different scale intervals;

[0078] In the aforementioned application scenario, the dial image contains multiple scale segments arranged in an arc pattern, and the pointer rotates along the central axis to indicate the current reading. Because the scale lines are arranged in a fan shape, the direction and angle of each scale segment in the image are different, and the structural distribution has spatial differences.

[0079] In existing technologies, image recognition typically takes the entire dial image as input and uses end-to-end image regression networks or image segmentation algorithms to directly predict the pointer angle or scale value.

[0080] It is understandable that existing technologies have the following shortcomings:

[0081] Full-image recognition requires the model to learn the entire scale structure, which involves a large amount of training data.

[0082] The model results lack interpretability, making it difficult to determine attribution and trace location.

[0083] When the pointer swings in the boundary scale area, the reading may become unclear or jump.

[0084] This embodiment employs a structured recognition scheme, which rotates the entire dial image from multiple angles to divide it into multiple target image regions according to scale segments. Each region corresponds to a specific scale interval, containing complete scale frames and digit frame information. Within these structurally standardized regions, the pointer frame can establish a clear relative relationship with the scale lines, facilitating pixel position measurement and reading calculation.

[0085] Those skilled in the art will understand that the fundamental purpose of rotation processing is to transform the differently oriented scale structures in the original image to a unified orientation, in order to support subsequent independent identification and attribution analysis of multiple regions. This application is not based on global modeling of the entire image, but rather on the geometric structure analysis of local scale segments; therefore, rotation processing constitutes a necessary processing step in the method of this application.

[0086] S3: Crop the rotated image to obtain a target image region including a scale frame and a number frame;

[0087] In this embodiment, image traversal is performed on each rotated image to identify edge structure line segments with a length exceeding a set threshold. These line segments must satisfy the vertical structure judgment condition and their positional relationship must meet the horizontal alignment requirement. Based on the pixel coordinate boundaries between the selected scale line pairs, the corresponding target image region is cropped. Each target region contains a continuous scale structure and its corresponding numerical marker.

[0088] S4: The target image region is processed sequentially using an edge detection algorithm to determine the relative position of the target image region to which the pointer belongs and the pointer frame corresponding to the pointer relative to the scale line;

[0089] In this embodiment, edge detection is performed on each target image region. Standard algorithms such as Canny, Sobel, and Laplacian can be used to obtain binary images of the image edges. Closed edge regions are then extracted based on contour tracking, and pointer bounding box structures are extracted by combining edge length, direction, and relative position parameters. If closed pointer bounding box structures are detected in multiple regions, the pointer bounding box structures within each region are inversely transformed to the original image coordinates for geometric synthesis. The synthesized structure is then projected back into each rotated image, and the degree of overlap between it and the structures within the region is calculated to select the assigned region.

[0090] S5: Based on the image region to which the pointer belongs and the relative position of the pointer frame to the scale line, calculate the pixel distribution of the pointer frame to obtain the scale value indicated by the pointer;

[0091] In this embodiment, after determining the target image region to which the pointer frame belongs, the structures of the two main tick marks adjacent to the pointer frame within that region are extracted, and the actual tick mark difference range is calculated based on their pixel spacing. Based on the offset distance of the center pixel of the pointer frame relative to the starting tick mark, the precise tick mark value corresponding to the pointer frame is calculated using a linear interpolation method. Pixel offset can be measured using Euclidean distance, or angle quantization can be achieved using the standard vector inner product formula after coordinate transformation.

[0092] Before detailing the specific technical aspects of each processing step in the embodiments of this application, it must be emphasized again that in a typical square pointer dial structure, because the scale intervals are arranged in an arc shape along the edge of the dial, and their direction varies significantly with the angle, directly processing the entire image in a uniform manner results in problems such as structural misalignment, distorted boundary positions, and incomparable pixel distribution. Furthermore, the pointers in the image are elongated, high-contrast edge structures that may appear simultaneously in multiple cropped views, especially when pointing to the boundary areas of scale segments, making it highly likely that the same pointer will be repeatedly identified in multiple areas.

[0093] The processing method employed in this embodiment does not directly target the pointer position by determining the entire image. Instead, it aligns and standardizes the structures of different scale segments through multi-angle rotation, and then performs local edge extraction and structure judgment in each view to achieve classification separation under structured recognition. In engineering practice, this is reflected in guiding view cropping through structural direction and establishing a correspondence between local structures and pixel behaviors under a unified coordinate system. Compared to methods that directly regress based on the entire image structure, this method is more suitable for scenarios with square pointer tables where scale segments are scattered, especially in applications with limited image resolution or high sensitivity to boundary recognition, exhibiting stronger structural adaptability and localization robustness.

