CN111008628B - A method and device for automatic reading of a pointer instrument robust to light - Google Patents

Abstract

The invention discloses an automatic reading method and device of a pointer meter that is robust to light. The method includes the following steps: step A: image preprocessing; step B: detecting the pointer outline of the meter and determining the value of each pointer in the image Direction; Step C: Check the direction of the 0 scale line; Step D: Complete the final reading of the meter. The invention can still ensure high-precision reading of the meter image under the condition that the illumination and the attitude of the meter are uncontrollable, and can be applied to the automatic reading of various meters.

Description

A method and device for automatic reading of pointer-type instruments robust to illumination

Technical Field

The invention belongs to the field of computer vision and image processing learning, and in particular relates to an automatic reading method and device for a pointer-type instrument which is robust to illumination.

Background Art

Pointer instruments (such as pointer water meters) are widely used in all walks of life because of their simple structure, easy use, low cost and anti-electromagnetic interference. Traditional instrument readings are all done manually, which is not only inefficient, but also the physiological reaction of visual fatigue of personnel can easily affect the accuracy of readings. With the development of the information age, in order to solve the problems caused by manual readings, automatic reading of instruments has also become a research field. Automatic reading of pointer instruments is of great significance to improving the productivity of industrial production and civil facilities and reducing labor costs. For example, the reading of the instrument is calibrated during production to ensure that the accuracy of the instrument meets the standard requirements of the industry; patrol robots perform mobile measurement and monitoring of instrument equipment, etc.

At present, most automatic reading methods of pointer instruments based on computer vision complete the reading by measuring the relative angle between the direction of the pointer and the direction of the 0 scale line. In the stage of detecting the pointer direction, existing studies mostly use Huffman transform, least squares method and template matching method to detect linear pointers with simple structures. However, for instruments with more complex structures and pointer shapes and multiple sub-dials (such as most pointer water meters), many literatures usually use mathematical morphology and pointer structure information to extract the pointer contour and determine the pointer direction. However, the pointer detection results of these algorithms are affected by lighting conditions. In complex situations (such as uneven light and poor camera angle, poor imaging quality), it is often difficult to accurately detect the pointer contour.

In the stage of identifying the direction of the 0 scale line on the instrument dial, existing research can be divided into two strategies. The first strategy is to estimate the direction of the 0 scale line by aligning the instrument image with the template image. The second strategy is to directly detect the direction of the 0 scale line. Considering that the angle between the horizontal line and the zero scale line is unchanged when aligning the instrument image, the first type of strategy is to align the instrument image with the template image by aligning the camera. However, there are situations where the instrument image cannot be easily aligned with the template image, for example, because the camera has an out-of-plane rotation, so many studies have proposed methods for directly detecting the 0 scale. Among them, Ma et al. used an improved least squares method to detect the 0 scale line after removing the interference image. Yi et al. used K-means clustering to identify the 0 scale line. However, this type of method is affected by the image quality, and it is difficult to directly detect the zero scale line for images with poor imaging quality and blurred images. In addition, by utilizing the inherent relationship of the pointer rotation center, some methods can indirectly extract the direction of the 0 scale line, which is applicable to instruments with multiple pointers. They obtain the center of the instrument based on the structural feature that the pointer rotation center of each sub-dial is on the same circle. The 0 scale line is detected based on the angular relationship between the connecting line between the center of the instrument and the rotation center of the pointer and the 0 scale line. The reading results of these methods depend to a large extent on the detection results of the rotation center of the pointer. That is to say, the detection of the direction of the 0 scale line in this method depends on the inherent relationship between the pointers. The detection results of the pointers will further affect the detection of the direction of the 0 scale line, which lacks practicality and scalability.

From the above methods, we can see that the automatic instrument reading method is limited in practical application, mainly due to two deficiencies: 1) Affected by lighting conditions and instrument posture, when there is certain interference information on the instrument, the pointer and the 0 scale line cannot be accurately detected; 2) The existing instrument reading method depends on the unique structure of the instrument, and there is currently no general method applicable to various instrument readings.

