WO2022057103A1

WO2022057103A1 - Deep-learning-based method for automatically reading pointer instrument

Info

Publication number: WO2022057103A1
Application number: PCT/CN2020/134266
Authority: WO
Inventors: 熊继平; 李金红; 陈泽辉; 朱凌云
Original assignee: 浙江师范大学
Priority date: 2020-09-15
Filing date: 2020-12-07
Publication date: 2022-03-24
Also published as: CN112149667A

Abstract

A deep-learning-based method for automatically reading a pointer instrument. The method comprises the following steps: S1, inputting an instrument image, which needs to be subjected to detection, into an instrument disk pointer detection model trained by using a convolutional neural network, and performing detection on said instrument image, so as to obtain the position of an instrument disk and the position of a pointer; S2, performing binarization processing on said instrument image, so as to obtain a black and white binarized image; S3, according to the obtained position information of the pointer, cropping the binarized black and white image to obtain a pointer area; and S4, according to the obtained pointer area, obtaining the deviation angle of the pointer, and then obtaining a corresponding degree according to the measuring range of an instrument, thereby realizing the reading of a pointer instrument. An instrument disk and a pointer are detected by means of a deep learning method, the deviation angle of the pointer is obtained by using obtained position information of the instrument disk and the pointer, and the reading of the instrument is then obtained according to the measuring range of the instrument. The accuracy is high, the steps are simple, and the practicability is high.

Description

A method of automatic reading of pointer meter based on deep learning

technical field

The invention relates to the technical field of computer vision, and more particularly to a method for automatic reading of a pointer meter based on deep learning.

Background technique

With the continuous improvement and development of pattern recognition technology, computer technology and other technologies, machine vision has made great progress and development. At present, in many enterprises, there are a large number of meters, and the reading of the meters is done manually. The workload is large and the error rate is relatively high. Therefore, it is very necessary to develop automatic reading of pointer meters.

Most of the existing automatic readings of pointer-type meters are based on traditional image processing technology, and some methods propose to first use the convolutional neural network model to detect the meter target image; then use the improved effective and accurate scene text detector (EAST) algorithm to detect The instrument target image is used for text detection, and then the printed number recognition model is used to filter out the meter scale numbers, and the position information and value of the instrument scale numbers are obtained. The output value is combined with the angle method to complete the meter reading identification. In this method, the image quality has a great influence on the extraction result in the steps of extracting the pointer line and the center of the instrument, and the whole research method has many steps and is more complicated. (Xu Fabing, Wu Huaiyu, Chen Zhihuan, Yu Han. Research on the detection and recognition of pointer-type instruments based on deep learning [J]. High-tech Communications, 2019, 29(12): 1206-1215.) There are other methods that are based on depth The learned Faster-RCNN algorithm locates the dial; then, according to the characteristics and geometric characteristics of the instrument, the methods of connected domain analysis, image threshold segmentation, and least squares are used to locate the center of the dial, segment and refine the pointer; The numbers are divided, and the LeNet-5 convolutional neural network based on deep learning is used to identify the dial numbers and combine the scale values to calculate the division values, and on this basis, the numbers are obtained. However, this method has shortcomings. When dividing and recognizing numbers, if the inclination angle is too large and the dial has a certain degree of occlusion on the scale, the reading and interpretation may be inaccurate, which has certain limitations. (Liu Kui. Recognition of Pointer Instrument Recognition Based on Deep Learning [D]. Huazhong University of Science and Technology, 2017.) In addition, another research method is to use the convolutional neural network model to detect the bounding box of the instrument target in the current field of view position, calculate the deviation value from the center of the field of view and the image ratio, adjust the camera position and zoom factor accordingly, so that the coordinates of the center point of the image coincide with the coordinates of the center point of the bounding box, and then eliminate the distortion of the instrument image through perspective transformation, bilateral filtering, MSRCR The algorithm enhances the image, and finally the Hough transform detects the dial and pointer of the instrument to complete the identification of the instrument reading. This research method also has the problem of complex steps, high requirements on hardware computing resources, and some missed detections. (Xing Haoqiang, Du Zhiqi, Su Bo. Detection and identification method of pointer instrument in substation [J]. Journal of Instrument and Meter, 2017,38(11):2813-2821.)