[0094] It should be noted that the scale interval division described in this embodiment is based on... Figure 1 Taking the square pointer dial as an example, the dial scale structure is divided into three intervals: 0–50, 50–100, and 100–150. The division method is intended to intuitively demonstrate the processing flow of the method described in this application for extracting the target image region at different rotation angles. It is only an exemplary illustration for easy understanding and does not constitute a limitation on the scale range, scale density, or number of intervals.

[0095] It should be understood that the structural configuration used in this embodiment is only for... Figure 1 The description is presented in the form shown. In specific applications, it can be flexibly adjusted according to the actual scale layout of the dial, recognition accuracy requirements, or image resolution settings. This application does not limit this.

[0096] Next, the part of the method of this application regarding image rotation will be further elaborated.

[0097] For example, regarding image rotation, please refer to... Figure 3 To understand, Figure 3 This is a schematic diagram of a rotated image according to an embodiment of this application. Figure 3 The image shows the effect of rotating the original dial image, with the scale range 100–150 rotated to an approximately horizontal position.

[0098] Understandably, by rotating the original image structure as a whole, the local scale segments that were originally arranged along the fan shape present a regular and linearly aligned layout in the new coordinate system. The pointer direction, scale box boundaries and number marks in the image are more easily separated in space, which significantly improves the reliability of edge detection and pointer box attribution judgment.

[0099] In an optional embodiment, the present application rotates the dial image at preset first angles, second angles, and third angles respectively to generate corresponding first rotated image, second rotated image, and third rotated image.

[0100] Specifically, the preset rotation angle can be set according to the scale distribution characteristics of the target dial, so as to... Figure 1 Taking the structure shown as an example, the directions of each scale segment in the original image are different, so the rotation angles can be set to 10 degrees, 45 degrees and 75 degrees respectively, corresponding to the three interval segments [100°, 150°], [50°, 100°] and [0°, 50°] respectively.

[0101] Furthermore, the rotation operation can be implemented based on an affine transformation model, using a two-dimensional rotation matrix to rotate and map the image coordinate system, with the center point being the geometric center of the image. To avoid edge cropping or pixel loss caused by rotation, the image size can be expanded before rotation. The filling method can be a fixed value, edge copying, or mirror reflection. The specific strategy can be set according to the surface of the specific square pointer table, and this application does not limit it. During the rotation process, the coordinates (x, y) of each pixel in the image will be transformed into new coordinates (x', y') according to the set rotation angle θ, which can be achieved through the following two-dimensional affine matrix:

[0102]

[0103] Where, x c and y c The coordinates are the center coordinates of the dial image. The rotated image retains its original resolution, which facilitates subsequent cropping operations and structural alignment.

[0104] In a preferred embodiment, the rotation angle of this application is not fixedly set as a preset value, but is determined by automatically calculating the main direction angle of the scale interval in the current image through image analysis processing of the scale line direction in the original dial image. The specific steps are as follows:

[0105] S2.1: Extract the character region from the dial image and identify the bounded numeric marker regions in the image using an image segmentation algorithm;

[0106] Specifically, the purpose of this step is to separate the numerical portions representing the scale from the dial image, serving as the structural basis for subsequent spatial positioning and orientation derivation. In a typical square analog dial structure, the numerical markers for each scale segment are usually located near the scale lines, exhibiting a stable font shape and clear boundary structure. Therefore, image segmentation techniques can be used to detect character regions in the image, thereby identifying image blocks with numerical meaning.

[0107] In this embodiment, character region extraction employs a processing logic combining connected component analysis and contour filtering. First, the original dial image is grayscaled and edge-enhanced. Then, preliminary binarization is performed by setting a binary threshold to extract high-contrast regions that may contain characters. In the resulting binary image, closed regions are extracted using connected component analysis, and non-character regions are filtered out based on geometric features (such as aspect ratio, area, and rectangle fitting accuracy). To further improve recognition accuracy, lightweight character detection networks (such as CRNN or EAST) can be used to supplement the remaining regions, enhancing the ability to recognize occluded or low-contrast characters.

[0108] S2.2: Output the minimum bounding rectangle corresponding to each digit-marked area and calculate its center pixel coordinates;

[0109] Specifically, this step is a prerequisite for establishing a point-to-space structure for deriving the direction of the scale interval. The required geometric data is the center coordinates of the digital marker area. The character area itself may exhibit slight rotation or font distortion in the image, so directly using boundary points for positioning is prone to noise interference.

[0110] In this embodiment, for each confirmed digital marker region, a rectangle fitting algorithm is used to obtain the bounding rectangle of its boundary, thus obtaining the bounding rectangle of the region. The bounding rectangle can be determined based on the minimum area criterion, ensuring that the generated rectangle shape closely fits the character body, avoiding excessive expansion or inclusion of redundant background caused by irregular character structure.

[0111] Furthermore, after obtaining the minimum bounding rectangle, its center pixel coordinates are calculated based on the rectangle's geometric center formula. These center coordinates will serve as a reference point for the digital marker in the image coordinate system, used for subsequent scale pair construction and direction vector calculation.