In this context, it is particularly important to develop a method that can well resist light conditions and accurately complete the automatic reading of the instrument without relying on the unique structure of the instrument.

Summary of the invention

The technical problem to be solved by the present invention is to provide a pointer-type instrument automatic reading method which is robust to illumination and can accurately complete the automatic reading of the instrument in view of the deficiencies of the prior art.

The technical solution adopted by the present invention is as follows:

A method for automatic reading of a pointer-type instrument robust to illumination comprises the following steps:

Step A, preprocessing the instrument image;

Step B: for the instrument image preprocessed in step A, detecting the pointer contour therein and determining the pointer direction;

Step C: for the instrument image preprocessed in step A, detecting the direction of the 0 scale line therein;

Step D, complete the reading.

There is no restriction on the execution order of the above steps B and C. Step B may be executed first and then step C, or step C may be executed first and then step B, or steps B and C may be executed simultaneously. The division into steps B and C is only for the convenience of description, and does not constitute a limitation on the execution order of the steps in the technical solution of the present invention.

Furthermore, the specific processing process of step A is as follows:

Step A1, using the Huffman circle detection algorithm to extract the ROI area of the instrument image;

Step A2, first cut out all areas outside the minimum circumscribed matrix of the ROI area on the instrument image; then use the characteristics of the circle to determine whether the point q(i, j) in the minimum circumscribed matrix of the ROI area is within the ROI area. If q(i, j) is outside the ROI area, set the pixel value p(i, j) of the point to 255 to obtain the ROI image;

Step A3, enhancing the pointer information in the ROI area based on the color difference transformation; specifically: first grayscale the ROI image to obtain a grayscale image, and then perform an R-Y operation on the grayscale image to obtain an RY grayscale image; the pixel value Y of each pixel in the grayscale image is:

Y＝0.299*R+0.587*G+0.114*B

The pixel value R-Y of each pixel in the R_Y grayscale image is:

R-Y＝0.701*R-0.587*G-0.114*B

Among them, R, G and B are the three components of the pixel value of the corresponding pixel in the ROI image.

Furthermore, the specific processing process of step B is as follows:

Step B1, using the MSER algorithm to detect all maximum stable extreme value regions (MSER) on the R Y grayscale image, and taking each maximum stable extreme value region (MSER) as a candidate pointer contour;

Step B2: matching the correspondence between each candidate outline of the pointer and the instrument pointer based on the improved NMS algorithm (ANMS algorithm), and determining the final outline of each pointer on the instrument;

Step B3: Based on the final outline of each pointer on the instrument, determine the starting point s ( _xs , _ys ) and end point t ( _xt , _yt ) of each pointer respectively, fit the straight line where each pointer is located, and obtain the direction of each pointer.

Furthermore, the step B2 specifically includes the following steps:

Step B21, calculating the minimum circumscribed matrix area of all candidate pointer contours;

Step B22: For any two candidate contours of the pointer, calculate the area overlap IOU between their minimum circumscribed matrices:

Among them, _S1 and _S2 represent the areas of the minimum circumscribed matrices of the two candidate contours of the pointers, S represents the overlapping area of the minimum circumscribed matrices of the two candidate contours of the pointers, and the calculation formula of S is as follows:

和

取两个指针候选轮廓的最小外接矩阵的左上角坐标值中较大的坐标值；Among them, x _{right_min} and y _{right_min} take the smaller coordinate value of the lower right corner coordinate value of the minimum circumscribed matrix of the two pointer candidate contours.

and

Take the larger coordinate value of the upper left corner of the minimum circumscribed matrix of the two candidate contours of the pointer;

Step B23, setting the region overlap threshold IOU _th ;

For any two candidate pointer contours, compare their corresponding IOU and IOU _th. If the IOU is greater than IOU _th , they are considered to correspond to the same pointer on the instrument.

