Pointer instrument reading method and device based on lightweight rotating box target detection, equipment and medium

Abstract

A pointer instrument reading method, device, equipment and medium based on lightweight rotating frame target detection are disclosed, the method comprising: detecting a pointer instrument dial in a patrol image, cropping to obtain a dial image, and then enhancing and preprocessing; performing lightweight rotating frame target detection on the enhanced dial image to obtain dial center, pointer, main scale line and scale number information; combining adjacent scale numbers into scale values, and one-to-one corresponding the scale values with the main scale line; correcting the enhanced dial image using the dial center and the main scale line and adopting a perspective transformation method, and mapping the dial center, the pointer, the main scale line and the corresponding scale values onto the corrected dial image; and calculating the reading result of the pointer instrument according to the corrected dial image using an angle method. The present application significantly reduces the parameter quantity of the rotating frame target detection model, greatly improves the instrument reading speed, and facilitates meeting the edge lightweight deployment requirements.

Description

Technical Field

[0001] This invention belongs to the field of instrumentation and image recognition technology, specifically relating to a pointer instrument reading method, device, equipment, and medium based on lightweight rotating frame target detection. Background Technology

[0002] Pointer instruments, with their high efficiency, reliability, and economic advantages, are widely used in important sectors of the national economy such as power, energy, and chemical industries. Timely and accurate readings of these instruments are crucial for the safe and stable operation of many critical pieces of equipment. Traditional manual reading methods are not only time-consuming and labor-intensive but also prone to errors. Furthermore, reading instruments on high-temperature, high-pressure, high-speed rotating, or high-voltage equipment poses significant safety risks. In recent years, with the rapid development of image processing technology, image-based pointer instrument reading recognition methods have emerged, enabling non-contact, automatic acquisition of pointer instrument readings and attracting widespread attention from researchers.

[0003] Currently, many substations, power plants, and industrial production workshops are equipped with surveillance cameras or inspection robots, which can replace manual periodic inspections and achieve automatic reading of pointer instrument readings without human intervention, effectively reducing the cost and risk of manual inspections. The instrument reading recognition methods used in practical applications mainly include traditional image processing methods and mainstream deep learning methods. Traditional methods mostly use Hough transform or template matching to identify instrument readings, which are simple and intuitive, but their accuracy and robustness are insufficient. Deep learning methods mostly use object detection methods to locate the instrument dial, then use semantic segmentation methods to extract information such as the center point, pointer, and scale lines, and use character recognition methods to extract scale values, finally achieving the reading. Compared with traditional image processing methods, deep learning-based methods have stronger generalization and robustness. However, due to the existence of semantic segmentation and character recognition models, existing methods always face the problems of large model parameters and slow recognition speed, making it difficult to adapt to lightweight deployment requirements. At the same time, the shooting environment of pointer instruments is complex and diverse, and current methods have poor adaptability to different indoor and outdoor environments, severely limiting their widespread application. Therefore, researching fast and efficient methods for identifying pointer instrument readings is of great significance for promoting the intelligent transformation of factory workshops and ensuring the safe and stable operation of power energy systems. Summary of the Invention

[0004] This invention provides a pointer instrument reading method, apparatus, device, and medium based on lightweight rotating frame target detection, aiming to solve the technical problems of existing methods having large computational loads that are difficult to deploy in a lightweight manner, and being prone to misidentification due to insufficient adaptability to complex indoor and outdoor shooting environments.

[0005] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0006] A pointer instrument reading method based on lightweight rotating frame target detection includes:

[0007] Step 1: Perform target detection on the pointer instrument dial in the inspection image and crop it to obtain the dial image;

[0008] Step 2: Perform enhancement preprocessing on the dial image;

[0009] Step 3: Perform lightweight rotating bounding box target detection on the enhanced preprocessed dial image to obtain dial information: dial center, hands, main scale lines, and scale numbers;

[0010] Step 4: Based on the distance between the scale numbers, combine adjacent scale numbers into scale values, and match the scale values ​​with the main scale lines one by one;

[0011] Step 5: Using the dial center and main scale line in the dial information, the perspective transformation method is used to correct the enhanced preprocessed dial image, and the dial center, hands, main scale line and corresponding scale value are mapped onto the corrected dial image.

