CN105740829A - Scanning line processing based automatic reading method for pointer instrument - Google Patents

Abstract

The invention discloses an automatic reading method of a pointer instrument based on scanning line processing, which includes the steps of: firstly, using a single-scale Retinex algorithm to perform illumination processing on the original image, then binarizing the image, and then extracting Feature pixel point is carried out, Hough transformation is carried out to feature pixel point to detect pointer straight line, adopt angle method to calculate reading at last; Described method based on line scanning line processing refers to scanning each line in the image, extracting the continuous segment in this line , search for a continuous segment whose length is less than or equal to the width of the pointer line, and use the midpoint of the continuous segment as a feature pixel. The present invention can reduce the impact of non-uniform illumination on subsequent algorithms by adopting the Retinex algorithm, and can reduce the number of pixels participating in the Hough transform by adopting a method based on row scanning line processing, thereby reducing the amount of calculation.

Description

An automatic reading method of pointer instrument based on scanning line processing

technical field

The invention relates to the field of image processing research, in particular to an automatic reading method of a pointer meter based on scanning line processing.

Background technique

Due to the advantages of simple structure, low price and easy maintenance, pointer instruments are still widely used in electricity, transportation and residents' lives. The accuracy and speed of manual readings are easily affected by their own factors. Especially in the face of the problem of instrument reading in some harsh environments where people are not suitable for approaching, an automatic instrument reading technology is needed to replace manual reading. Because machine vision has more advantages than human physiological vision, it is more accurate, objective and stable, so image recognition technology has become an important application means for automatic reading of pointer instruments.

Most of the automatic reading algorithms of pointer instruments based on image recognition are carried out in an ideal environment, and are generally implemented in three steps: the first step is pointer extraction, and the implementation methods include subtraction method, color space method, threshold method, Edge extraction method, etc.; the second step is pointer positioning, and the implementation method includes using Hough transform to detect pointers, etc.; the third step is reading calculation, and the implementation methods include angle method and distance method. However, in practical applications, factors such as uneven illumination or occlusion have a serious impact on the accuracy and speed of readings, and the ideal results cannot be obtained by directly applying the above algorithm.

Aiming at uneven illumination, many literatures have also proposed a solution algorithm, such as Li Xuecong, Wang Renhuang and others proposed a six-step preprocessing method for pointer instrument images in the journal "Electrical Measurement and Instrument" in 2012; Wang Zhimin, Wang Renhuang in 2014 in "Research on Image Preprocessing and Segmentation of Pointer Instruments" published in the journal "Industrial Control Computer" mentioned the use of homomorphic filtering technology to deal with uneven illumination. Although the above methods can solve the problem of certain illumination changes, they have a large amount of computation and are not suitable for occasions with high real-time requirements.

In addition, for the problem that the Hough transform itself has a relatively large amount of calculation, there are also many literatures that propose a solution algorithm. For example, Tao Bingjie, Han Jiale, etc. published "A Practical Pointer Instrument Reading Recognition Method" in the "Optoelectronic Engineering" magazine in 2011. This method proposes to use a double-threshold Hough transform, and only the pixels within the angle range of the pointer are processed. Transformation; "Improvement of Hough Transformation and Its Application in Pointer Water Meter Recognition Process" published by Li Jing, Zhang Ning, etc. in 2015, this method proposes to transform the pixel points in the water meter whose gray value is within a certain range; Duan Rujiao et al. proposed a fast line detection algorithm based on the improved Hough transform. The algorithm first clusters the foreground pixels, then performs perceptual grouping and subdivides them into many small straight line segments, and finally uses random Hough transform to detect the straight line segments. To a certain extent, these methods can reduce the number of pixels for Hough transform and improve the detection speed, but the detection accuracy is still not ideal.

For this reason, it is of great application value to study an automatic reading algorithm of a pointer instrument with high reading precision and fast reading speed.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and provide an automatic reading method for pointer instruments based on scanning line processing. The algorithm can still achieve high precision and high reading speed under non-uniform illumination, and the reading speed adjustable.