[0112] Understandably, this step does not perform character content recognition, but only locates the spatial position of the number markers to avoid being affected by the accuracy of character recognition, thereby improving the stability of the direction determination process.

[0113] S2.3: Sort the minimum bounding rectangles of all extracted digit marking regions according to their horizontal positions, construct scale digit pairs one by one, and filter out invalid combinations that are too close or suspected of being duplicate markings, retaining digit pairs that meet the spatial distance threshold and have high character recognition confidence.

[0114] Specifically, this step aims to create directional reference lines for calculating rotation angles. The lines connecting the scale pairs structurally correspond to an actual scale interval. Since images may contain duplicate annotations, anomaly detection, or blurred areas, constructing all point pairs without filtering would introduce a large number of structurally unstable angular references, affecting subsequent judgments.

[0115] In this embodiment, all rectangles are first sorted from left to right according to the x-coordinate of their center pixels to form a sequence structure. Then, adjacent digit marker regions are sequentially selected using a sliding window method to form digit pairs, initially creating candidate scale segments. To improve the stability of the constructed angle reference line, a spatial spacing check is performed on each digit pair, excluding combinations where the center point spacing is less than a set threshold, thus avoiding misclassifying redundant digit recognition results as valid markers.

[0116] Preferably, if the recognition stage also has the function of character recognition, the character recognition confidence level can be combined for screening to retain combinations with high recognition confidence and clear structure, so as to improve the representativeness of the evaluation. This application will not elaborate on this point.

[0117] S2.4: For each valid scale digit pair, calculate the direction vector corresponding to the line connecting its center points, and use the angle between the direction vector and the horizontal axis of the image as the local rotation angle to obtain a candidate angle set;

[0118] Specifically, the goal of this step is to establish the directional representation of the scale segments in a geometric coordinate system, and to construct the set of angles required for image rotation. In this application, the image as a whole remains undistorted; the change in scale direction mainly stems from the fan-shaped layout of the dial structure. Therefore, the center line connecting two digit markers can be considered as the main axis of the scale for that segment, and the direction of the line represents the directional trend of the scale in that segment.

[0119] In this embodiment, for each confirmed valid digit pair, a spatial straight line is constructed based on its two center pixels, and the direction vector of this line in the image coordinate system is recorded. Then, the angle between the direction vector and the direction vector of the horizontal axis of the image coordinate system is calculated to obtain the direction angle value of the current digit pair's scale segment. This direction angle can be solved based on vector projection relationships, and the direction angle value is added to a candidate angle set, representing possible rotation correction directions in the image.

[0120] Understandably, the constructed candidate angle set comprehensively covers the main scale segment directions included in the current dial structure, providing data for subsequent selection of representative angles and execution of image rotation operations. The candidate angle set originates from the physical positional relationships of the digital markers themselves, possessing structural consistency and enabling adaptive derivation of rotation angles under various shooting tilts and dial style changes.

[0121] S2.5: Perform median filtering on the candidate angle set, exclude abnormal angles with large deviations, retain several representative angles as the target rotation angle set, and perform rotation operation on the original image using the target rotation angle set to generate a rotated image;

[0122] Specifically, the purpose of this step is to select angles with stable orientation and strong representativeness from multiple candidate angles, avoiding the introduction of incorrect rotation directions due to individual outliers, which could affect the accuracy of cropping the target image region. Images may contain noisy digital markers, recognition errors, or abnormal digital arrangement; some values ​​in the candidate angle set may significantly deviate from the trend of most directions. Without processing, this can easily cause image structure distortion.

[0123] In this embodiment, the candidate angle set is first analyzed for angle distribution. By calculating the median of the angle values, the overall trend direction is determined, and abnormal angles that deviate from the threshold range are eliminated based on this median. Subsequently, from the remaining angle values, three representative directional angles (or the number can be set according to the dial) are selected as the final rotation angle set based on the directional distribution range.

[0124] Furthermore, affine rotation operations are performed on the original image using the set of rotation angles, generating a rotated image for each angle. This rotated image serves as the source image for subsequent structural cropping and target image region extraction. To avoid image edge truncation or cropping distortion after rotation, fixed pixel compensation or mirror filling can be performed on the image edges before rotation to ensure that the rotation center remains consistent and the pixel space is continuous.

[0125] Next, we will further elaborate on the image cropping aspect of the method in this application.

[0126] It is easy to understand that the above steps yielded three rotated images, and the horizontally aligned area in each rotated image is the image area to be cropped.

[0127] In one example, the specific steps of S3 are as follows:

[0128] S3.1: Traverse the rotated image and identify multiple scale lines whose length is greater than a preset length threshold;

[0129] Specifically, the core objective of this step is to identify structural edges in the rotated image that conform to the geometric features of the tick marks, providing a basis for the subsequent construction of tick mark pairs.