B24. For all candidate contours of the same pointer on the corresponding instrument, calculate the average value of their areas (the area of the candidate contour of the pointer can be taken as the number of pixels in the area it contains), and select the candidate contour of the pointer with the area closest to the average value as the final contour of the corresponding pointer on the instrument.

Furthermore, the step B3 specifically includes the following steps:

B31. Select the area contained in the final outline of the pointer on the ROI image, that is, the centroid s( _xs , _ys ) in the pointer area as the starting point of the pointer, and use the grayscale centroid method to calculate the centroid s( _xs , _ys ) of the outline. The formula is as follows:

Wherein, the formula f(u, v) represents the pixel value of the pixel at the coordinate (u, v) in the pointer region Ω on the ROI image;

B32. Determine the end point of the pointer based on the minimum angle method;

Assume that the final outline of a pointer consists of n points; for any three adjacent points among these n points, calculate the angle value corresponding to the middle point, so as to obtain a series of angle values; compare the sizes of all calculated angle values, and the middle point corresponding to the minimum angle value is the end point t(x _t , y _t ) of the pointer;

Suppose three adjacent points are A, B, and C, and the middle point is B. Calculate the size of ∠B in △ABC, that is, the angle value corresponding to the middle point B. The size of ∠B is calculated using the cosine theorem:

Furthermore, the specific processing process of step C is as follows:

Step C1, selecting a front image of an instrument as a template image, and manually marking the direction of the 0 scale line on the template image;

Step C2, using accelerated robust feature technology to detect feature points on the template image and the instrument image respectively;

Step C3, matching the feature points on the template image and the instrument image, and using the RANSC algorithm to remove mismatched points to further ensure the accuracy of the matching;

Step C4: based on the direction of the 0 scale line on the template image and the matching feature points between the template image and the instrument image, obtain the direction of the 0 scale line on the instrument image.

Furthermore, the step C4 specifically includes the following steps:

Step C41, assuming that there are k feature points on the template image that match the instrument image;

Step C42, randomly select two feature points p ( _xp , _yp ) and w ( _xw , _yw ) from the k matching feature points on the template image, _xp < _xw , to form a vector pw;

Step C43, calculate the angle θ between the vector pw on the template image and the 0 scale line:

Wherein, N(x _N , y _N ) and M(x _M , y _M ) are the starting point and end point of the 0 scale line on the template image respectively;

Step C44: Assume that the matching feature points on the instrument image corresponding to p( _xp , _yp ) and w( _xw , _yw ) on the template image are p′( _xp ′, _yp ′) and w′(xw _′ , _yw ′);

Find a straight line p′f′ in the instrument image with an angle of θ with p′w′, and use its direction as a candidate 0 scale line direction; let the coordinates of f′ be (x _f ′, y _f ′), and its calculation formula is:

Step C45, repeating steps C42 to C44 T times, and ensuring that the two feature points randomly selected from the k matching feature points on the template image are different each time; through this operation, T candidate 0 scale line directions are found in the instrument image;

C46, adopt a voting strategy to determine a final 0 scale line direction from T candidate 0 scale line directions;

In the voting strategy, an angle threshold γ is set to evaluate the accuracy of each candidate 0 scale line direction; for each candidate 0 scale line direction, its score value is equal to the number of other candidate 0 scale line directions whose angles with it are less than the angle threshold γ; the candidate 0 scale line direction with the highest score value is selected as the 0 scale line direction in the instrument image.

Furthermore, the specific processing process of step D is as follows:

Step D1, calculate the angle δ between the pointer direction and the 0 scale line direction, the calculation formula is:

Among them, v1 is the direction vector of the pointer, and v2 is the direction vector of the 0 scale line;

Step D2, determine the meter reading according to the angle δ between the pointer direction and the 0 scale line direction.

The present invention also provides a pointer-type instrument automatic reading device that is robust to illumination, comprising a processor. The processor adopts the above-mentioned pointer-type instrument automatic reading method to realize the pointer-type instrument automatic reading.