The object of the present invention is achieved through the following technical solutions: a method for automatic reading of pointer meters based on scanning line processing, comprising the steps of: firstly adopting a single-scale Retinex algorithm to carry out illumination processing on the original image, then binarizing the image, and then based on The method of row scanning line processing extracts feature pixels, performs Hough transform on feature pixels to detect pointer straight lines, and finally adopts angle method to calculate readings; the method based on row scan line processing refers to scanning each row in the image, Extract the continuous segment in the row, search for the continuous segment whose length is less than or equal to the width of the pointer line, and use the midpoint of the continuous segment as the feature pixel. The influence of non-uniform illumination on subsequent algorithms can be reduced by adopting the Retinex algorithm, and the number of pixels participating in the Hough transform can be reduced by adopting the method based on row scanning line processing, thereby reducing the amount of calculation.

Preferably, the step of using the single-scale Retinex algorithm to perform illumination processing on the original image is:

r(x,y)=logR(x,y)

$<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}\frac{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ;</span>$

≈logi(x,y)-log(F(x,y)*i(x,y))

Among them, r(x,y) is the image processed by single-scale Retinex algorithm, F(x,y)=λexp(-(x ² +y ² )/σ ² ), ∫∫F(x,y)dxdy= 1, is a Gaussian function, i(x,y) is the gray value at the image position (x,y), R(x,y), L(x,y) are the image at point (x,y) The reflection component and the illumination component at .

Preferably, the image binarization adopts the method of maximum between-class variance (OTSU).

Preferably, the following operations are performed after image binarization: perform morphological processing, connect the pointers of the breakpoints, and then perform connected domain analysis. When the length and width of the bounding rectangle of the connected domain are less than a certain threshold, set the connected domain As the background part, the numbers, symbols and text on the instrument are filtered out. In this way, irrelevant data is filtered out as much as possible before pointer recognition, which further reduces the calculation amount of subsequent algorithms and improves the processing speed.

Preferably, feature pixel points are extracted based on the method of row scanning line processing, and the specific steps are as follows:

(1-1) Define binary image size as h×w, initialize variable i=0, set , the value of the foreground pixel is 1; according to the image resolution, calculate the number of pixels l _zhizhen occupied by the width direction of the pointer line;

(1-2) Scan the i-th row of pixels in the image, i∈[1,2,3...h], regard the continuous foreground pixels as a line segment, set a total of k segments, and mark them as b _j , j=1,2,3...k, and record the number of pixels in each segment as l _j ,l _j ∈[1,2,3...w];

(1-3) Compare the number of pixels l _j of each segment with l _zhizhen , if l _j ≤ l _zhizhen , save the coordinates of the midpoint of the line segment j in the original binary image to the set H;

(1-4) Determine whether i is equal to h, if not, set i=i+1, and return to step (1-2), otherwise, perform Hough transform on the coordinates in the set H to detect a straight line.

The foreground block of the non-pointer part can be removed by adopting the row scanning line processing method, such as a large-area foreground block caused by uneven illumination or occlusion shadows. Only the midpoint of the line segment that conforms to the pointer line width is reserved as the foreground point for line detection, and the scanning process is simple and fast. After being processed by the above algorithm, the number of foreground pixels for Hough transform is greatly reduced, thereby improving the detection speed of straight lines.

Furthermore, in the step (1-4), n is set to represent the number of row and column intervals for scanning line processing, and after the i-th row is scanned, i+n rows are scanned next. The straight line detection speed can be adjusted by setting n. The larger n is, the less time is required for the Hough transformation, that is, the faster the straight line detection speed, so the calculation speed of the algorithm can be further improved.