[0130] In the example of this application, the length of the scale lines is not uniform. For example, the scale lines for the intervals of 0 degrees, 50 degrees, 100 degrees and 150 degrees are longer than the other scale lines. What needs to be done in this embodiment is to extract the longer scale lines separately in order to determine the interval range of the target image area.

[0131] Those skilled in the art will understand that, given the known differences in the lengths of the scale lines, how to process a rotated image to identify scale lines whose lengths exceed a preset threshold is prior art, and this application will not elaborate on it here.

[0132] S3.2: The scale lines are filtered one by one according to the preset judgment conditions to obtain the scale line pairs that meet the judgment conditions. The image area between the scale line pairs is used as the target image area for cropping. The judgment conditions include that the two scale lines are perpendicular in direction and aligned in the horizontal position in the rotated image.

[0133] Specifically, this step involves selecting tick line pairs with spatial symmetry and structural regularity from all tick lines that meet the length threshold. The boundary range of the target image region formed by these tick line pairs is then used as the positioning basis for subsequent cropping operations. Since the tick structure in the rotated image is standardized and aligned, directional consistency and lateral alignment can be used as the criteria for constructing tick line pairs.

[0134] In this embodiment, for all tick marks that meet the extraction conditions of S3.1, the direction vector of each line segment is calculated, and line segments with excessively large deviations from the vertical direction are filtered out to ensure that the constructed tick mark pairs have a basis for structural alignment. Subsequently, all line segments are sorted according to their horizontal coordinates, and line segment pairs are constructed sequentially. The horizontal coordinate spacing between each pair of tick marks must be greater than the minimum spacing threshold to prevent misidentification of adjacent dashed line segments as valid tick marks due to line jitter. Each pair of candidate line segments is further judged to determine whether its direction difference is within the angle threshold range and whether it has a similar vertical coverage area to ensure that its projection range in the image has structural consistency.

[0135] Furthermore, after selecting the scale line pair, the closed image block between the two scale lines can be extracted based on their boundary positions in the image and used as the target image region for cropping.

[0136] S3.3: Perform iterative operations until the first rotated image, the second rotated image, and the third rotated image are all cropped to obtain the first target image region, the second target image region, and the third target image region;

[0137] Next, we will further elaborate on the pointer positioning part of the method in this application.

[0138] It should be noted that in the image processing flow adopted in this application, although multiple rotated images are generated from the original dial image and corresponding target image regions are cropped, the physical characteristics of the pointer structure determine that it only points to a specific scale position at any given time. Therefore, the pointer frame structure cannot exist simultaneously in all of the cropped image regions. In other words, in any frame of image data, only one or at most two target image regions may contain a complete pointer frame structure, while the remaining image regions do not contain pointers or only contain edge noise information.

[0139] In engineering scenarios, if the pointer's assigned image region is not determined before performing pointer position calculation and scale recognition on the entire target image region, problems such as duplicate recognition, numerical jumps, or misassignment will inevitably arise. Especially when the pointer is near the boundary between two scale segments, it is easy to form an almost complete closed structure in two adjacent image regions simultaneously, making it difficult to determine its true assignment without a judgment mechanism.

[0140] Understandably, this application establishes pointer attribution determination logic before calculating the scale, which is used to identify the valid existence area of ​​the pointer frame structure in the current image and uniquely assign it to a certain target image region. The assigned image region will serve as a coordinate system reference and participate in subsequent processing such as calculating the pixel spacing of the relative pointer positions and linear interpolation of scale values.

[0141] Please refer to Figure 4 , Figure 4 This is a schematic diagram of the pointer positioning process in an embodiment of this application.

[0142] In one example, the specific steps for pointer positioning are as follows:

[0143] S4.1: Perform edge detection on the first target image region, the second target image region, and the third target image region respectively to obtain edge features;

[0144] Specifically, this step extracts structural boundary line segment information from the target image region, providing basic input for recognizing closed pointer box structures. Because the target image region has been rotated and cropped, it possesses characteristics of local orientation alignment and consistent geometric scale, making edge detection highly effective at this stage. Without this processing, subsequent structure recognition will struggle to obtain pointer box regions with closed shapes.

[0145] In one example, edge detection is performed on the first target image region, the second target image region, and the third target image region, respectively, including:

[0146] S4.1.1: The target image region is processed by an edge detection algorithm to obtain a binary edge map;

[0147] S4.1.2: Perform contour tracing on the binary edge map and extract a set of edge segments composed of continuous pixels in the binary edge map, wherein the set of edge segments corresponds to the structural boundary in the binary edge map;

[0148] S4.1.3: Extract parameter features for each edge segment in the set of edge segments. The parameter features include the start and end coordinates, pixel length, direction, edge strength, and connectivity between adjacent edge segments.

