Character wheel-pointer type intelligent camera shooting water meter high-precision reading method and system

Abstract

The invention relates to the technical field of water meters, in particular to a character wheel-pointer type intelligent camera shooting water meter high-precision reading method and system. Comprising the following steps: positioning a character wheel area in a water meter image by adopting a rotating target detection model, and automatically correcting dial plate deflection by analyzing a deflection angle and geometrical characteristics of the character wheel area; and extracting a pointer and a pointer end region based on the corrected image, and calculating and obtaining a pointer reading by utilizing a long side slope of a pointer end bounding box. And finally, based on the pointer reading and the character wheel region, constructing a convolutional neural network classification model containing a carry state, and realizing digital reading by combining a dual-threshold carry correction algorithm, thereby realizing composite water meter reading. According to the method provided by the invention, high-precision reading can be realized under the influence of any deflection angle, indication error and mechanical carry precision difference, the processing speed reaches 74 milliseconds / frame, and the average pointer angle identification error is lt; and the method is particularly suitable for water meter automatic reading scenes in complex environments.

Description

Technical Field

[0001] This application relates to the field of water meter technology, and in particular to a high-precision reading method and system for a pointer-type intelligent camera water meter. Background Technology

[0002] Smart infrastructure is crucial for realizing smart cities, and water meters, as key instruments for water resource measurement and management, are essential for accurate billing, water usage analysis, and leakage control in smart water systems. However, traditional manual meter reading methods not only consume significant manpower and resources but are also costly, inefficient, and prone to inaccuracies and data verification issues. Therefore, the technology of automating water meter reading by installing camera-equipped cameras to achieve low-cost deployment and high-precision identification is considered the core of smart water management development.

[0003] Existing vision-based pointer-type water meter reading technologies include digital reading methods using convolutional neural networks and template matching, and pointer reading methods based on Hough transform and deep learning for keypoint detection. For example, in the existing invention patent with publication number CN111783785B, a text localization network is used to locate the numerical region in an image, followed by the extraction of image features using a convolutional neural network, and then combined with a recurrent neural network and other classification modules to achieve digital readings within the pointer region of the water meter. In the existing invention patent with publication number CN118918573A, this method uses Hough transform to detect the pointer line in the meter image and uses the pointer axis and pointer bounding box to accurately locate the pointer position and achieve pointer readings.

[0004] In practical applications, camera-based water meters often suffer from image distortion and flipping issues due to installation environment limitations. This makes it difficult for traditional text detection or corner matching methods to accurately extract the digit wheel area. Furthermore, affected by image resolution and indication errors, the reading of the higher-order pointer often relies on the lower-order pointer for correction. Relying solely on line detection or digital positioning methods cannot guarantee reading accuracy in low-quality images. Simultaneously, mechanical carry effects cause digits to exhibit unique transition states, and the differences in carry accuracy among different water meters further increase the difficulty of digit recognition, reducing the overall recognition accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide a high-precision digital wheel-pointer type water meter intelligent reading system to address the problems of inaccurate intelligent readings caused by arbitrary deflection angles, low resolution, indication errors, and differences in water meter carry-over accuracy.

[0006] The technical solution of this invention is as follows: a high-precision reading system for a pointer-type intelligent camera water meter, the system comprising a water meter image acquisition module, an image correction module, a pointer recognition module, a digital reading module, and a combined reading module; a first end of the water meter image acquisition module is connected to a first end of the image correction module; a second end of the image correction module is connected to a first end of the pointer recognition module, and a third end of the image correction module is connected to a first end of the digital reading module; a second end of the pointer recognition module is connected to a second end of the digital reading module, and a third end of the pointer recognition module is connected to a first end of the combined reading module; a third end of the digital reading module is connected to a second end of the combined reading module. The water meter image acquisition module is used to acquire images of a pointer-type water meter; The image correction module is used to obtain the correction angle of the image of the dial-pointer water meter, and to obtain the corrected image based on the correction angle; The pointer recognition module acquires the pointer reading of a digit-pointer type water meter from a corrected image; The digital reading module acquires the digital reading in a digit wheel-pointer water meter based on the corrected image and pointer reading. The combined reading module is used to obtain the reading result of the pointer-pointer water meter based on the pointer reading and the digital reading.