Preferably, the steps of using the angle method to calculate the reading are:

(2-1) Set the range of the pointer as [a, b], the pointer swings from left to right, the angle between the smallest scale is θ ₁ , the largest scale is θ ₂ , Z is the reading value of the instrument, ε is the current pointer The included angle with the negative direction of the x-axis, and an xoy coordinate system is established with the point around which the pointer swings as the coordinate center;

(2-2) For instruments with uniform scale, there is a proportional formula:

$\frac{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}}{\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; ;</span>$

Therefore reading $Z=\frac{\mathrm{\&epsiv;}-{\mathrm{\&theta;}}_1}{{\mathrm{\&theta;}}_2-{\mathrm{\&theta;}}_1}\&CenterDot;(b-a)+a.$

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. The present invention uses a single-scale Retinex algorithm to preprocess the pointer image, so that the automatic reading algorithm is not affected by changing illumination. In the case of dim light or mirror reflection, it can also obtain good recognition effect, and the recognition speed is fast, completely Meet real-time requirements.

2. The present invention uses methods such as connected domain analysis and scan line processing to reduce the number of foreground points in Hough transform, quickly extract feature pixels for Hough transform, reduce the number of pixels participating in Hough transform, and innovatively refine and discrete the pointer image Combined with Hough transform, etc., the pointer detection time is greatly reduced. At the same time, the reading speed can be adjusted, so that the detection speed of the pointer line can be more than 10 times faster than the traditional algorithm, which is very practical.

Description of drawings

Fig. 1 is the flowchart of the method of the present invention.

Figure 2(a) is the pointer image before lighting processing.

Figure 2(b) is the effect image after preprocessing with the single-scale Retinex algorithm.

Figure 3 is a schematic diagram of the structure of a straight line.

Fig. 4 is a schematic diagram of the geometric relationship obtained according to the straight line structure.

FIG. 5 is a schematic diagram of scanning processing of pixel gray values of a row of pixels in this embodiment.

Figure 6(a) is a binary image of a pointer.

FIG. 6( b ) is a structure diagram after performing all row scanning processing on the binary image shown in FIG. 6( a ).

Figure 7(a) is an example diagram of a straight line detection.

Figures 7(b)-(d) are respectively for Figure 7(a), when n=1, 5, 10, the binary image is scanned and processed, corresponding to the distribution diagram of the small grid count variable value in the parameter space.

Fig. 8 is a schematic diagram of the coordinate system established according to the instrument in this embodiment.

Figure 9 is the effect diagram obtained by using different pretreatment methods.

Fig. 10 is a pointer detection speed test image.

detailed description

The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example 1

Referring to Fig. 1, a method for automatic reading of pointer instruments based on scanning line processing in this embodiment can be mainly divided into two parts, the first part is pointer image preprocessing, and the second part is pointer detection and reading, as follows The methods involved in each part are described in detail with reference to the accompanying drawings.

1. Pointer image preprocessing

During the image acquisition process, due to uneven ambient light (such as being blocked by foreign objects) or local reflections caused by the glass lens of the instrument itself, the brightness of the collected image will also be uneven, which will seriously affect the subsequent images. accurate segmentation. This embodiment uses the single-scale Retinex algorithm to perform illumination processing on the original image, then binarizes the image, then performs morphological processing, connects the pointers of breakpoints, and finally performs connected domain analysis to filter out numbers and symbols on the instrument ,Word.

1.1 Using single-scale Retinex algorithm for lighting processing

Assuming that the gray value of the pixel at point (x, y) in the meter image is i(x, y), according to the Lambertian illumination model, there is

i(x,y)=R(x,y) L(x,y)(1)

where R(x,y) and L(x,y) are the reflection component and illumination component of the image at point (x,y), respectively. Since the illumination component L(x,y) belongs to the low-frequency component of the image and changes slowly, it is generally considered smooth, that is

L(x+Δx,y)≈L(x,y)(2)

L(x,y+Δy)≈L(x,y)(3)