[0149] S4.1.4: Calculate edge features based on the parameter features.

[0150] As a preferred embodiment, for each target image region, edge extraction can be performed using a multi-level edge enhancement and combined gradient operator approach. First, the image is subjected to Gaussian filtering for noise reduction to weaken the effects of digital background, uneven grayscale, or low-frequency shadows. Then, a combined Canny algorithm and Sobel thresholding method are used to calculate the gradient magnitude map of the image, obtaining candidate boundaries for potential edge regions. To further enhance line closure, morphological processing of closing operations can be performed on the edge map to connect broken line segments and fill gaps. Finally, all edge information is structured and output as a set of edge pixels.

[0151] S4.2: Based on the extracted edge features, identify the edge features that form closed pointer boxes and determine the image region where the corresponding pointer boxes are located;

[0152] Specifically, this step aims to filter out closed regions that conform to the geometry of the pointer frame from all edge segments. Combining this with the pointer's integrity and spatial envelope in the image coordinate system, the image region to which the pointer frame belongs is determined. This attribution determination is a prerequisite for ensuring the correct calculation of the pointer coordinates; if the region where the pointer frame is located cannot be accurately identified, subsequent scale value calculations will result in significant deviations.

[0153] In this embodiment, parameters such as area, aspect ratio, and approximate rectangle fit are calculated for each edge feature.

[0154] Furthermore, a set of filtering conditions for pointer frame shapes is set, including but not limited to: the outline area is within an empirical range, the aspect ratio of the fitted rectangle is within a reasonable range, the edge grayscale contrast is greater than a set threshold, and the outline centroid is near the vertical midline of the image. Closed outlines that meet this set of conditions are considered candidate pointer frames.

[0155] It is important to note that when multiple candidate pointer boxes exist simultaneously in multiple image regions, the offset distance between the center point of the contour and the center of the image region is further calculated, and the region with the smallest center offset is determined as the image region where the true pointer box is located.

[0156] It should be further pointed out that in practical applications, there may be multiple image regions where closed pointer boxes that conform to the preset shape constraints are identified, and the offset of their outline center point from the geometric center of the image region is similar, making it impossible to directly determine the image region they belong to based on the magnitude of the offset distance.

[0157] Understandably, this situation typically occurs when the pointer is positioned exactly at the boundary of two scale segments, such as around 50 or 100. The pointer structure is repeatedly captured across two image regions and exhibits strong boundary symmetry. In this case, without an additional structure determination strategy, the uniqueness of the pointer frame's affiliation cannot be clearly determined, leading to inaccurate subsequent scale calculations or abrupt changes in its affiliation.

[0158] Therefore, when the above judgment is insufficient to uniquely determine the image attribution, this application further introduces a processing logic based on a spatially synthesized pointer structure to infer the attribution image region. Specifically, the closed pointer frames identified in the two candidate image regions are mapped back to the original dial image coordinate system through their corresponding rotation angles to obtain the geometric positions of the two candidate pointer frames in the same coordinate system. Based on this, a synthesized pointer structure is constructed, which can be the minimum bounding rectangle covering the outlines of the two pointer frames, a boundary-fitted line segment, or a closed polygon region.

[0159] Subsequently, the synthesized pointer structure is projected onto the coordinate systems of all candidate image regions at their respective rotation angles, and the structural representation completeness of the synthesized structure in each image region is calculated. This completeness is jointly evaluated across multiple dimensions, including structural envelope relationship, boundary closure degree, and centroid placement. Finally, the image region where the synthesized structure maintains the strongest closure and largest boundary containment relationship is selected as the image region to which the pointer box belongs.

[0160] In one example, if the image region containing the pointer box is not unique, the method further includes:

[0161] S4.2.1: Based on the rotation angle corresponding to the image region, the geometric structure of the identified pointer frame is inversely transformed from the coordinate system of the corresponding image region to the original image coordinate system;

[0162] Specifically, to achieve a unified comparison of pointer structures across multiple image regions, all candidate pointer frames must first be mapped from their respective image region coordinate systems back to the coordinate system of the initial dial image. Since each image region originates from a local area cropped after rotating the original image at different angles, their coordinate systems differ significantly. Directly comparing the spatial relationships or geometric shapes of the pointer frames will lead to distortion and deviation. Therefore, based on the rotation parameters used during image rotation, an inverse coordinate transformation of the current pointer frame's boundary point set is required; that is, restoring it to the original image coordinate system through affine transformation or a rotation matrix.

[0163] In this embodiment, the inverse transformation method can construct an affine transformation matrix based on a known rotation angle, perform matrix operations on the set of contour boundary points that constitute the pointer frame, and reconstruct them to a unified coordinate system.