[0012] Step 6: Based on the calibrated dial image mapped with dial information, calculate the reading of the pointer instrument using the angle method.

[0013] Furthermore, a lightweight rotating bounding box target detection model based on YOLOv8 OBB is used to perform lightweight rotating bounding box target detection on the dial image. The lightweight improvement includes replacing the Conv downsampling convolutional modules in the YOLOv8 OBB backbone network and neck network with partially depth-separable convolutional modules, except for the first and second convolutional modules in the backbone network.

[0014] The depthwise separable convolutional module, abbreviated as PDSConv module, consists of three layers: the first layer is a 3×3 depthwise convolution, batch normalization, and ReLU activation layer; the second layer is a 1×1 pointwise convolution, batch normalization, and ReLU activation layer; and the third layer is a concatenation of the output of the first layer and the output of the second layer along the channel dimension.

[0015] Furthermore, a lightweight rotating bounding box target detection model based on YOLOv8 OBB is used to perform lightweight rotating bounding box target detection on the dial image; the lightweight improvement includes replacing the C2f feature optimization modules in the YOLOv8 OBB backbone network and neck network with multi-scale feature fusion modules based on partial convolution.

[0016] The multi-scale feature fusion module based on partial convolution is abbreviated as CSP-PMSFA module. It replaces the Bottleneck module in the C2f feature optimization module entirely with a redesigned PMSFA module. The PMSFA module includes: First, the input features are subjected to a 3×3 convolution operation; then, they are split equally according to the number of channels, with the first branch subjected to a 5×5 convolution operation, and the second branch remaining unchanged; next, the features obtained from the 5×5 convolution operation are split equally according to the number of channels, with the first branch subjected to a 7×7 convolution operation, and the second branch remaining unchanged; then, the first branch and the second branch obtained from the two splits are concatenated along the channel dimension; next, the concatenated features are fused using a 1×1 convolution; finally, the fused features are added to the input features.

[0017] Furthermore, step 4 specifically includes:

[0018] Obtain the center coordinates of each scale number detection box and decimal point detection box, and combine each scale number and decimal point with the distance between the center coordinates within the corresponding preset threshold range into a scale value in order of position, thereby obtaining the detection box for each scale value.

[0019] Obtain the center coordinates of each scale value detection box and the main scale line detection box, and match the scale values ​​whose distance between the center coordinates is within the corresponding preset threshold range with the main scale line.

[0020] Furthermore, step 6 specifically includes:

[0021] Based on the calibrated dial image with the dial information mapped, the hands and main scale lines are obtained. By calculating and comparing the distance between the hands and each main scale line, the nearest main scale lines on the left and right sides of the hands are determined and recorded as the left scale line and the right scale line, respectively.

[0022] Using the center of the dial as the origin of the coordinate system, the angle A between the left and right scale lines is calculated based on the center coordinates of the detection frames of the left and right scale lines. The angle B between the pointer and the left scale line is calculated based on the center coordinates of the detection frames of the left scale line and the pointer. The angle C between the pointer and the right scale line is calculated based on the center coordinates of the detection frames of the right scale line and the pointer.

[0023] The scale value of the left scale line Using the included angle B as a reference value, calculate the instrument reading. ;

[0024] The scale value of the left scale line Using the included angle C as a reference value, calculate the instrument reading. ;

[0025] The scale value of the right-hand scale line Using the included angle C as a reference value, calculate the instrument reading. ;

[0026] The scale value of the right-hand scale line Using the included angle B as a reference value, calculate the instrument reading. ;

[0027] The average of the above four instrument readings is taken as the final instrument reading result V:

[0028] .

[0029] A pointer instrument reading device based on lightweight rotating frame target detection, used to implement the pointer instrument reading method based on lightweight rotating frame target detection as described in any of the above claims, comprising:

[0030] The dial target detection module is used to: perform target detection on the dial of pointer instruments in the inspection image and crop it to obtain the dial image;

[0031] The preprocessing module is used to: crop the detected dial area to obtain a dial image, and enhance the dial image;

[0032] The rotating bounding box target detection module is used to perform lightweight rotating bounding box target detection on the enhanced preprocessed dial image to obtain dial information: dial center, hands, main scale lines, and scale numbers.