In general, the traditional method of detecting pointers with image technology is easily affected by image quality, and changes in illumination and shooting angles will cause interference. Resource requirements are high. Existing methods based on deep learning rely on the recognition of digital characters, and there are problems such as recognition accuracy. Compared with the existing method, the present invention uses the deep learning to detect the pointer, has extremely strong anti-interference, can adapt to different environments for positioning and identification, does not need to preprocess the image before detection, and the steps of the whole research process are relatively simple , the algorithm is more efficient.

Therefore, how to provide a deep learning-based pointer meter reading method with high recognition accuracy is an urgent problem to be solved by those skilled in the art.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a method for reading a pointer-type meter based on deep learning. The method of deep learning is used to detect the meter disk and the pointer, and the obtained position information of the meter disk and the pointer is used to obtain the deviation of the pointer. angle, and then obtain the reading of the meter according to the range of the meter, which not only has high accuracy, but also has simple steps and strong practicability.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method for automatic reading of an analog meter based on deep learning, which is used to determine the reading of the meter through image information, including the following steps:

S1. Input the instrument image to be detected using the instrument disc pointer detection model trained by the convolutional neural network for detection, and obtain the position of the instrument disc and the pointer;

S2. Perform binarization processing on the instrument image to be detected to obtain a black and white binarized image;

S3. According to the obtained position information of the pointer, the black and white image after binarization is clipped to obtain the pointer area;

S4. According to the obtained pointer area, the deflection angle of the pointer is obtained, and then according to the range of the meter, the corresponding degree is obtained, so as to realize the reading of the pointer type meter.

Preferably, the specific method of using the convolutional neural network to train the instrument disc pointer detection model in S1 is:

Collect pointer meter data and mark it;

The collected pointer-type meter data and labels are input into a convolutional neural network as sample data to train the convolutional neural network to obtain the meter disc pointer detection model.

Preferably, the specific content of S1 includes:

Input the image of the pointer instrument to be detected into the trained instrument disc pointer detection model for detection, and mark the positions of the instrument disk and the pointer with a rectangular frame on the original image of the pointer instrument to be detected.

Preferably, the specific content of S2 includes:

Binarize the original image to obtain a binarized image with only pixels with pixel values of 0 and 255.

Preferably, the specific content of S3 includes:

Obtain the coordinates of the rectangular frame according to the pointer position marked by the rectangular frame, and crop the rectangular frame on the binarized image to obtain the pointer area.

Preferably, the specific content of S4 includes:

Refine the pointer according to the image in the pointer area, detect the central axis of the pointer, obtain the slope of the straight line where the central axis of the pointer is located, and then convert it into the deflection angle in the coordinate system with the center of the disc as the origin;

According to the obtained deflection angle and the range of the meter, the indication of the meter is calculated to obtain the reading of the pointer meter.

As can be seen from the above technical solutions, compared with the prior art, the present invention provides a method for automatic reading of a pointer-type meter based on deep learning, and the deep learning method is used for pointer detection, which can accurately determine where the pointer is located. This effectively solves the problem of low accuracy of the detection method in the prior art. Secondly, compared with the traditional image processing based on the pointer to fit a straight line, the present invention does not need to perform many preprocessing steps on the image in the early stage, which effectively simplifies The overall method steps can be reduced, and the dependence on image quality can also be reduced, and it has strong anti-interference.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only It is an embodiment of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to the provided drawings without creative work.

Fig. 1 accompanying drawing is the overall method flow chart provided by the present invention;

Fig. 2 accompanying drawing is the flow chart of training instrument disc pointer detection model provided by the present invention;

[Correction 24.12.2020 according to Rule 91]
FIG. 3 is a flow chart of determining the meter number provided by the present invention.

detailed description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The embodiment of the present invention discloses a deep learning-based automatic reading method of a pointer-type meter, which is used to determine the display number of the meter through image information. As shown in FIG. 1 , the method includes the following steps:

S4. According to the obtained pointer area, find out the deflection angle of the pointer, and then according to the range of the meter, get the corresponding degree, so as to realize the reading of the pointer meter.