[0164] S4.2.2: Perform geometric synthesis of each pointer frame in the original image coordinate system and calculate the pointer geometry;

[0165] Specifically, after completing the inverse transformation, the geometric positions of the pointer bounding boxes in all candidate image regions under the original image coordinate system can be obtained. At this point, these structures can be further geometrically synthesized to construct a unified synthetic pointer structure. The synthetic pointer structure aims to express the common shape of the pointer in each rotated image.

[0166] In this embodiment, the synthesis operation can be performed based on the aggregation processing of boundary point sets, including but not limited to the following methods:

[0167] First, the minimum bounding rectangle method is used to calculate the minimum rectangular region containing the boundaries of all inverse transform pointers, which serves as the boundary of the synthesized structure.

[0168] Secondly, perform contour fitting to connect the connectivity boundaries between multiple point sets and construct a unified closed curve;

[0169] Third, perform the main axis extraction operation based on the boundary point set to obtain the pointer direction vector and reconstruct the synthesized pointer line segment accordingly.

[0170] Furthermore, during the synthesis process, a linear weighting strategy can be introduced based on image structural features to assign higher retention weights to boundary segments with high overlap, thereby suppressing the disturbance effect of misidentified boundaries.

[0171] S4.2.3: Project the pointer geometry onto the image region coordinate system respectively;

[0172] Specifically, after constructing the synthetic structure, in order to evaluate its structural representation ability in different image regions, the structure needs to be remapped to the coordinate system of each candidate image region. The purpose of this step is to deduce the image region to which the structure is most suitable by comparing the overlap of the same physical structure in different image regions.

[0173] S4.2.4: Calculate the degree of boundary overlap and region envelope relationship of the pointer geometry in the corresponding image region coordinate system to obtain the first overlap and the second overlap;

[0174] Specifically, after projection is completed, the consistency between the original pointer structure and the synthesized structure projection in each image region needs to be evaluated. A dual-index joint modeling approach, using boundary overlap and envelope inclusion, is preferred for calculation.

[0175] In this embodiment, the boundary overlap refers to the proportion of the intersection of the contour pixels of the synthesized structure and the original structure in the image region. It can be calculated by rendering the contours of the two structures into binary images and then performing a pixel intersection divided by a union operation. The envelope containment refers to whether the synthesized structure completely encloses the original structure, or whether there is excessive extension and misalignment in the spatial form. The calculation method includes the coverage rate of contour points, the containment relationship of boundary centers, and the proportion of overlapping areas.

[0176] Furthermore, the two indicators mentioned above are calculated separately and then weighted and fused to form a comprehensive structural expression score, which serves as an indicator of the structural expression ability of the image region, namely, the degree of overlap.

[0177] S4.2.5: Based on the first overlap and the second overlap, determine the image region corresponding to the mapping structure with the larger overlap, and use it as the target image region to which the pointer belongs;

[0178] Specifically, after obtaining the synthetic structure representation score for each image region, the image region with higher overlap, more closed structural boundaries, and more reasonable spatial envelope relationship is selected as the final pointer assignment region. In this embodiment, the judgment criterion can be the one with the higher comprehensive score; when the scores are similar, secondary factors such as structural contour connectivity and pointer direction vector stability can be introduced to update the exclusion factor; if a decision still cannot be made, offset median elimination or spatial fuzzy aggregation mechanism can be executed to further compress the fuzzy judgment space.

[0179] It should be further explained that in the processing flow of this application, different rotation angles will crop out multiple image regions, each of which covers only one scale interval. Since the pointer actually spans two intervals, the complete pointer frame structure may be recognized in both image regions. However, physically, there can only be one pointer on the actual dial. Therefore, it is necessary to determine which image region it truly belongs to in order to correctly calculate the subsequent scale values.

[0180] To address this issue, this embodiment introduces a synthetic pointer structure as a logical entity. The synthetic pointer structure refers to mapping the pointer frames identified in each of the two image regions back to the original dial image coordinate system through an inverse rotation transformation. Within this coordinate system, the two overlapping or similar pointer structures are merged to form a more complete and continuous geometric representation. This synthetic structure is essentially an estimation model of the restored real pointer, reproducing as accurately as possible the geometric state of the pointer in its actual physical location.

[0181] Furthermore, after the synthetic structure is constructed, it is mapped back to the coordinate systems of the two candidate image regions to observe where it falls and the degree of overlap in each region. This application argues that the image region where the pointer truly belongs will provide a better interpretation of the synthetic structure; that is, the synthetic structure will fit more closely with the originally identified pointer bounding box in this region, with a higher degree of overlap and more consistent boundaries. Conversely, in non-belonging regions, problems such as shape deviation, boundary misalignment, or incomplete contour closure may occur.

[0182] Ultimately, by determining which image region can more completely accommodate the synthesized pointer structure, this region can be used as the actual target image region where the pointer is located. Combined with the scale line information in this region, the corresponding pointer scale value can be calculated to ensure the accuracy and consistency of recognition.