[0033] The scale value acquisition and mapping module is used to: combine adjacent scale numbers into scale values ​​based on the distance relationship between scale numbers, and map the scale values ​​to the main scale lines one by one;

[0034] The image correction module is used to: use the dial center and main scale line in the dial information to correct the enhanced preprocessed dial image using the perspective transformation method, and map the dial center, hands, main scale line and corresponding scale value onto the corrected dial image;

[0035] The reading calculation module is used to calculate the reading result of the pointer instrument using the angle method based on the calibrated dial image that maps the dial information.

[0036] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to implement the pointer instrument reading method based on lightweight rotating frame target detection as described above.

[0037] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the pointer instrument reading method based on lightweight rotating frame target detection as described in any of the preceding claims.

[0038] Beneficial effects

[0039] This invention discloses a method, apparatus, device, and medium for reading pointer instruments based on lightweight rotating bounding box target detection. First, a YOLOv8n target detection model is used to locate the dial image. Then, a lightweight improved YOLOv8 OBB rotating bounding box target detection model is used to extract information such as the dial center, pointer, main scale lines, and scale numerals from the dial image. Finally, the final reading of the pointer instrument is calculated using perspective transformation correction and the average angle method. Compared with existing technologies, this invention significantly reduces the number of model parameters and greatly improves the instrument reading speed while ensuring high accuracy, making it more suitable for lightweight deployment at edge devices. Details are as follows:

[0040] (1) The present invention is a fast instrument reading recognition method that does not rely on semantic segmentation and OCR character recognition. It uses a rotating target detection model to extract key information of the dial such as the center of the dial, pointer, main scale line, and scale number at one time, which greatly improves the reading speed and accuracy.

[0041] (2) The existing YOLOv8 OBB model has been improved by lightweighting. The PDSConv module and CSP-PMSFA module were designed to replace the Conv downsampling convolution module and C2f feature optimization module in the YOLOv8 OBB network model, respectively, which effectively improved the reading speed of the pointer instrument. Attached Figure Description

[0042] Figure 1 These are the main steps of the method described in the embodiments of the present invention.

[0043] Figure 2 These are indoor and outdoor installation scenarios for pointer instruments, with sub-figures (a), (b), and (c) representing different installation scenarios.

[0044] Figure 3 This is the architecture of the lightweight and improved rotating box target detection model according to an embodiment of the present invention.

[0045] Figure 4 This is the structure of the PDSConv module described in the embodiments of the present invention. Figure 4 (a) is a module structure diagram. Figure 4 (b) is a schematic diagram of the feature extraction process.

[0046] Figure 5 This is the structure of the CSP-PMSFA module described in the embodiments of the present invention.

[0047] Figure 6These are diagrams illustrating the effect of dial information extraction using the method described in this embodiment of the invention, where sub-graphs (a), (b), (c), and (d) respectively extract dial information from different pointer instruments.

[0048] Figure 7 This is a schematic diagram of the angle method instrument reading processing in an embodiment of the present invention. Detailed Implementation

[0049] The embodiments of the present invention will be described in detail below. These embodiments are based on the technical solutions of the present invention and provide detailed implementation methods and specific operation processes to further explain the technical solutions of the present invention.

[0050] This embodiment provides a pointer instrument reading method based on lightweight rotating frame target detection, referencing... Figure 1 As shown, it includes the following steps:

[0051] Step 1: Perform target detection on the pointer instrument dial in the inspection image and crop to obtain the dial image.

[0052] The inspection images can be captured by surveillance cameras or inspection robots during the inspection of pointer instruments, such as... Figure 2 As shown.

[0053] The target detection in this step can be achieved by training your own pointer instrument target detection model based on the YOLOv8n target detection algorithm.

[0054] Step 2: Perform enhancement preprocessing on the dial image to obtain a dial image with better image quality.

[0055] Image enhancement methods that can be used in this step include, but are not limited to, contrast stretching and contrast-limited adaptive histogram equalization.

[0056] Step 3: Perform lightweight rotating bounding box target detection on the enhanced preprocessed dial image to obtain dial information: dial center, hands, main scale lines, and scale numbers.