Furthermore, the pointer-type instrument automatic reading device further comprises an image acquisition module for acquiring instrument images; the processor completes the pointer-type instrument automatic reading based on the instrument images acquired by the image acquisition module.

The present invention can still ensure high-precision reading of instrument images when the illumination and instrument posture are uncontrollable, and can be applied to automatic reading of various instruments.

Beneficial Effects

The present invention discloses a method and device for automatic reading of pointer-type instruments that is robust to illumination, which has the following advantages: 1) Based on local invariant features, the structural information is locally encoded to determine the direction of the pointer and the zero scale line, thereby improving the accuracy of the reading. It neither directly detects the zero scale nor relies on the inherent structure between the pointers, and can be applied to the automatic reading of other types of instruments; 2) Under different environments, the accuracy of the pointer detection and reading results is significantly better than the most advanced existing methods. 3) The method can simultaneously meet the requirements of robustness, accuracy, and the need to control a large amount of auxiliary information such as the illumination conditions and imaging angles of the image, and is highly practical.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flow chart of a method for automatic reading of a pointer-type instrument that is robust to illumination in an example of the present invention.

FIG2 is a test set, in which FIG2(a) is an instrument image with uniform illumination in an example of the present invention; FIG2(b) is an instrument image with uneven illumination in an example of the present invention; and FIG2(c) is an instrument image with blurred and dim imaging.

FIG. 3 is a flow chart showing the processing of a sample (test image) in the image preprocessing stage in an example of the present invention.

FIG. 4 is a flow chart of the pointer detection phase in the example of the present invention.

FIG. 5 is a pointer candidate outline detected by MSER in an example of the present invention.

Figure 6 is the final contour of the pointer detected in the example of the present invention, wherein the contour marked in Figure 6(a) is the only contour determined after redundant contours are eliminated using the ANMS algorithm in the example of the present invention; the contour centroid in Figure 6(b) is the starting point of the instrument pointer detected in the example of the present invention, the black dot is the end point of the instrument pointer detected in the example of the present invention, and the white straight line is the final fitted pointer straight line.

FIG. 7 is a flow chart of determining the direction of the 0 scale line in an example of the present invention.

FIG8 is a template image used in an example of the present invention and the direction of the 0 scale line therein, wherein FIG8(a) is the template image used in an example of the present invention; FIG8(b) is the direction of the 0 scale line marked in the template image.

FIG9 is a result of feature point matching between a template image and an instrument image in this example.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings:

The following embodiment is for automatic reading of pointer-type water meters, and the complete process is shown in Figure 1. The test set in the embodiment consists of 145 samples, which are divided into three categories, namely 45 meter images imaged under uniform illumination conditions [as shown in Figure 2(a)], 45 meter images imaged under uneven illumination conditions [as shown in Figure 2(b)], and 45 meter images with blurred and dim imaging [as shown in Figure 2(c)].

Embodiment 1:

This embodiment provides a method for automatic reading of a pointer-type instrument that is robust to illumination, comprising the following steps:

Step A: preprocessing the instrument image to remove useless information in the instrument image;

Step B: based on the image preprocessed in step A, detecting the outline of the pointer in the instrument and determining the direction of the pointer;

Step C, based on the image preprocessed in step A, detecting the direction of the 0 scale line in the instrument;

Step D, complete the reading.

Embodiment 2:

In this embodiment, based on the embodiment 1, the implementation process of step A is shown in FIG3 , and the specific processing process is as follows:

Step A1, using Huffman circle detection to extract the ROI area of the instrument image;

Step A2, based on the characteristics of the circle, the instrument image is operated to eliminate useless information outside the ROI area; the specific method is: first, all areas outside the minimum circumscribed matrix of the ROI area on the instrument image are cropped; then, the characteristics of the circle are used to determine whether the point q(i, j) in the minimum circumscribed matrix of the ROI area is within the ROI area; if g(i, j) is outside the ROI area, the pixel value p(i, j) of the point is set to 255 to obtain the ROI image;

Step A3, enhancing the pointer information in the ROI area based on the color difference transformation; specifically: first grayscale the ROI image to obtain a grayscale image, and then perform an R-Y operation on the grayscale image to obtain an RY grayscale image; the pixel value Y of each pixel in the grayscale image is:

Y＝0.299*R+0.587*G+0.114*B

The pixel value R-Y of each pixel in the R_Y grayscale image is:

R-Y＝0.701*R-0.587*G-0.114*B

Among them, R, G and B are the three components of the pixel value of the corresponding pixel in the ROI image.