[0007] Furthermore, the image correction module includes a module for training a rotating target detection model to extract information about the character wheel region, judging the image flip state, and image correction.

[0008] Furthermore, the pointer recognition module includes pointer region extraction and deflection, pointer tip region in pointer image, pointer angle calculation, and pointer reading process.

[0009] Furthermore, the digital reading module includes the number of digits in the character wheel and the character wheel image, digital image segmentation, segmented image standardization and grayscale conversion, dual-threshold last digit recognition, and digital reading process.

[0010] The present invention provides a high-precision reading method for a dial-pointer type intelligent camera water meter, comprising the following steps: The water meter image acquisition module acquires an image of the dial-pointer water meter; the image correction module acquires the correction angle of the dial-pointer water meter image and acquires a corrected image based on the correction angle; the pointer recognition module acquires the pointer reading of the dial-pointer water meter on the corrected image; the digital reading module acquires the digital reading of the dial-pointer water meter based on the corrected image and the pointer reading; and the combined reading module acquires the reading result of the dial-pointer water meter based on the pointer reading and the digital reading.

[0011] Furthermore, the image correction module includes a module for training a rotating target detection model to extract information about the character wheel region, judging the image flipping state, and an image correction module; The word wheel region information extracted by the trained rotating target detection model includes the center point of the word wheel region. The width and height of the type wheel area are respectively , Type wheel rotation angle ,in The coordinates of the four vertices are ; The image flip state The formula for judgment is as follows: In the formula, For logical "OR", that is, when , or If one or more are true, then If true; if all three are false, then It is false. Among them, , and The calculation formula is as follows: In the formula, , For the image deflection angle, and These represent the width and height of the pointer-type water meter image acquired by the water meter image acquisition module, respectively. For weighting coefficients. When The image is in a flipped state when it is in a flipped state; otherwise, the image is not in a flipped state. Furthermore, the correction angle for the digit-pointer water meter image in the image correction module includes the deflection angle of the digit wheel. and the image flip state If the image is rotated to the right... To correct the angle, the calculation formula is as follows: In the formula, The function rounds a number to the nearest integer. (Correction angle) .

[0012] Furthermore, the pointer recognition module includes pointer region extraction and deflection, pointer tip region in pointer image, pointer angle calculation, and pointer reading process; Pointer region extraction and deflection involves training a target detection model to extract all pointer region images based on the corrected image obtained by the image correction module. The number of pointer region images in the water meter image is... The pointer region image is located at the center coordinates of the water meter image. And the boundary point coordinate information, then, the pointer area image is rotated to the right by a fixed angle; The pointer tip region in the pointer image is extracted by training a rotated target detection model based on the rotated pointer image. The pointer angle calculation is based on the pointer tip region and the rotated pointer image to calculate the pointer rotation angle. Taking a 90-degree right rotation during pointer region extraction and deflection as an example, the pointer rotation angle is... The calculation formula is as follows: In the formula, , It is the average value of the boundary points closest to the center of the pointer image in the pointer tip region. The average value of boundary points in the pointer tip region far from the center of the pointer image. Single pointer reading. and pointer rotation angle The relationship between them is as follows: The pointer reading process is based on the horizontal coordinate obtained during pointer region extraction and deflection. Sorted from left to right, the reading of the least significant bit pointer is: In the formula, This indicates rounding down. Indicates rounding up. For the first The readings of the first pointer, followed by the readings of the remaining pointers: In the formula, The final pointer read vector is represented as Its value is equal to .