[0057] This embodiment uses a lightweight improved rotating bounding box target detection model based on YOLOv8 OBB to perform lightweight rotating bounding box target detection on the dial image. (Reference) Figure 3As shown, the lightweight improvements of this invention include: (1) replacing the Conv downsampling convolution modules in the YOLOv8 OBB backbone and neck network with partially depth separable convolution modules (PDSConv modules), except for the first and second convolution modules in the backbone network, to effectively reduce the computational complexity of the model; (2) replacing all C2f feature optimization modules in the YOLOv8 OBB backbone and neck network with cross-stage partial network & partial multi-scale feature aggregation (CSP-PMSFA modules), which enhances the feature fusion effect and effectively improves the model inference speed. The YOLOv8 OBB model with improved PDSConv and CSP-PMSFA modules has significantly reduced parameters and slightly improved accuracy.

[0058] like Figure 4 As shown, the PDSConv module includes three layers: the first layer is a 3×3 depthwise convolution, batch normalization, and ReLU activation layer; the second layer is a 1×1 pointwise convolution, batch normalization, and ReLU activation layer; and the third layer is a concatenation of the output of the first layer and the output of the second layer along the channel dimension.

[0059] like Figure 5 As shown, the CSP-PMSFA module is derived by replacing the Bottleneck module in the C2f feature optimization module with a newly designed PMSFA module. The PMSFA module includes: First, the input features are subjected to a 3×3 convolution operation; then, they are split equally according to the number of channels, with the first branch subjected to a 5×5 convolution operation and the second branch remaining unchanged; next, the features obtained from the 5×5 convolution operation are split equally according to the number of channels, with the first branch subjected to a 7×7 convolution operation and the second branch remaining unchanged; then, the first branch and the second branch obtained from the two splits are concatenated along the channel dimension; next, the concatenated features are fused through a 1×1 convolution; finally, the fused features are added to the input features.

[0060] Before performing the lightweight rotated bounding box object detection operation, this step involves collecting at least 100 watch face images and meticulously annotating key watch face information such as the center, hands, main scale lines, and scale numbers. Then, a lightweight, improved YOLOv8 OBB model is used for training. The watch face information output by the lightweight, improved YOLOv8 OBB model can be used for the lightweight rotated bounding box object detection operation in this step. Figure 6 The diagram shows the effect of this invention in extracting information from the dials of various pointer instruments.

[0061] Step 4: Based on the distance between the scale numbers, combine adjacent scale numbers into scale values, and match the scale values ​​with the main scale lines one by one.

[0062] First, obtain the center coordinates of each scale number detection box and decimal point detection box. Then, combine the scale numbers and decimal points whose center coordinates are within the corresponding preset threshold range into a scale value in order of position. All detection boxes that combine a scale value are also merged to obtain the detection box of that scale value.

[0063] Then, obtain the center coordinates of each scale value detection box and the main scale line detection box, and match the scale values ​​with the Euclidean distance between the center coordinates within the corresponding preset threshold range with the main scale line.

[0064] Step 5: Using the dial center and main scale lines from the dial information, a perspective transformation method is used to correct the enhanced preprocessed dial image, and the dial center, hands, main scale lines, and corresponding scale values ​​are mapped onto the corrected dial image. This aims to avoid inaccurate readings caused by dial image distortion.

[0065] First, establish a coordinate system with the center of the dial as the origin, transform the coordinates of the center point of the main scale line into this coordinate system, and take the coordinates of the first four main scale lines as the coordinates before the transformation.

[0066] Then, similarly, the coordinates of the corresponding scale line under the front view of the dial are taken as the transformed coordinates, and perspective transformation is performed using the four sets of corresponding coordinates to correct the enhanced preprocessed dial image.

[0067] Finally, the center of the dial, the hands, the main scale lines, and the corresponding scale values ​​are mapped onto the calibrated dial image.

[0068] Step 6: Based on the calibrated dial image mapped with dial information, calculate the reading of the pointer instrument using the angle method.

[0069] First, based on the calibrated dial image with mapped dial information, the pointer and main scale lines are obtained. By calculating and comparing the distance between the pointer and each main scale line, the nearest main scale lines on the left and right sides of the pointer are determined and recorded as the left scale line and the right scale line, respectively.