According to the Retinex theory, the incident light and the reflected object are the two main factors that constitute the imaging of the human eye, while the illumination part belongs to the low frequency part, and the reflection part belongs to the high frequency part. The product of the two is the image brightness perceived by the human eye imaging, such as the formula ( 1) as shown. The low-frequency component L(x,y) can be estimated by filtering the image with low-frequency filtering technology, that is, L(x,y)≈F(x,y)*i(x,y). Therefore, the single-scale Retinex algorithm can be expressed by the following formula:

r(x,y)=logR(x,y)

$<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}\frac{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

≈logi(x,y)-log(F(x,y)*i(x,y))

Where F(x,y)=λexp(−(x ² +y ² )/σ ² ), ∫∫F(x,y)dxdy=1. F(x,y) is the environmental function, here we choose Gaussian function as the environmental function, i(x,y) is the gray value of the image position (x,y), R(x,y), L(x,y) are the reflection component and the illumination component respectively; r(x, y) is the image processed by the single-scale Retinex algorithm. It can be seen from the above formula that the processed image is the logarithmic form of the reflection component, which has nothing to do with the illumination component, so the obtained image has the characteristics of illumination expression invariance. For the original image in Figure 2(a), the effect after preprocessing with the single-scale Retinex algorithm is shown in Figure 2(b).

1.2 Binarization processing

Binarization is performed on the image after illumination processing, and the purpose is to segment the pointer image. The OTSU method is the most commonly used adaptive binarization technology. Its basic idea is to find a threshold in all gray levels, which can divide the image into two parts, the foreground and the background with the largest variance.

Assuming that the gray value t is the threshold for foreground and background segmentation, the proportion of foreground points in the image is ω ₀ , and the average gray level is μ ₀ ; the proportion of background points in the image is ω ₁ , and the average gray level is μ ₁ . Then the total average gray level of the image is μ and the variance g of the foreground and background images is shown in (5) and (6) respectively:

μ＝ω ₀ ·μ ₀ +ω ₁ ·μ ₁ (5)

g＝ω ₀ ·(μ ₀ -μ) ² +ω ₁ ·(μ ₁ -μ) ² (6)

When g in formula (6) attains the maximum value, the corresponding segmentation threshold t is the threshold value sought by the OTSU method.

1.3 Connected domain analysis

After using the OTSU method to binarize the image, first perform morphological processing, connect the pointers of the breakpoints, and then perform connected domain analysis. When the length and width of the bounding rectangle of the connected domain are less than a certain threshold, set the connected domain as the background part , to filter out numbers, symbols, text, etc. on the meter.

2. Pointer detection and reading

In this embodiment, feature pixels are extracted based on the method of row scanning line processing, Hough transform is performed on the feature pixels to detect the pointer straight line, and finally the angle method is used to calculate the reading. The above binarization process has divided the instrument image into foreground part and background part, the foreground part is the pointer candidate part, and the detection of the pointer is essentially the straight line detection of the foreground image. Hough transform is widely used in instrument pointer detection. According to its detection principle, this paper proposes a fast detection method of instrument image pointer based on scan line processing.

2.1 Hough transform detection line principle

The basic idea of Hough transform is the duality of point and line, that is, the problem in the image space is converted to the parameter space to solve it. The parametric equation of the line:

ρ=x·cosθ+y·sinθ(7)

In the parameter space, x and y are constants (here only the case of not being 0 at the same time is discussed), and the formula (7) can be changed to

$\begin{array}{c}\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; x</span>}}{\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; the\; y</span>}}{\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

in

It can be known from (7) (8) that the point (x, y) in the image space corresponds to the sinusoidal curve in the θ-ρ space, and the point in the θ-ρ space corresponds to the straight line in the image space. Therefore, it is easy to obtain the detection line method: in the θ-ρ space, the According to a certain step length, it is divided into many small grids (w, h are the width and height of the picture respectively), each small grid is assigned a count variable, and each time the sinusoidal curve passes through the small grid, the value of the corresponding small grid count variable is increased by 1, The value of the small grid count variable corresponds to the number of collinear pixels in the image space, and the θ, ρ in the small grid center coordinates (θ, ρ) whose count variable value is greater than a certain threshold are substituted into (7) to obtain the xy space The equation of the line that is the line you are looking for.