[0013] Furthermore, the digital reading module includes the number of digits in the character wheel and the character wheel image, digital image segmentation, segmentation image standardization and grayscale conversion, dual-threshold last digit recognition, and digital reading process; The number of digits in the character wheel and the number of digits in the character wheel image. This refers to the label information for the character wheel region in the image correction module. The character wheel image is obtained through... Corrected type wheel area; The digital image segmentation is based on the number of digits. and the width of the character wheel area Divide the parts proportionally; The segmented image standardization and grayscale conversion involves converting the segmented single digital color image to grayscale and standardizing the image size. The dual-threshold last-digit recognition identifies standardized and grayscale single digits. It first expands the original digit classification range to 20 categories, including half-character carry states. Its CNN classification models for 0-9 and 00-19 are as follows: In the formula, For digital images, and The prediction results are for the 0-9 and 00-19 classification models, respectively. (Through...) Predict the category of the last digit, when predicting the category When it is a non-odd number and the pointer reading Less than the threshold At that time, the reading result for When it exceeds the threshold The reading result is Similarly, when predicting categories When the number is odd, then consider Rounding up or down, pointer reading Less than the threshold Round up, otherwise round down; however, this process... ,as well as When rounding up, an overflow may occur, requiring correction of both the current and previous digits. Specifically, when... When it is -1, it is automatically corrected to 9. When the value is 10, it is automatically corrected to 0, and a carry flag is set. ; The digit reading process involves identifying the remaining digits sequentially from the least significant digit to the most significant digit, based on the recognition of the last digit. First, the digits are divided into carry-over and non-carry-over states. When the previous digit is... Prediction Category If the previous digit's final identification result is 0 or 9, then the current digit is determined to be in a carry state; there are two cases for the carry state: Case 1: If the current digit When in a carry state and the lower bit is 0; use predict ,like It is an odd number, and the digits are read as follows. Otherwise, the digital reading If corrected If the value is 10, then force it to be set to 0 and set the flag. ; Case 2: If the current digit When in a carry-over state and the lowest digit is 9; use predict ,like It is an odd number, and the digits are read as follows. Otherwise, the digital reading If corrected If the value is -1, then it is forced to be set to 9, and the flag is set. ; Otherwise, the digital reading is as follows: Predict the data and clear the carry flag; the final digital reading result is calculated as follows: Furthermore, the joint reading module is used to read the pointer value. With digital reading Obtain the readings of a pointer-type water meter.

[0014] Compared with the prior art, the present invention has the following advantages: This invention systematically solves the three major bottleneck problems of traditional water meter identification technology through multi-model fusion and dynamic correction mechanisms. In terms of geometric correction, it innovatively combines rotating target detection and geometric relationship analysis, overcoming the limitations of traditional corner detection methods in correcting large-angle deflections and image flipping, achieving adaptive standardized correction at any installation angle. Addressing the challenge of low-resolution pointer recognition, it proposes a "nearest-farthest edge" dual-feature point modeling and long-side slope calculation method, effectively overcoming low resolution and mechanical indication errors through a recursive correction mechanism from low to high digits. Addressing the industry challenge of digital carry-over in mechanical water meters, it utilizes a dual-threshold linkage correction method, explicitly distinguishing half-character states through a 20-category digit classification model, and dynamically triggering carry-over compensation based on pointer readings. It also designs a carry-over flag transmission and overflow handling mechanism to resolve ambiguity in the last digit and cumulative deviation from continuous carry-overs. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the reading system of a pointer-type water meter in one embodiment; Figure 2 In the figure, (a) is a diagram of a digit wheel-pointer water meter under deflection and flipping states in one embodiment, (b) is a diagram of a digit wheel-pointer water meter under indication error in one embodiment, and (c) is a diagram of a digit wheel-pointer water meter under mechanical carry accuracy difference in one embodiment; Figure 3This is a schematic diagram of the structure of a specific dial-pointer type water meter reading system in one embodiment; Figure 4 In the image, (a) is a pointer area diagram of a dial-pointer water meter in one embodiment, and (b) is a pointer image after deflection and the pointer tip area in the pointer image in one embodiment; Figure 5 This is a diagram illustrating the process of identifying the last digit of a double-threshold digit in one embodiment; Figure 6 In the figures, (a) is the image of the pointer-type water meter to be detected in Example 1; (b) is the image of the pointer-type water meter after extracting the pointer region information in Example 1; (c) is the image of the pointer-type water meter after deflection correction in Example 1; (d) is the image of the pointer region extracted after correction in Example 1; (e) is the image of the pointer region recognition result in Example 1; and (f) is the image of the pointer after deflection and the image of the pointer tip region recognition result in Example 1. Figure 7 In the diagram, (a) shows a clear dial-pointer water meter diagram in Example 2, and (b) shows a dial-pointer water meter diagram with external interference in Example 2. Figure 8 In the diagram, (a) is the water meter chart obtained on day 1 of Example 3, (b) is the water meter chart obtained on day 31 of Example 3, and (c) is the water meter chart obtained on day 61 of Example 3. Figure 9 In the figure, (a) is a graph showing the continuous reading results of the wheel-pointer water meter in Example 3, and (b) is a graph showing the error distribution of the pointer reading in Example 3. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0017] Example 1 like Figure 1 As shown, this embodiment provides a high-precision reading method and system for a pointer-type intelligent camera water meter. It includes a water meter image acquisition module, an image correction module, a pointer recognition module, a digital reading module, and a combined reading module. The first end of the water meter image acquisition module is connected to the first end of the image correction module; the second end of the water meter image acquisition module is connected to the first end of the pointer recognition module, and the third end is connected to the first end of the digital reading module; the second end of the pointer recognition module is connected to the second end of the digital reading module, and the third end is connected to the first end of the combined reading module; the third end of the digital reading module is connected to the second end of the combined reading module.