[0183] S4.3: Extract the edge features of the tick marks in the image region to which the pointer frame belongs;

[0184] Specifically, given that the image region to which the pointer belongs has been determined, this step is used to extract the tick mark features in that region that serve as a reference for the pointer, in order to calculate the relative position. Since there may be multiple tick marks with different lengths and uneven densities in different image regions, it is crucial to extract only the tick marks that form a positional reference relationship with the pointer direction.

[0185] In this embodiment, firstly, all vertical main line segments are identified in the edge map of the image region. A line segment detection algorithm is used to fit a set of main tick lines whose length meets a set threshold, and their horizontal positions and start and end coordinates in the image are recorded. Then, the spacing between all tick lines is calculated, and based on the shortest distance relationship with the pointer frame edge, two tick lines located on the left and right sides of the pointer frame with the closest distance are selected as adjacent reference tick lines, serving as the left reference tick line and the right reference tick line, respectively. These two tick lines will be used to define the pixel-tick value mapping relationship in the image region and serve as the boundary reference for linear interpolation calculations.

[0186] S4.4: Calculate the pixel coordinates of the center position of the pointer frame in the image coordinate system, and determine the relative position of the pointer frame in the corresponding image area based on the pixel coordinates and the pixel spacing between the scale line edge features;

[0187] Specifically, the core objective of this step is to obtain the percentage or offset of the pointer bounding box's position within the current image region, providing input for subsequent tick value regression calculations. The position calculation does not rely on absolute coordinates, but rather on the normalized position of the pointer's center relative to the pixel distances between its two tick marks; therefore, this position is scale-adaptive and orientation-robust.

[0188] In this embodiment, based on the pointer frame outline structure extracted in S4.2, its minimum bounding rectangle is calculated, and the center point of the rectangle is taken as the center position of the pointer frame. The absolute coordinate value of this point in the image coordinate system is obtained and recorded as a reference pixel. Subsequently, the pixel distance between this reference point and the horizontal coordinates of the left and right reference tick lines is compared, and the normalized distance value between the two tick lines, i.e., the relative offset, is calculated. The normalized distance value is usually in the range of 0–1.

[0189] Furthermore, a consistency constraint can be added on the angle between the scale line and the pointer direction angle to verify whether the current calculation direction is consistent with the dial structure logic. If they are inconsistent, the rotation angle can be reversed for correction.

[0190] In one example, the specific steps of S5 are as follows:

[0191] S5.1: In the image area to which the pointer frame belongs, determine the actual scale values ​​of the first and second scale lines adjacent to the pointer frame;

[0192] S5.2: Calculate the scale value corresponding to the pointer frame by linear interpolation based on the relative position and the actual scale value.

[0193] In one example, this application provides a reading recognition system for a square pointer meter, the system comprising:

[0194] The image acquisition module is used to acquire the dial image of the target square pointer watch;

[0195] The image processing module is used to rotate the dial image from multiple angles and crop the rotated image to obtain a target image area including a scale frame and a number frame.

[0196] The edge analysis module is used to perform edge detection on the target image region, identify the edge features that form a closed pointer box, and determine the relative position of the pointer box with respect to the scale line.

[0197] The scale calculation module is used to calculate the scale value indicated by the pointer frame based on the actual scale difference between the relative position and the scale line.

[0198] The edge analysis module includes:

[0199] The edge detection unit is used to perform edge detection on the target image region and generate the corresponding binary edge map.

[0200] The contour extraction unit is used to perform contour tracking operation on the binary edge map and extract a set of edge line segments composed of continuous pixels.

[0201] An edge parameter calculation unit is used to extract parameter features, including start and end coordinates, pixel length, direction, edge strength, and connectivity, from the edge segments in the edge segment set.

[0202] The pointer recognition unit is used to identify the pointer frame that constitutes a closed contour structure based on the parameter characteristics of the edge line segment, and to determine the image area where the pointer frame is located.

[0203] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for recognizing the readings of a square pointer meter, characterized in that, The method includes: Obtain the dial image of the target square pointer watch; The dial image is rotated at multiple angles to generate rotated images corresponding to different scale intervals; The rotated image is cropped to obtain a target image region including a scale frame and a number frame; The target image region is processed sequentially by an edge detection algorithm to determine the relative position of the target image region to which the pointer belongs and the pointer frame corresponding to the pointer relative to the tick line. Based on the image region to which the pointer belongs and the relative position of the pointer frame relative to the scale line, the pixel distribution of the pointer frame is calculated to obtain the scale value indicated by the pointer; The cropping of the rotated image to obtain an image region including a scale frame and a number frame includes: Traverse the rotated image to identify multiple scale lines whose length is greater than a preset length threshold; The scale lines are filtered one by one according to the preset judgment conditions to obtain the scale line pairs that meet the judgment conditions. The image area between the scale line pairs is used as the target image area for cropping. The judgment conditions include: the line segments of the two scale lines constituting the scale line pair both satisfy the vertical structure judgment condition, and the positional relationship between the two scale lines satisfies the horizontal alignment requirement.