[0070] Then, refer to Figure 7 As shown, with the center of the dial as the origin of the coordinate system, the angle A between the left and right scale lines is calculated based on the center coordinates of the detection boxes of the left and right scale lines. The angle B between the pointer and the left scale line is calculated based on the center coordinates of the detection boxes of the left scale line and the pointer. The angle C between the pointer and the right scale line is calculated based on the center coordinates of the detection boxes of the right scale line and the pointer.

[0071] Then use the scale value of the left scale line Using the included angle B as a reference value, calculate the instrument reading. ;

[0072] The scale value of the left scale line Using the included angle C as a reference value, calculate the instrument reading. ;

[0073] The scale value of the right-hand scale line Using the included angle C as a reference value, calculate the instrument reading. ;

[0074] The scale value of the right-hand scale line Using the included angle B as a reference value, calculate the instrument reading. ;

[0075] Finally, the average of the four instrument readings is taken as the final instrument reading result V:

[0076] .

[0077] The following uses the average relative error and average relative scale error Two indicators are used to evaluate the accuracy of pointer instrument readings, and the calculation formula is as follows:

[0078]

[0079]

[0080] Where i represents the i-th instrument image. and Let represent the relative error and relative scale error of the i-th instrument reading, respectively. This indicates the true value of the instrument reading. This indicates the predicted value of the instrument reading. This indicates the maximum scale value of the instrument reading, and N represents the number of instrument images.

[0081] Considering that the relative error of instrument readings is generally allowed to be within the range of 1% or 5%, therefore, it is also adopted. and These two accuracy metrics evaluate the overall performance of the instrument readings, and the calculation formulas are as follows:

[0082]

[0083] in, , The set tolerance threshold is typically 0.01 or 0.05.

[0084] The speed of reading a pointer instrument can be expressed as the number of image frames processed per second (FPS), calculated as follows:

[0085]

[0086] Where N represents the number of instrument images, and T represents the time required to process N instrument images.

[0087] This invention uses object detection and rotated bounding box object detection models as its basic framework. Compared with existing frameworks based on object detection, semantic segmentation, and OCR character recognition, this invention's network model has fewer parameters, faster inference speed, and is more suitable for lightweight deployment at edge devices. Furthermore, this invention features a lightweight design based on the YOLOv8 OBB network structure, innovatively incorporating two lightweight modules: PDSConv and CSP-PMSFA. Ultimately, while maintaining the accuracy of dial information extraction, this invention significantly reduces the number of model parameters by approximately 33%. Tests on a dataset containing 3053 instrument images show that the method achieves an average relative error of 0.49% and an average relative scale error of 0.61% for the final instrument readings; with a maximum allowable relative error of 1%, the reading accuracy is 90.52%; and with a maximum allowable relative error of 5%, the reading accuracy reaches as high as 99.14%. The method achieves a running speed of 9.42fps on a Jetson Nano embedded board, significantly higher than other methods relying on semantic segmentation and OCR character recognition.

[0088] The above embodiments are preferred embodiments of this application. Those skilled in the art can make various changes or improvements based on them. Without departing from the overall concept of this application, such changes or improvements should fall within the scope of protection claimed in this application.