[0018] The water meter image acquisition module is used to acquire images of the dial-pointer water meter; the captured images of the dial-pointer water meter in deflection and flipping states are shown below. Figure 2 As shown in (a); the diagram of the digit wheel-pointer type water meter with indication error is as follows. Figure 2 As shown in (b), the figure illustrates that directly reading pointer A is ambiguous; the higher pointer must be corrected by combining the reading of the lower pointer. The mechanical carry accuracy difference in the digit wheel-pointer type water meter diagram is shown below. Figure 2 As shown in (c), the last digit of the two water meter images is in a non-carrying state at the same time, but the pointers are in different positions. Since the first pointer B0 is between 9 and 0, the last digit B needs to be corrected by subtracting 1.

[0019] The image correction module is used to obtain the correction angle of the image of the dial-point water meter and obtain a corrected image based on the correction angle. The pointer recognition module obtains the pointer reading of the dial-point water meter from the corrected image. The digital reading module obtains the digital reading of the dial-point water meter based on the corrected image and the pointer reading.

[0020] Specifically, the structural schematic diagram of the specific dial-pointer water meter reading system in this embodiment is as follows: Figure 3 As shown, the image correction module includes a module for training a rotating target detection model to extract character wheel region information, an image flipping state determination module, and an image correction module. The character wheel region information extracted by the training rotating target detection model is as follows: In the formula, Indicates the center point of the character wheel area; and Indicates the width and height of the character wheel area; It is the rotation angle of the type wheel, among which . This represents the coordinates of four vertices arranged in clockwise or counterclockwise order.

[0021] Furthermore, image flip state The formula for judgment is as follows: In the formula, For logical "OR", that is, when , or If one or more are true, then If true; if all three are false, then It is false. Among them, , and The calculation formula is as follows: In the formula, , For the image deflection angle, and These represent the width and height of the pointer-type water meter image acquired by the water meter image acquisition module, respectively. These are the weighting coefficients. The larger the value, the stricter the criteria for determining the flip condition. When The image is in a flipped state if it is flipped, otherwise the image is not flipped.

[0022] Furthermore, the correction angle for the digit-pointer water meter image in the image correction module includes the deflection angle of the digit wheel. and the image flip state Now, let's look at the image rotated to the right. For example, the formula is calculated as follows: In the formula, The function rounds a number to the nearest integer. (Correction angle) Given the right turn angle Image correction is then performed.

[0023] Specifically, the structural schematic diagram of the specific dial-pointer water meter reading system in this embodiment is as follows: Figure 3 As shown, the pointer recognition module includes pointer region extraction and deflection, pointer tip region in the pointer image, pointer angle calculation, and pointer reading process. The pointer recognition module includes pointer region extraction and deflection, which is based on the corrected image obtained by the image correction module, training a target detection model to extract all pointer region images. The number of pointer region images in the water meter image... The pointer region image is located at the center coordinates of the water meter image. And boundary point coordinate information, such as Figure 4 As shown in (a). Then, the pointer area image is rotated to the right by a fixed angle.

[0024] The pointer tip region in the pointer image is extracted by training a rotated target detection model based on the rotated pointer image. The pointer tip region is as follows: Figure 4 As shown in (b).