2.2 Extract feature pixels based on row scan line processing method

From Section 2.1, it can be seen that the calculation amount of the Hough transform detection line itself is quite large, and the calculation amount increases linearly with the increase of the number of transformed pixels. Therefore, reducing the number of pixels for Hough transform is an effective method to improve the detection of straight lines. The method of the invention performs row scanning processing on the image, uses the midpoint of the line segment within the pointer width range as the foreground point of line detection, and rapidly reduces the number of pixels participating in the Hough transformation, thereby reducing the amount of calculation.

In order to facilitate the description of the principle, some concepts are defined as follows:

Definition: A straight line segment with a width (not a single pixel width) is called a straight line, a single-pixel wide edge line on both sides of the straight line is called a straight line edge line, and a point with the same distance from the two edge lines is called a feature of a straight line Points, and the line formed by the feature points is called the feature line of the line bar. The straight line structure is shown in Figure 3. In the figure, 1 represents the edge line, 2 represents the feature point, and 3 represents the feature line. It can be seen from Figure 3 that the two edge lines of the same straight line are parallel to each other, the feature line is the line expected to be found, and the feature points are selected as the feature points for Hough transformation. For feature points, it can be determined by the following theorem.

Theorem 1: Assume that the two edge straight lines l ₁ , l ₂ of a straight line intersect with a horizontal line (or a vertical line) at two points A and B, and a vertical line ( or a horizontal line), and intersect with straight line l ₁ , l ₂ at C and D respectively, then point O is also the midpoint of the line connecting C and D, and point O is a feature point located on the feature line of the straight line .

Proof: The known conditions are shown in Figure 4. Because the two edge lines l ₁ and l ₂ of the same straight line are parallel, ∠OCA and ∠ODB are their internal alternate angles, so:

∠OCA＝∠ODB(9)

The straight line AB and the straight line CD are perpendicular to each other, and O is the midpoint of the line segment AB, so the following formula holds:

∠AOC＝∠BOD＝π/2(10)

AO=BO(11)

To two congruent right triangles:

$\mathrm{<span\; class="notranslate">\; \&Delta;\; AOC</span>}<span\; class="notranslate">\; \&cong;</span>\mathrm{<span\; class="notranslate">\; \&Delta;BOD</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 12</span>}<span\; class="notranslate">\; )</span>$

From (12) formula: OC=OD, the distance from point O to AC is equal to the distance from O to BD. According to the definition, the point O is the midpoint of the line segment CD, which is also the characteristic point of the straight line, and is located on the characteristic straight line of the straight line. Certificate completed

The included angle between the pointer of the meter used in this embodiment and the negative direction of the x-axis is within the range of 45° and 135°, so the present invention adopts row scanning processing. The principle of scanning the pixel gray value of a row of pixels is shown in FIG. 5 . In Figure 5, the pixel with a pixel gray value of 1 is the foreground point, otherwise it is the background point, and x and y represent the pixel position and pixel gray value respectively. The width of the foreground part in the middle is too large and is not within the range of the pointer width, and the width of the foreground pixels on the other two sides is within the range of the pointer width. After scanning line processing, only the middle pixels are extracted as Hough transform pixels. Figure 6(b) shows the result after all line scanning processing of the binary image shown in Figure 6(a). It can be seen from FIG. 6 that this scanline processing algorithm can remove non-pointer foreground blocks, such as large-area foreground blocks caused by uneven illumination or occlusion shadows. In addition, only the midpoint of the line segment conforming to the line width of the pointer is reserved as the foreground point for line detection, and the scanning process is simple and fast. After being processed by the proposed algorithm, the number of foreground pixels for Hough transform is greatly reduced, thereby improving the detection speed of straight lines.