Embodiment 3:

In this embodiment, based on Embodiment 2, the specific processing process of step B is as follows:

Step B1, using the MSER algorithm to detect all maximum stable extreme value regions (MSER) on the R_Y grayscale image, and taking each maximum stable extreme value region (MSER) as a candidate pointer contour;

Step B2: matching the correspondence between each candidate outline of the pointer and the instrument pointer based on the improved NMS algorithm (ANMS algorithm), and determining the final outline of each pointer on the instrument;

Step B3: based on the final outline of each pointer on the instrument, determine the starting point s ( _xs , _ys ) and the end point t ( _xt , _yt ) of each pointer respectively, fit the straight line where each pointer is located, and obtain the direction of each pointer;

Embodiment 4:

In this embodiment, based on Embodiment 3, step B2 specifically includes the following steps:

Step B21, calculating the minimum circumscribed matrix area of all candidate pointer contours;

Step B22: For any two candidate contours of the pointer, calculate the area overlap IOU between their minimum circumscribed matrices:

Among them, _S1 and _S2 represent the areas of the minimum circumscribed matrices of the two candidate contours of the pointers, S represents the overlapping area of the minimum circumscribed matrices of the two candidate contours of the pointers, and the calculation formula of S is as follows:

和

取两个指针候选轮廓的最小外接矩阵的左上角坐标值中较大的坐标值；Among them, x _{right_min} and y _{right_min} take the smaller coordinate value of the lower right corner coordinate value of the minimum circumscribed matrix of the two pointer candidate contours.

and

Take the larger coordinate value of the upper left corner of the minimum circumscribed matrix of the two candidate contours of the pointer;

Step B23, setting the region overlap threshold IOU _th (IOU _th is an empirical parameter, and in this embodiment, IOU _th is set to 0.1);

For any two candidate pointer contours, compare their corresponding IOU and IOU _th. If the IOU is greater than IOU _th , they are considered to correspond to the same pointer on the instrument.

Step B24: For all candidate contours of the same pointer on the corresponding instrument, calculate the average value of their areas (the area of the candidate contour of the pointer can be taken as the number of pixels in the area it contains), and select the candidate contour of the pointer with the area closest to the average value as the final contour of the corresponding pointer on the instrument [as shown in Figure 6(a)].

Embodiment 5:

In this embodiment, based on Embodiment 3, step B3 specifically includes the following steps:

Step B31, select the area contained in the final outline of the pointer on the ROI image, that is, the centroid s( _xs , _ys ) in the pointer area as the starting point of the pointer, and use the grayscale centroid method to calculate the centroid s( _xs , _ys ) of the outline. [As shown in Figure 6(b), the white point is the starting point s( _xs , _ys )], the formula is as follows:

The formula f(u, v) represents the pixel value of the pixel at the coordinate (u, v) in the pointer region Ω on the ROI image.

Step B32: Determine the end point of the pointer based on the minimum angle method;

Assume that the final outline of a pointer consists of n points; for any three adjacent points among these n points, calculate the angle value corresponding to the middle point to obtain a series of angle values; compare all the calculated angle values, and the middle point corresponding to the minimum angle value is the end point t(x _t , y _t ) of the pointer [as shown in Figure 6(b) , where the black dot is the end point t(x _t , y _t )].