[0025] Furthermore, the pointer angle calculation is based on the pointer tip region and the rotated pointer image to calculate the pointer rotation angle. Taking a 90-degree right rotation during pointer region extraction and deflection as an example, the pointer rotation angle... The calculation formula is as follows: In the formula, . It is the average value of the boundary points closest to the center of the pointer image in the pointer tip region. The average value of the boundary points in the pointer tip region that are far from the center of the pointer image.

[0026] Subsequently, single-pointer reading and pointer rotation angle The relationship between them is as follows: Furthermore, the pointer reading process is based on the horizontal coordinate obtained during pointer region extraction and deflection. Sorted from left to right, the reading of the least significant bit pointer is: In the formula, This indicates rounding down. Indicates rounding up. For the first The readings of the first pointer, followed by the readings of the remaining pointers: In the formula, The final pointer read vector is represented as follows: Its value is equal to .

[0027] Specifically, the structural schematic diagram of the specific dial-pointer water meter reading system in this embodiment is as follows: Figure 3 As shown, the digital reading module includes counting the number of digits in the character wheel and the character wheel image, digital image segmentation, normalization and grayscale conversion of the segmented image, dual-threshold last digit recognition, and the digital reading process. The number of digits in the character wheel and the number of digits in the character wheel image are also described. This refers to the label information for the character wheel region in the image correction module. The character wheel image is obtained through... The corrected character wheel area. The digital image segmentation is based on the number of digits... and the width of the character wheel area The images are divided proportionally. The standardization and grayscale conversion of the segmented images involves converting the segmented individual digital color images to grayscale and unifying the image size.

[0028] Furthermore, the dual-threshold last digit recognition identifies a standardized and grayscale single digit, the process of which is as follows: Figure 5 As shown. First, the original digit classification range was expanded to 20 categories, including half-character carry states. The CNN classification models for 0-9 and 00-19 are as follows: In the formula, For digital images, and These are the prediction results for the 0-9 and 00-19 classification models, respectively. (Carry indicator) Initialize to 0, through Predict the category of the last digit, when predicting the category When it is a non-odd number and the pointer reading Less than the threshold At that time, the reading result for When it exceeds the threshold The reading result is Similarly, when predicting categories... When the number is odd, then consider Rounding up or down, pointer reading Less than the threshold Round up, otherwise round down. However, this process... ,as well as When rounding up, an overflow may occur, requiring correction of both the current and previous digits. Specifically, when... When it is -1, it is automatically corrected to 9. When the value is 10, it is automatically corrected to 0, and a carry flag is set. .

[0029] Furthermore, the digit reading process involves identifying the remaining digits sequentially from the least significant digit to the most significant digit, based on the recognition of the last digit. First, the digits are divided into carry-over and non-carry-over states; when the previous digit is... Prediction Category If the previous digit's final identification result is 0 or 9, then the current digit is determined to be in a carry state; there are two cases for the carry state. Case 1: If the current digit When in a carry state and the lower bit is 0; use predict ,like It is an odd number, and the digits are read as follows. Otherwise, the digital reading If corrected If the value is 10, then force it to be set to 0 and set the flag. .

[0030] Case 2: If the current digit When in a carry-over state and the lowest digit is 9; use predict ,like It is an odd number, and the digits are read as follows. Otherwise, the digital reading If corrected If the value is -1, then it is forced to be set to 9, and the flag is set. .

[0031] Otherwise, the digital reading is as follows: Predict the data and clear the carry flag. The final digital reading is calculated as follows: Furthermore, the joint reading module is used to read the pointer value. With digital reading Obtain the reading result of the pointer-type water meter. .

[0032] by Figure 6 (a) The image shows a pointer-type water meter as the case to be detected. The results of training the rotating target detection model to extract information from the pointer region are as follows: Figure 6 As shown in (b), where Figure 6 (b) In this context, "Pointer" represents the pointer region; the deflection angles of the two images of the test case are shown. The center point of the character wheel area Image flip state Image right rotation angle As shown in Table 1. Image deflection correction dial-point water meter, as shown... Figure 6 As shown in (c); Figure 6 (e) is a diagram showing the results of pointer region recognition; Figure 6 (f) shows the pointer image after deflection and the recognition result of the pointer tip area. The pointer readings and the final reading results are shown in Table 1.