2.3 Control of linear detection speed

According to the principle of Hough detection of straight lines, it is known that the parameters of the straight line in the image space are obtained by comparing the count values on the small grids in the parameter space. It can be imagined that if the values of counting variables of all the small cells in the parameter space are reduced by the same value at the same time, which does not change their size relationship, then the parameters ρ, θ corresponding to the small cell with the largest count value are returned unchanged, and also That is, they are mapped to the image space as the same straight line.

In fact, as long as the binarized image is scanned at equal intervals, it is equivalent to reducing the value of the small variable in the parameter space at the same time. By adjusting the number of intervals of line scanning to further change the number of pixels of Hough transform, and then make the detection speed of the straight line controllable. The double-threshold Hough transform essentially selects the pixels between the minimum angle and the maximum angle of the pointer to perform Hough transform. The specific steps of the double-threshold Hough transform fast detection line algorithm based on scan line processing are as follows:

(1) Input a binary image whose size is h×w, initialize variable i=0, set H=φ; calculate the number of pixels l _zhizhen occupied by the pointer line width direction according to the image resolution;

(2) Scan the pixels in the i-th row of the image, i∈[1,2,3...h], regard the continuous foreground pixels as a line segment, assuming there are k segments in total, mark them as b _j , j=1,2 ,3...k, and record the number of pixels in each segment as l _j ,l _j ∈[1,2,3...w];

(3) Compare the number of pixels l _j in each segment with l _zhizhen , if l _j ≤ l _zhizhen , save the coordinates of the midpoint of the line segment j in the original binary image to the set H;

(4) i=i+n, if i<h, return to step (2), otherwise enter step (5);

(5) Perform double-threshold Hough transform on the coordinates in the set H to detect straight lines.

Among them, n represents the number of row and column intervals processed by the scan line, and is a parameter for adjusting the detection speed of a straight line. When n=1, 5, 10, the distribution of the small grid count variable value in the corresponding parameter space of the binary image Fig. 7(a) after scanning is shown in Fig. 7(b), (c), and (d). Among them, when n=1, 5, 10, the time required for the Hough transformation to determine the parameters of the straight line is 7.83ms, 2.55ms, and 1.86ms respectively. The small grid step size of parameter space ρ, θ is 1, and the small grids in the statistical range are arranged in a row. It can be seen from Figure 7 that the distribution shape of the small grid count variable values has a very high similarity. The larger the scanning interval n is, the less time is required for Hough transformation, that is, the faster the detection speed of the straight line, the smaller the small grid count variable value is at the same time. , the maximum value of the small cell count variable corresponds to the same small cell or nearby small cells, which is consistent with the previous analysis.

2.4 Reading Calculation

Since the angle range of the straight line detected by the Hough transform is 0 to 180 degrees, the angles between the minimum scale and the maximum scale of the instrument pointer used in this embodiment and the negative direction of the x-axis are 45 degrees and 135 degrees, so this embodiment only considers the pointer deflect within this range. Suppose the range of the pointer is [a,b], the pointer swings from left to right, the angle between the minimum scale is θ ₁ , and the maximum scale is θ ₂ , Z is the reading value of the meter, and the point around which the pointer swings is An xoy coordinate system is established for the coordinate center, as shown in Fig. 8 .

For an instrument with uniform scale, the proportional formula can be obtained from Figure 7:

$\frac{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}}{\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; )</span>$

Therefore reading $Z=\frac{\mathrm{\&epsiv;}-{\mathrm{\&theta;}}_1}{{\mathrm{\&theta;}}_2-{\mathrm{\&theta;}}_1}\&Center\; Dot;(b-a)+a---\left(14\right)$

3. Compared with existing technology

In order to verify the robustness of the method of the present invention to illumination, the classic histogram equalization and grayscale transformation with faster processing speed are selected for comparison with the method of the present invention; in order to verify the effect of pointer detection speed, the canny operator and skeleton extraction method are selected to be compared with the method of the present invention Invented method for comparison.