Suppose three adjacent points are A, B, and C, and the middle point is B. Calculate the size of ∠B in △ABC, that is, the angle value corresponding to the middle point B. The size of ∠B is calculated using the cosine theorem:

Embodiment 6:

In this embodiment, based on the embodiment 1, the implementation process of step C is shown in FIG7 , and the specific processing process is as follows:

Step C1, select a front image of the instrument as a template image [as shown in FIG8(a)], and manually mark the direction NM of the zero scale line on the template image [as shown in FIG8(b)];

Step C2, using the speeded up robust feature (SURF) technology to detect feature points on the template image and the instrument image (preprocessed instrument image, ROI image);

Step C3, matching the feature points on the template image and the instrument image [as shown in FIG9 ], and using the RANSC algorithm to remove mismatched points, to further ensure the accuracy of the matching.

Step C4: based on the direction of the 0 scale line on the template image and the matching feature points between the template image and the instrument image, obtain the direction of the 0 scale line on the instrument image.

Embodiment 7:

In this embodiment, based on Embodiment 6, step C4 specifically includes the following steps:

Step C41, assuming that there are k feature points on the template image that match the instrument image;

Step C42, randomly select two feature points p ( _xp , _yp ) and w ( _xw , _yw ) from the k matching feature points on the template image, _xp < _xw , to form a vector pw;

Step C43, calculate the angle θ between the vector pw on the template image and the 0 scale line:

Wherein, N(x _N , y _N ) and M(x _M , y _M ) are the starting point and end point of the 0 scale line on the template image respectively;

Step C44: Assume that the matching feature points on the instrument image corresponding to p( _xp , _yp ) and w( _xw , _yw ) on the template image are p′( _xp ′, _yp ′) and w′(xw _′ , _yw ′);

Find a straight line p′f′ in the instrument image with an angle of θ with p′w′, and use its direction as a candidate 0 scale line direction; let the coordinates of f′ be (x _f ′, y _f ′), and its calculation formula is:

Step C45, repeating steps C42 to C44 for T times (T is the optimal parameter obtained from the experiment, and T=6 is set in this embodiment), and ensuring that the two feature points randomly selected from the k matching feature points on the template image are different each time; through this operation, T candidate 0 scale line directions are found in the instrument image;

Step C46: adopt a voting strategy to determine a final 0 scale line direction from T candidate 0 scale line directions;

In the voting strategy, an angle threshold γ (γ is an empirical parameter, and γ=1 is set in this embodiment) is set to evaluate the accuracy of each candidate 0 scale line direction; for each candidate 0 scale line direction, its score value is equal to the number of other candidate 0 scale line directions whose angles with it are less than the angle threshold γ; the candidate 0 scale line direction with the highest score value is selected as the 0 scale line direction in the instrument image.

Embodiment 8:

In this embodiment, based on the embodiment 1, the specific processing process of step D is as follows:

Step D1, calculate the angle δ between the pointer direction and the 0 scale line direction, the calculation formula is:

Among them, v1 is the direction vector of the pointer, and v2 is the direction vector of the 0 scale line;

Step D2, determine the meter reading according to the angle δ between the pointer direction and the 0 scale line direction.

其中δ₁、δ₂、δ₃、δ₄分别为四个子表盘中指针方向与0刻度线方向的夹角。In this embodiment, the pointer water meter has four sub-dials, each with a pointer. The scale values in each sub-dial have 10 levels, each value is evenly distributed, the angle between each scale value is 36 degrees, and the scale value units in the four sub-dials are 0.1, 0.01, 0.001, and 0.0001 respectively. The final reading of the pointer can be determined by the above scheme.

Wherein δ ₁ , δ ₂ , δ ₃ , and δ ₄ are the angles between the pointer directions and the 0 scale line directions in the four subdials, respectively.

Embodiment 9:

The present embodiment provides a pointer-type instrument automatic reading device that is robust to illumination, including a processor. The processor adopts the above-mentioned pointer-type instrument automatic reading method to realize the pointer-type instrument automatic reading.