[0033] Table 1 For example, the water meter image acquisition module in this embodiment can be any camera-based image transmission system (such as the ESP32 vision module, RER-USB48MP02, WonderMV vision module, etc.). This module can capture, store, and upload images of the water meter to the cloud, including the image of the dial-pointer type water meter described in Embodiment 1. The rotating target detection model in this embodiment can be any rotation-based target detection algorithm (such as YOLOv8 OBB, YOLOv11 OBB, etc.). This model is used to extract information about the dial, pointer, and pointer tip area, including the image correction module and pointer recognition module in Embodiment 1.

[0034] This embodiment provides a reading method and system for a dial-pointer water meter based on dual-threshold carry correction. It employs a rotating target detection model to locate the dial region in the water meter image and automatically corrects dial deflection by analyzing the deflection angle and geometric features of the dial region. Based on the corrected image, the pointer and needle tip regions are extracted, and the pointer reading is calculated using the slope of the long side of the needle tip bounding box. Finally, based on the pointer reading and the dial region, a convolutional neural network classification model incorporating carry states is constructed, and combined with a dual-threshold carry correction algorithm to achieve digital reading. This enables a composite water meter reading, effectively overcoming the influence of arbitrary deflection angles, indication errors, and differences in mechanical carry accuracy on the water meter reading.

[0035] Example 2 To better verify the reading method and system of the pointer-type water meter based on dual threshold carry correction in Example 1, the following explanation is provided in conjunction with specific data and publicly available data.

[0036] The specific process is as follows: The programming environment for this embodiment is Python 3.9. Model training was performed using the PyTorch deep learning framework. Experimental configuration details are shown in Table 2.

[0037] Table 2 In terms of model reading ability, accuracy (Acc) and pointer recognition accuracy are used. ) and digit recognition accuracy ( The accuracy of the model is evaluated using the following formula: In the formula, This is the total number of test dial images; and Representing the first Predicted readings and actual ground readings for each sample; This is a conditional function; it returns 1 if the condition within the parentheses is true, and 0 otherwise. Pointer recognition accuracy. and digit recognition accuracy The calculation formula is as follows: In the formula, and It is the first The total count of pointers and numbers in each test sample. and They represent the first The number of pointer read errors and the number of number read errors in each test sample.

[0038] The system's performance was empirically evaluated based on a constructed dataset of 1500 test samples. The clear and noisy data were evenly distributed. Figure 7 (a) A clear diagram of a dial-pointer water meter; Figure 7 (b) is a diagram of a dial-pointer water meter with external interference.

[0039] The proposed reading method and system for a pointer-type water meter based on dual-threshold carry correction achieved an accuracy of 96.93% on clear data and maintained an accuracy of 91.07% on interfering data, demonstrating robust performance with an overall accuracy of 94.0%. This system also excels in pointer recognition (…). : 98.45%) and digital recognition ( It exhibits exceptional performance in terms of accuracy (99.40%), with only a slight decrease in performance (0.63-1.99 percentage points) under interference conditions, while maintaining a detection time of approximately 74.29 ms per sample, as shown in the digital pointer reading results in Table 3.

[0040] Table 3 In the publicly available WMeter5K database, the first 800 data points were selected sequentially for verification, and the water meter readings are shown in Table 4.

[0041] Table 4 The accuracy was low (84.50%) when directly using the existing model for validation. This was due to the unrecognized pointers on some dials. After expanding the data by 10%, the accuracy improved to 92.88%. The accuracy reached 98.35%, an average improvement of 7.33% compared to existing methods. Among the erroneous data, 40.3% contained errors in the last pointer digit. Therefore, when the error was limited to the last pointer digit, and with difficult data accounting for 10%, the accuracy reached 95.75%. The results from both sets of data demonstrate that the proposed dual-threshold carry correction-based reading method and system for pointer-type water meters can effectively achieve high accuracy readings in complex environments.

[0042] Example 3 This embodiment uses continuous monitoring data from a water meter as an application scenario to further verify the effectiveness and practicality of this application. The water meter data obtained in this embodiment is as follows: Figure 8 As shown, where Figure 8 (a) A graph of a dial-pointer water meter obtained on day 1; Figure 8 (b) A graph of a dial-pointer water meter obtained on day 31; Figure 8(c) is the digit wheel-pointer water meter chart obtained on day 61.