In order to further realize the above technical solution, as shown in FIG. 2 , in this embodiment, a sample image of a pointer-type meter is obtained, annotating software is used to annotate it, and a convolutional neural network is used to train an instrument disc pointer detection model, wherein:

The obtained pointer meter samples should include those showing different indications and different shooting angles;

The convolutional neural network adopts a smaller network structure to improve the detection speed.

Use the labeling software to label the instrument disc and pointer in the obtained pointer instrument image, that is, frame and label each category in the sample image;

The labeled samples are input into a shallow convolutional neural network for training, and a detection model for instrument discs and pointers is obtained.

In order to further realize the above technical solution, the specific content of S1 includes:

Input the image of the pointer instrument to be detected into the trained instrument disc pointer detection model for detection, and mark the position of the instrument disk and the pointer with a rectangular frame on the original image of the pointer instrument to be detected.

In order to further realize the above technical solution, as shown in Figure 3, the specific content of S2 includes:

In order to further realize the above technical solution, the specific content of S3 includes:

In order to further realize the above technical solution, the specific content of S4 includes:

Using the feature that the pixel value of the pointer part in the pointer area is 255, and the pixel value of the other parts of the rectangle is 0, the pointer is refined in the pointer area image, and the axis of the pointer is detected. The slope of the straight line where the axis is located, and then converted into the deflection angle in the coordinate system with the center of the disc as the origin;

The invention discloses and provides a method for automatic reading of a pointer-type meter based on deep learning. A convolutional neural network is used to detect the disk and pointer of the meter, and then according to the obtained position information, it is determined in a coordinate system centered on the disk. The deflection angle of the pointer is used to obtain the indication of the meter. Specifically, the present invention will firstly use the convolutional neural network to train a network model for detecting the meter disc and the pointer through samples, and secondly obtain the meter disc and the pointer obtained through the model. The position information of the pointer is calculated to obtain the deflection angle of the pointer in the coordinate system with the center of the disc as the origin, and the indication of the meter is finally obtained according to the range of the meter.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method.

The above description of the disclosed embodiments enables any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

A method for automatic reading of a pointer-type meter based on deep learning, which is used to determine the indication of the meter through image information, characterized in that it includes the following steps:

S1. Input the instrument image to be detected using the instrument disc pointer detection model trained by the convolutional neural network for detection, and obtain the position of the instrument disc and the pointer;

S2. Perform binarization processing on the instrument image to be detected to obtain a black and white binarized image;

S3. According to the obtained position information of the pointer, the black and white image after binarization is clipped to obtain the pointer area;

S4. According to the obtained pointer area, the deflection angle of the pointer is obtained, and then according to the range of the meter, the corresponding degree is obtained, so as to realize the reading of the pointer type meter.
The method for automatic reading of a pointer-type meter based on deep learning according to claim 1, characterized in that, in S1, the specific method of using convolutional neural network to train the meter disc pointer detection model is:

Collect pointer meter data and mark it;

The collected pointer-type meter data and labels are input into a convolutional neural network as sample data to train the convolutional neural network to obtain the meter disc pointer detection model.
The method for automatic reading of an analog meter based on deep learning according to claim 1, wherein the specific content of S1 includes:

Input the image of the pointer instrument to be detected into the trained instrument disc pointer detection model for detection, and mark the positions of the instrument disk and the pointer with a rectangular frame on the original image of the pointer instrument to be detected.
The method for automatic reading of an analog meter based on deep learning according to claim 1, wherein the specific content of S2 includes:

Binarize the original image to obtain a binarized image with only pixels with pixel values of 0 and 255.
The method for automatic reading of an analog meter based on deep learning according to claim 3, wherein the specific content of S3 includes:

Obtain the coordinates of the rectangular frame according to the pointer position marked by the rectangular frame, and crop the rectangular frame on the binarized image to obtain the pointer area.
The method for automatic reading of an analog meter based on deep learning according to claim 1, wherein the specific content of S4 includes:

Refine the pointer according to the image in the pointer area, detect the central axis of the pointer, obtain the slope of the straight line where the central axis of the pointer is located, and then convert it into the deflection angle in the coordinate system with the center of the disc as the origin;

According to the obtained deflection angle and the range of the meter, the indication of the meter is calculated to obtain the reading of the pointer meter.