Claims

1. A pointer instrument reading method based on lightweight rotating frame target detection, characterized in that, include: Step 1: Perform target detection on the pointer instrument dial in the inspection image and crop it to obtain the dial image; Step 2: Perform enhancement preprocessing on the dial image; Step 3: Perform lightweight rotating bounding box target detection on the enhanced preprocessed dial image to obtain dial information: dial center, hands, main scale lines, and scale numbers; A lightweight rotating bounding box target detection model based on YOLOv8 OBB was used to perform lightweight rotating bounding box target detection on the dial image. The lightweight improvement includes replacing the Conv downsampling convolutional modules in the YOLOv8 OBB backbone network and neck network with partially depth-separable convolutional modules, except for the first and second convolutional modules in the backbone network. The partially depth-separable convolutional module, abbreviated as PDSConv module, consists of three layers: the first layer is a 3×3 depthwise convolution, batch normalization, and ReLU activation layer; the second layer is a 1×1 pointwise convolution, batch normalization, and ReLU activation layer; and the third layer is a concatenation of the output of the first layer and the output of the second layer along the channel dimension. The lightweight improvements include replacing all C2f feature optimization modules in the YOLOv8 OBB backbone network and neck network with multi-scale feature fusion modules based on partial convolution. The multi-scale feature fusion module based on partial convolution is abbreviated as CSP-PMSFA module. It replaces the Bottleneck module in the C2f feature optimization module entirely with a redesigned PMSFA module. The PMSFA module includes: First, the input features are subjected to a 3×3 convolution operation; then, they are split equally according to the number of channels, with the first branch subjected to a 5×5 convolution operation, and the second branch remaining unchanged; next, the features obtained from the 5×5 convolution operation are split equally according to the number of channels, with the first branch subjected to a 7×7 convolution operation, and the second branch remaining unchanged; then, the first branch and the second branch obtained from the two splits are concatenated along the channel dimension; next, the concatenated features are fused using a 1×1 convolution; finally, the fused features are added to the input features. Step 4: Based on the distance between the scale numbers, combine adjacent scale numbers into scale values, and match the scale values ​​with the main scale lines one by one; Step 5: Using the dial center and main scale line in the dial information, the perspective transformation method is used to correct the enhanced preprocessed dial image, and the dial center, hands, main scale line and corresponding scale value are mapped onto the corrected dial image. Step 6: Based on the calibrated dial image mapped with dial information, calculate the reading of the pointer instrument using the angle method.

2. The pointer instrument reading method based on lightweight rotating frame target detection according to claim 1, characterized in that, Step 4 specifically includes: Obtain the center coordinates of each scale number detection box and decimal point detection box, and combine each scale number and decimal point with the distance between the center coordinates within the corresponding preset threshold range into a scale value in order of position, thereby obtaining the detection box for each scale value. Obtain the center coordinates of each scale value detection box and the main scale line detection box, and match the scale values ​​whose distance between the center coordinates is within the corresponding preset threshold range with the main scale line.

3. The pointer instrument reading method based on lightweight rotating frame target detection according to claim 1, characterized in that, Step 6 specifically includes: Based on the calibrated dial image with the dial information mapped, the hands and main scale lines are obtained. By calculating and comparing the distance between the hands and each main scale line, the nearest main scale lines on the left and right sides of the hands are determined and recorded as the left scale line and the right scale line, respectively. Using the center of the dial as the origin of the coordinate system, the angle A between the left and right scale lines is calculated based on the center coordinates of the detection frames of the left and right scale lines. The angle B between the pointer and the left scale line is calculated based on the center coordinates of the detection frames of the left scale line and the pointer. The angle C between the pointer and the right scale line is calculated based on the center coordinates of the detection frames of the right scale line and the pointer. The scale value of the left scale line Using the included angle B as a reference value, calculate the instrument reading. ; The scale value of the left scale line Using the included angle C as a reference value, calculate the instrument reading. ; The scale value of the right-hand scale line Using the included angle C as a reference value, calculate the instrument reading. ; The scale value of the right-hand scale line Using the included angle B as a reference value, calculate the instrument reading. ; The average of the above four instrument readings is taken as the final instrument reading result V: 。 4. A pointer instrument reading device based on lightweight rotating frame target detection, used to implement the pointer instrument reading method based on lightweight rotating frame target detection as described in any one of claims 1-3, characterized in that, include: The dial target detection module is used to: perform target detection on the dial of pointer instruments in the inspection image and crop it to obtain the dial image; The preprocessing module is used to: crop the detected dial area to obtain a dial image, and enhance the dial image; The rotating bounding box target detection module is used to perform lightweight rotating bounding box target detection on the enhanced preprocessed dial image to obtain dial information: dial center, hands, main scale lines, and scale numbers. The scale value acquisition and mapping module is used to: combine adjacent scale numbers into scale values ​​based on the distance relationship between scale numbers, and map the scale values ​​to the main scale lines one by one; The image correction module is used to: use the dial center and main scale line in the dial information to correct the enhanced preprocessed dial image using the perspective transformation method, and map the dial center, hands, main scale line and corresponding scale value onto the corrected dial image; The reading calculation module is used to calculate the reading result of the pointer instrument using the angle method based on the calibrated dial image that maps the dial information.

5. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, the processor enables the processor to implement the pointer instrument reading method based on lightweight rotating frame target detection as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the pointer instrument reading method based on lightweight rotating frame target detection as described in any one of claims 1 to 3.

Images