[0043] During the 100-day continuous monitoring period, the continuous reading results of the dial-pointer water meter were as follows: Figure 9 As shown in (a), the cumulative water consumption shows a continuously increasing trend with no reading errors. To verify the accuracy of the proposed pointer recognition module, we compared the reading of the first pointer of the meter with the actual pointer reading (e.g., Figure 8 (a) The actual pointer reading is 1.955, and the estimated first pointer reading is 1.9928. Figure 9 (b) shows the distribution of absolute errors between the estimated and true values. During the 100-day continuous monitoring period, the frequency decreased with increasing error magnitude. The average absolute measurement error of the pointer reading was 0.117 (equivalent to a volume error of 0.0000117 m³), ​​corresponding to an angle estimation error of 4.21°. This result strongly demonstrates the reliability of the method for pointer readings under harsh low-resolution (41×41px) conditions.

[0044] The water meter image acquisition module is implemented through hardware (camera and image transmission), while the image correction module, pointer recognition module, digital reading module, and joint reading module are implemented through software code that executes the corresponding algorithms.

[0045] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0046] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A high-precision reading system for a pointer-type intelligent camera water meter, characterized in that, The system includes a water meter image acquisition module, an image correction module, a pointer recognition module, a digital reading module, and a combined reading module. The first end of the water meter image acquisition module is connected to the first end of the image correction module; the second end of the image correction module is connected to the first end of the pointer recognition module, and the third end of the image correction module is connected to the first end of the digital reading module; the second end of the pointer recognition module is connected to the second end of the digital reading module, and the third end of the pointer recognition module is connected to the first end of the combined reading module; the third end of the digital reading module is connected to the second end of the combined reading module. The water meter image acquisition module is used to acquire images of a pointer-type water meter; The image correction module is used to obtain the correction angle of the image of the dial-pointer water meter, and to obtain the corrected image based on the correction angle; The pointer recognition module acquires the pointer reading of a digit-pointer type water meter from a corrected image; The digital reading module acquires the digital reading in a digit wheel-pointer water meter based on the corrected image and pointer reading. The combined reading module is used to obtain the reading result of the pointer-pointer water meter based on the pointer reading and the digital reading.

2. The high-precision reading system for a pointer-type intelligent camera water meter according to claim 1, characterized in that, The image correction module includes a module for training a rotating target detection model to extract information about the character wheel region, judging the image flipping state, and image correction.

3. The high-precision reading system for a pointer-type intelligent camera water meter according to claim 1, characterized in that, The pointer recognition module includes pointer region extraction and deflection, pointer tip region in pointer image, pointer angle calculation, and pointer reading process.

4. The high-precision reading system for a pointer-type intelligent camera water meter according to claim 1, characterized in that, The digital reading module includes the number of digits in the character wheel and the character wheel image, digital image segmentation, segmentation image standardization and grayscale conversion, dual-threshold last digit recognition, and digital reading process.

5. A high-precision reading method for a dial-pointer type intelligent camera water meter, characterized in that, Includes the following steps: The water meter image acquisition module acquires an image of the dial-pointer water meter; the image correction module acquires the correction angle of the dial-pointer water meter image and acquires a corrected image based on the correction angle; the pointer recognition module acquires the pointer reading of the dial-pointer water meter on the corrected image; the digital reading module acquires the digital reading of the dial-pointer water meter based on the corrected image and the pointer reading; and the combined reading module acquires the reading result of the dial-pointer water meter based on the pointer reading and the digital reading.

6. The high-precision reading method for a pointer-type intelligent camera water meter according to claim 5, characterized in that, The image correction module includes a training rotating target detection model to extract word wheel region information, an image flipping state judgment, and an image correction module. The word wheel region information extracted by the trained rotating target detection model includes the center point of the word wheel region. The width and height of the type wheel area are respectively , Type wheel rotation angle ,in The coordinates of the four vertices are ; The image flip state The formula for judgment is as follows: In the formula, For logical "OR", that is, when , or If one or more are true, then If true; if all three are false, then It is false, among which, , and The calculation formula is as follows: In the formula, , For the image deflection angle, and These represent the width and height of the pointer-type water meter image acquired by the water meter image acquisition module, respectively. For weighting coefficients, when The image is in a flipped state if it is flipped, otherwise the image is not flipped.