Embodiment 10:

In this embodiment, based on the ninth embodiment, the pointer-type instrument automatic reading device further includes an image acquisition module for acquiring instrument images; and the processor completes the pointer-type instrument automatic reading based on the instrument images acquired by the image acquisition module.

The effect of the present invention is verified by using samples in the test set. The results show that the present invention can accurately complete the pointer detection and reading results for instrument images imaged under uniform lighting conditions, instrument images imaged under uneven lighting conditions, and 45 blurred and dim instrument images. The accuracy of the pointer detection and reading results is significantly better than the most advanced existing methods. The method of the present invention can simultaneously meet the requirements of robustness, accuracy, and the need to control a large amount of auxiliary information such as the lighting conditions and imaging angles of the image, and has strong practicality.

Claims (7)

1. A method for automatic reading of a pointer-type instrument robust to illumination, characterized in that it comprises the following steps: Step A, preprocessing the instrument image; Step B: for the instrument image preprocessed in step A, detecting the pointer contour therein and determining the pointer direction; Step C: for the instrument image preprocessed in step A, detecting the direction of the 0 scale line therein; Step D, complete reading; The specific processing process of step A is as follows: Step A1, using the Huffman circle detection algorithm to extract the ROI area of the instrument image; Step A2, firstly cut out all the areas outside the minimum circumscribed matrix of the ROI area on the instrument image; then for each pixel point q(i, j) in the minimum circumscribed matrix of the ROI area, determine whether it is within the ROI area; if q(i, j) is outside the ROI area, set its pixel value p(i, j) to 255, thereby obtaining the ROI image; Step A3: First, grayscale the ROI image to obtain a grayscale image, and then perform an R-Y operation on the grayscale image to obtain an R_Y grayscale image; the pixel value Y of each pixel in the grayscale image is: Y＝0.299*R+0.587*G+0.114*B The pixel value R-Y of each pixel in the RY grayscale image is: R-Y＝0.701*R-0.587*G-0.114*B Among them, R, G and B are the three components of the pixel value of the corresponding pixel in the ROI image; The specific processing process of step B is as follows: Step B1, using the MSER algorithm to detect all maximum stable extreme value regions on the R_Y grayscale image, and taking each maximum stable extreme value region as a candidate contour of a pointer; Step B2: matching the corresponding relationship between each candidate outline of the pointer and the instrument pointer based on the improved NMS algorithm, and determining the final outline of each pointer on the instrument; Step B3, based on the final outline of each pointer on the instrument, determine the starting point and the end point of each pointer respectively, fit the straight line where each pointer is located, and obtain the direction of each pointer; The step B2 specifically comprises the following steps: Step B21, calculating the minimum circumscribed matrix area of all candidate pointer contours; Step B22: For any two candidate contours of the pointer, calculate the area overlap IOU between their minimum circumscribed matrices:

Among them, _S1 and _S2 represent the areas of the minimum circumscribed matrices of the two candidate contours of the pointers, S represents the overlapping area of the minimum circumscribed matrices of the two candidate contours of the pointers, and the calculation formula of S is as follows: S＝(x _{right_min} -x _{left_max} )*(y _{right_min} -y _{left_max} ) Among them, x _{right_min} and y _{right_min} take the smaller coordinate value of the lower right corner coordinate value of the minimum circumscribed matrix of the two candidate contours of the pointer, and x _{left_max} and y _{left_max} take the larger coordinate value of the upper left corner coordinate value of the minimum circumscribed matrix of the two candidate contours of the pointer; Step B23, setting the region overlap threshold _IOUth ; For any two candidate pointer contours, compare their corresponding IOU and IOU _th. If the IOU is greater than IOU _th , they are considered to correspond to the same pointer on the instrument. Step B24: For all candidate contours of the same pointer on the corresponding instrument, calculate the average value of their areas, and select the candidate contour of the pointer with the area closest to the average value as the final contour of the corresponding pointer on the instrument. 2. The illumination-robust pointer instrument automatic reading method according to claim 1, characterized in that the step B3 specifically comprises the following steps: Step B31, for each pointer on the instrument, select the area contained in the final outline of the pointer on the ROI image, that is, the center of gravity in the pointer area as the starting point of the pointer; Step B32: for each pointer on the instrument, determine the end point of the pointer based on the minimum angle method; Assume that the final outline of a pointer is composed of n pixels; for any three adjacent pixels among the n pixels, calculate the angle value corresponding to the middle pixel point, so as to obtain a series of angle values; compare all the angle values, and the middle pixel point corresponding to the minimum angle value is the end point of the pointer; the angle value corresponding to the middle pixel point is calculated as follows: suppose the three adjacent pixels are A, B, and C, and the middle pixel point is B, calculate the size of ∠B in ΔABC, that is, the angle value corresponding to the middle pixel point B. 3. The illumination-robust pointer instrument automatic reading method according to claim 1, characterized in that the specific processing process of step C is as follows: Step C1, selecting a front image of an instrument as a template image, and manually marking the direction of the 0 scale line on the template image; Step C2, using accelerated robust feature technology to detect feature points on the template image and the instrument image respectively; Step C3, matching the feature points on the template image and the instrument image, and using the RANSC algorithm to remove mismatched points; Step C4: based on the direction of the 0 scale line on the template image and the matching feature points between the template image and the instrument image, obtain the direction of the 0 scale line on the instrument image. 4. The illumination-robust pointer instrument automatic reading method according to claim 3, characterized in that the step C4 specifically comprises the following steps: Step C41, assuming that there are k feature points on the template image that match the instrument image; Step C42, randomly select two feature points p ( _xp , _yp ) and w ( _xw , _yw ) from the k matching feature points on the template image, _xp < _xw , to form a vector pw; Step C43, calculate the angle θ between the vector pw on the template image and the 0 scale line:

Wherein, N(x _N , y _N ) and M(x _M , y _M ) are the starting point and end point of the 0 scale line on the template image respectively; Step C44: Assume that the matching feature points on the instrument image corresponding to p( _xp , _yp ) and w( _xw , _yw ) on the template image are p′( _xp ′, _yp ′) and w′(xw _′ , _yw ′); Find a straight line p′f′ in the instrument image with an angle of θ with p′w′, and use its direction as a candidate 0 scale line direction; let the coordinates of f′ be (x _f ′, y _f ′), and its calculation formula is:

Step C45, repeating steps C42 to C44 T times, and ensuring that the two feature points randomly selected from the k matching feature points on the template image are different each time; through this operation, T candidate 0 scale line directions are found in the instrument image; Step C46: adopt a voting strategy to determine a final 0 scale line direction from T candidate 0 scale line directions; In the voting strategy, an angle threshold γ is set to evaluate the accuracy of each candidate 0 scale line direction; for each candidate 0 scale line direction, its score value is equal to the number of other candidate 0 scale line directions whose angles with it are less than the angle threshold γ; the candidate 0 scale line direction with the highest score value is selected as the 0 scale line direction in the instrument image. 5. The illumination-robust pointer instrument automatic reading method according to claim 1, characterized in that the specific processing process of step D is as follows: Step D1, calculate the angle δ between the pointer direction and the 0 scale line direction on the instrument image, and the calculation formula is:

Among them, v1 is the direction vector of the pointer on the instrument image, and v2 is the direction vector of the 0 scale line on the instrument image; Step D2, determining the meter reading according to the angle δ between the pointer direction and the 0 scale line direction on the meter image. 6. An automatic reading device for pointer-type instruments that is robust to illumination, comprising a processor, wherein the processor implements automatic reading of pointer-type instruments using the method described in any one of claims 1 to 5. 7. The lighting-robust pointer-type instrument automatic reading device according to claim 6 comprises a processor, and is characterized in that it also comprises an image acquisition module for acquiring instrument images; the processor completes the automatic reading of the pointer-type instrument based on the instrument image acquired by the image acquisition module.

Images