2. The method for recognizing the readings of a square pointer meter according to claim 1, characterized in that, The dial image is rotated at multiple angles to generate rotated images corresponding to different scale intervals, including: The dial image is rotated at a preset first angle to generate a first rotated image; The dial image is rotated at a preset second angle to generate a second rotated image; The dial image is rotated at a preset third angle to generate a third rotated image.

3. The method for recognizing the readings of a square pointer meter according to claim 2, characterized in that, The step of cropping the rotated image to obtain an image region including a scale frame and a number frame also includes: Perform iterative operations until the first rotated image, the second rotated image, and the third rotated image are all cropped, resulting in the first target image region, the second target image region, and the third target image region.

4. The method for recognizing the readings of a square pointer meter according to claim 3, characterized in that, The step of processing the target image region sequentially using an edge detection algorithm to determine the relative position of the target image region to which the pointer belongs and the pointer frame corresponding to the pointer relative to the tick marks includes: Edge detection is performed on the first target image region, the second target image region, and the third target image region respectively to obtain edge features; Based on the extracted edge features, identify the edge features that form closed pointer boxes and determine the image region where the corresponding pointer boxes are located; Extract the edge features of the tick marks in the image region to which the pointer bounding box belongs; Calculate the pixel coordinates of the center position of the pointer frame in the image coordinate system, and determine the relative position of the pointer frame in the corresponding image region based on the pixel coordinates and the pixel spacing between the tick mark edge features.

5. The method for recognizing the readings of a square pointer meter according to claim 4, characterized in that, Edge detection is performed on the first target image region, the second target image region, and the third target image region, respectively, including: The target image region is processed by an edge detection algorithm to obtain a binary edge map; Contour tracing is performed on the binary edge map to extract a set of edge segments composed of continuous pixels in the binary edge map. The set of edge segments corresponds to the structural boundary in the binary edge map. For each edge segment in the set of edge segments, parameter features are extracted. The parameter features include the start and end coordinates, pixel length, direction, edge intensity, and connectivity between adjacent edge segments. Calculate edge features based on the parameter characteristics.

6. The method for recognizing the readings of a square pointer meter according to claim 4, characterized in that, Calculate the pixel distribution of the pointer frame to obtain the scale value indicated by the pointer, including: Within the image area to which the pointer frame belongs, determine the actual scale values ​​of the first and second scale lines adjacent to the pointer frame; Based on the relative position and the actual scale value, the scale value corresponding to the pointer frame is calculated by linear interpolation.

7. The method for recognizing the readings of a square pointer meter according to claim 4, characterized in that, If the image region where the pointer box is located is not unique, the method further includes: Based on the rotation angle corresponding to the image region, the geometric structure of the identified pointer frame is inversely transformed from the coordinate system of the corresponding image region to the original image coordinate system; In the original image coordinate system, the pointer frames are geometrically synthesized, and the pointer geometry is calculated. The pointer geometry is projected onto the image region coordinate system respectively; Calculate the degree of boundary overlap and region envelope relationship of the pointer geometry in the corresponding image region coordinate system to obtain the first overlap and the second overlap. The first overlap and the second overlap are used to determine the image region corresponding to the mapping structure with the larger overlap, which is then used as the target image region to which the pointer belongs.

8. A reading recognition system for a square pointer meter, used to implement the reading recognition method for a square pointer meter as described in any one of claims 1-7, characterized in that, The system includes: The image acquisition module is used to acquire the dial image of the target square pointer watch; The image processing module is used to rotate the dial image from multiple angles and crop the rotated image to obtain a target image area including a scale frame and a number frame. The edge analysis module is used to perform edge detection on the target image region, identify the edge features that form a closed pointer box, and determine the relative position of the pointer box with respect to the scale line. The scale calculation module is used to calculate the scale value indicated by the pointer frame based on the actual scale difference between the relative position and the scale line.

9. The reading recognition system for a square pointer meter according to claim 8, characterized in that, The edge analysis module includes: The edge detection unit is used to perform edge detection on the target image region and generate the corresponding binary edge map. The contour extraction unit is used to perform contour tracking operation on the binary edge map and extract a set of edge line segments composed of continuous pixels. An edge parameter calculation unit is used to extract parameter features, including start and end coordinates, pixel length, direction, edge strength, and connectivity, from the edge segments in the edge segment set. The pointer recognition unit is used to identify the pointer frame that constitutes a closed contour structure based on the parameter characteristics of the edge line segment, and to determine the image area where the pointer frame is located.

Images