This experiment is carried out in Pentium(R)Dual-CoreCPUE58003.2GHz memory 1.96GB, VS2010+Opencv4.8 environment.

During the experiment, it is the premise that the pointer can be detected. After the parameters are tested, the better parameters are selected. The skeleton extraction is the thinning technology. The parallel fast thinning algorithm proposed by T.Y.zhang et al. in the paper "Afastparallelalgorithmforthinningdigitalpatterns" The transformation uses the logarithmic transformation y=log(i+1)/b+a, i is the gray value of the pixel, a=10, b=0.3. Among them, in this paper, the step size of the small lattice θ∈[0°,180°) of the Hough transform is 1°, and the step size of ρ is 1 pixel unit; the parameter σ=2.5 of the low-pass filter in Retinex, the pointer image width threshold is 10 pixels wide.

3.1 Illumination robustness verification

The image processing results obtained in three different environments using three different lighting processing methods are shown in Figure 9. It can be seen from Figure 9 that the histogram equalization processing effect is the worst. Although the logarithmic transformation can perform grayscale dynamic compression, it is not ideal to deal with the illuminated areas with high contrast. After processing, there are still obvious differences between occluded and non-occluded areas. The contrast is not conducive to the accurate extraction of the pointer image. The single-scale Retinex method proposed by the present invention can handle the shadow part well, realize image enhancement, reduce the gray scale difference between the shadow part and the non-shadow part, thereby eliminating the influence of light.

Comparing the calculated results of the invented algorithm with the results obtained by manual visual inspection, some experimental data shown in Table 1 can be obtained:

Table 1 Invention Algorithm Calculation Results and Manual Visual Inspection Results Data

visual inspection reading value error 73 73.500 0.500 101 101.277 0.277 108 108.222 0.222 172 171.722 0.288 190 190.166 0.166

Wherein the last digit in the visual value is an artificial estimated value. It can be seen from Table 1 that the present invention is very accurate in positioning and reading of the pointer, and it can be seen that the algorithm used has quite good robustness to illumination.

3.2 Comparison of reading speed

After the instrument image is removed from the influence of illumination changes, it is subjected to binarization processing. In order to reduce the pixels for Hough transformation, the algorithm based on the line scan processing described in the present invention, the parallel fast thinning algorithm, and the Canny algorithm are respectively used for binarization. Image processing to compare the time required for reading. Assuming that the preprocessed binary image is shown in Figure 10, with a size of 640×253, test the time taken by different algorithms to extract Hough transform feature points and confirm the parameters of the straight line. The test results are shown in Table 2, where n is the scan interval .

Table 2 Reading time of different processing methods (unit: ms)

It can be drawn from Table 2 that the scanning line processing time of the present invention is the shortest, the larger the scanning interval n, the shorter the detection time, and when n=10, the detection speed is 15 times that of the Canny operator and 11 times that of the classic thinning algorithm . Here, the purpose of the Canny operator, the classical thinning algorithm and the algorithm of the present invention is to reduce the number of pixels for Hough transformation. The algorithm of the present invention extracts the Hough transformation pixel points, the process is simple, and the amount of calculation is small; and the refinement is the refinement of the two-dimensional image, and the Canny operator is a commonly used edge extraction algorithm, and they themselves have a large amount of calculation. The number of transformed pixels is also much larger than the proposed algorithm and is not adjustable. Therefore, the algorithm of the present invention is more superior.

To sum up, the method described in this embodiment uses single-scale Retinex to eliminate the influence of illumination, uses technologies such as connected domain analysis and scan line processing to reduce the number of foreground points in Hough transform, and innovatively refines and discretizes the pointer image with Hough The experiment proves that the proposed algorithm not only has good illumination robustness and high detection speed, but also has the advantages of adjustable speed.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (7)

1. the pointer instrument automatic reading method processed based on scanning line, it is characterized in that, including step: initially with single scale Retinex algorithm, original image is carried out photo-irradiation treatment, then by image binaryzation, the method being then based on horizontal scanning line process extracts feature pixel, feature pixel is carried out Hough transform detection pointer straight line, finally adopts preset angle configuration to calculate reading；The described method processed based on horizontal scanning line refers to each row in image is scanned, and extracts the continuous segment in this row, and the length of search continuous segment is less than or equal to the continuous segment of cursor line width, using the line segment midpoint of this continuous segment as feature pixel.