7. A high-precision reading method for a pointer-type intelligent camera water meter according to claim 5, characterized in that, The correction angle for the digit-pointer water meter image in the image correction module includes the deflection angle of the digit wheel. and the image flip state If the image is rotated to the right... To correct the angle, the calculation formula is as follows: In the formula, The function rounds a number to the nearest integer and corrects for the angle. .

8. A high-precision reading method for a pointer-type intelligent camera water meter according to claim 5, characterized in that, The pointer recognition module includes pointer region extraction and deflection, pointer tip region in pointer image, pointer angle calculation, and pointer reading process; Pointer region extraction and deflection involves training a target detection model to extract all pointer region images based on the corrected image obtained by the image correction module. The number of pointer region images in the water meter image is... The pointer region image is located at the center coordinates of the water meter image. And the boundary point coordinate information, then the pointer area image is rotated to the right by a fixed angle; The pointer tip region in the pointer image is extracted by training a rotated target detection model based on the rotated pointer image. The pointer angle calculation is based on the pointer tip region and the rotated pointer image to calculate the pointer rotation angle. Taking a 90-degree right rotation during pointer region extraction and deflection as an example, the pointer rotation angle is... The calculation formula is as follows: In the formula, , It is the average value of the boundary points closest to the center of the pointer image in the pointer tip region. The average value of the boundary points in the pointer tip region far from the center of the pointer image, single pointer reading. and pointer rotation angle The relationship between them is as follows: The pointer reading process is based on the horizontal coordinate obtained during pointer region extraction and deflection. Sorted from left to right, the reading of the least significant bit pointer is: In the formula, This indicates rounding down. Indicates rounding up. For the first The readings of the first pointer, followed by the readings of the remaining pointers: In the formula, The final pointer read vector is represented as Its value is equal to .

9. A high-precision reading method for a pointer-type intelligent camera water meter according to claim 5, characterized in that, The digital reading module includes the number of digits in the character wheel and the character wheel image, digital image segmentation, segmentation image standardization and grayscale conversion, dual threshold last digit recognition, and digital reading process; The number of digits in the character wheel and the number of digits in the character wheel image. This refers to the label information for the character wheel region in the image correction module. The character wheel image is obtained through... Corrected type wheel area; The digital image segmentation is based on the number of digits. and the width of the character wheel area Divide the parts proportionally; The segmented image standardization and grayscale conversion involves converting the segmented single digital color image to grayscale and standardizing the image size. The dual-threshold last digit recognition identifies standardized and grayscale single digits; it first expands the original digit classification range to 20 categories, including half-character carry states; The CNN classification models for 0-9 and 00-19 are as follows: In the formula, For digital images, and The prediction results are for the 0-9 and 00-19 classification models, respectively. Predict the category of the last digit, when predicting the category When it is a non-odd number and the pointer reading Less than the threshold At that time, the reading result for When it exceeds the threshold The reading result is Similarly, when predicting categories When the number is odd, then consider Rounding up or down, pointer reading Less than the threshold Round up, otherwise round down; however, this process... ,as well as When rounding up, an overflow may occur, requiring correction of both the current and previous digits. Specifically, when... When it is -1, it is automatically corrected to 9. When the value is 10, it is automatically corrected to 0, and a carry flag is set. ; The digit reading process involves identifying the remaining digits sequentially from the least significant digit to the most significant digit, based on the recognition of the last digit. First, the digits are divided into carry-over and non-carry-over states. When the previous digit is... Prediction Category If the previous digit's final identification result is 0 or 9, then the current digit is determined to be in a carry state; there are two cases for the carry state: Case 1: If the current digit When in a carry state and the lower bit is 0; use predict ,like It is an odd number, and the digits are read as follows. Otherwise, the digital reading If corrected If the value is 10, then force it to be set to 0 and set the flag. ; Case 2: If the current digit When in a carry-over state and the lowest digit is 9; use predict ,like It is an odd number, and the digits are read as follows. Otherwise, the digital reading If corrected If the value is -1, then it is forced to be set to 9, and the flag is set. ; Otherwise, the digital reading is as follows: Predict the data and clear the carry flag; the final digital reading result is calculated as follows: 。

Images