2. pointer instrument automatic reading method according to claim 1, it is characterised in that the step that original image is carried out photo-irradiation treatment by described employing single scale Retinex algorithm is:

$\begin{array}{c}r\left(x,y\right)=\log R\left(x,y\right)\\ =\log \frac{i\left(x,y\right)}{L\left(x,y\right)}\\ \&ap;\log i\left(x,y\right)-\log \left(F\left(x,y\right)*i\left(x,y\right)\right)\end{array};$

Wherein, r (x, the image after y) processing for single scale Retinex algorithm, F (x, y)=λ exp (-(x²+y²)/σ²), (x, y) dxdy=1, be a Gaussian function to ∫ ∫ F, (x, y) for picture position (x, y) gray value at place, R (x for i, y), L (x, y) respectively image at point (x, y) reflecting component at place and illumination component.

3. pointer instrument automatic reading method according to claim 1, it is characterised in that described image binaryzation adopts maximum variance between clusters.

4. pointer instrument automatic reading method according to claim 1, it is characterized in that, also proceed as follows after image binaryzation: carry out Morphological scale-space, the pointer of breakpoint is linked up, then connected domain analysis is carried out, when the length and width of connected domain boundary rectangle are less than certain threshold value, arranging this connected domain is background parts, filters the numeral in instrument, symbol, word.

5. pointer instrument automatic reading method according to claim 1, it is characterised in that the method processed based on horizontal scanning line extracts feature pixel, specifically comprises the following steps that

(1-1) define bianry image and be sized to h × w, initializing variable i=0, setForeground pixel point value is 1；According to image resolution ratio, calculate the number of pixels l shared by cursor line width_zhizhen；

(1-2) the i-th row pixel in scanogram, i ∈ [1,2,3...h], wherein will regard a line segment as by continuous print foreground pixel point, and if total k section, they will be labeled as b_j, j=1,2,3...k, and the number of pixels of every section is designated as l_j,l_j∈[1,2,3...w]；

(1-3) by the number of pixels l of every section_jWith l_zhizhenCompare, if l_j≤l_zhizhen, then the line segment midpoint of jth section coordinate in former bianry image is saved in set H；

(1-4) judge that whether i is equal to h, if it is not, then make i=i+1, returns step (1-2), otherwise the coordinate in set H is carried out Hough transform detection of straight lines.

6. pointer instrument automatic reading method according to claim 5, it is characterised in that in described step (1-4), sets the n ranks space-number representing that scanning line processes, after scanning through the i-th row, and then scanning i+n row.

7. pointer instrument automatic reading method according to claim 1, it is characterised in that employing preset angle configuration calculates the step of reading and is:

(2-1) setting needle deflections as [a, b], pointer is to swing from the left side to the right, and minimum scale place angle is θ₁, maximum scale place is θ₂, Z is instrument measurement reading value, and ε is current pointer and x-axis negative direction angle, during with the beat of pointer around point set up an xoy coordinate system for coordinate center；

(2-2) to the uniform instrument of scale, there is proportion expression:

$\frac{\mathrm{\&epsiv;}-{\mathrm{\&theta;}}_1}{{\mathrm{\&theta;}}_2-{\mathrm{\&theta;}}_1}=\frac{Z-a}{b-a};$

Therefore reading $Z=\frac{\mathrm{\&epsiv;}-{\mathrm{\&theta;}}_1}{{\mathrm{\&theta;}}_2-{\mathrm{\&theta;}}_1}\&CenterDot;(b-a)+a.$