CN107784307A - Pointer-type water meter recognition methods based on compressed sensing and Hough transformation - Google Patents

Abstract

The invention discloses a kind of pointer-type water meter recognition methods based on compressed sensing and Hough transformation, step includes：1) image preprocessing；2) dictionary D is constructed；3) parameter information of circle is obtained；4) pointer feature is extracted, the circle in correcting image is detected successively, finally gives the edge graph of all pointers；5) judge reading, the center using the cylindrical center of circle as pointer, using the point farthest from the center of circle as the needle point of pointer, obtain the angle of pointer rotation, find the center of circle of all circles of correcting image and the needle point of pointer；The anglec of rotation of each pointer is obtained according to the needle point in the round center of circle and pointer, then with look-up table, obtains the reading of each pointer, finally gives meter reading.The inventive method, step is simple, and accuracy is high, is easy to promote.

Description

Pointer-type water meter recognition methods based on compressed sensing and Hough transformation

Technical field

The invention belongs to image identification technical field, is related to a kind of pointer-type water meter based on compressed sensing and Hough transformation Recognition methods.

Background technology

In the industries such as electric power, water conservancy, pointer instrument is because its is simple in construction, directly perceived, operation is easy, low cost and other advantages It is employed extensively.Water meter measures the situation of water consumption there is also pointer instrument, and the dial plate in the work obtains main at present It is this using artificial reading is big with the method labor intensity of recording meter panel data, operating efficiency is low by being accomplished manually.

The realization of pointer instrument automatic identification is main, it is necessary to be SHAPE DETECTION to circle and the extraction of pointer in dial plate Detection method is to use Hough transformation (hough conversion) and generalised Hough transform.The main thought of this method is by XY domains Image is transformed into round-shaped parameter space, and then sparse peak value is found in parameter space, is again returned in XY domains, most The round-shaped center of circle and the radius in image are found eventually.But the detection method all can substantially for detection in addition to a straight line It is related to the parameter of 3 even more than 3, causes the memory space used in multivariable process big, calculates time length, and then calculate It is less efficient.Therefore, it is most important to improve identification of the efficiency of SHAPE DETECTION for pointer instrument.

The existing efficiency method thought for improving SHAPE DETECTION is in conjunction with the structure of image after being pre-processed to image Information reduces the scope residing for conversion pixel, and the time is calculated so as to reduce.This method is by Hough transformation to pointer-type water meter Round-shaped detection efficiency improves but improves the requirement to the pre-processed results of image, does not improve shape inherently The efficiency of detection, therefore the efficiency for improving SHAPE DETECTION still needs to further study.

The content of the invention

It is an object of the invention to provide a kind of pointer-type water meter recognition methods based on compressed sensing and Hough transformation, solves For prior art using Hough transformation extraction picture shape feature, it is big memory space to be present, calculates time length, efficiency is slower asks Topic.

The technical solution adopted in the present invention is a kind of pointer-type water meter identification side based on compressed sensing and Hough transformation Method, implement according to following steps：

Step 1, image preprocessing；

Step 2, construction dictionary D；

Step 3, the parameter information for obtaining circle；

Step 4, extraction pointer feature

The center of circle obtained according to step 3 and radius parameter, the data in circle are handled, retain the edge letter of pointer Breath；Expansion process is carried out to marginal information to obtain the expanding image of a pointer again；Edge is carried out to the expanding image again to carry Take, obtain the edge image of the pointer；

The circle in correcting image is detected successively, finally gives the edge graph of all pointers；

Step 5, judge reading

Center using the cylindrical center of circle as pointer, using the point farthest from the center of circle as the needle point of pointer, obtain pointer rotation The angle turned, finds the center of circle of all circles of correcting image and the needle point of pointer；

The anglec of rotation of each pointer is obtained according to the needle point in the round center of circle and pointer, then with look-up table, obtained each The reading of pointer, finally gives meter reading.

The invention has the advantages that step is simple, easy to operation, the identification accuracy to pointer-type water meter reading is notable Improve.

Brief description of the drawings

Fig. 1 is the pointer-type water meter original image of the inventive method identification object；

Fig. 2 is to the gray level image after Fig. 1 processing；

Fig. 3 is to the binary image after Fig. 2 processing；

Fig. 4 is to the correcting image after Fig. 3 processing；

Fig. 5 is Fig. 4 single pointer image；

Fig. 6 is the single circular image that Fig. 5 detects to obtain；

Fig. 7 is that detection obtains Fig. 5 pointer；

Fig. 8 is that detection obtains the edge image of pointer in Fig. 7；

Fig. 9 is that detection obtains the centered position of Fig. 4 institutes；

Figure 10 is that detection obtains all centers of circle of Fig. 4 and the position of pointer needle point；

Figure 11 is the pointer-type water meter original image of the embodiment of the present invention 2；

Figure 12 is that detection obtains all centers of circle of Figure 11 and the position of pointer needle point.

Embodiment

The present invention is described in detail with reference to the accompanying drawings and detailed description.

The technical thought of the present invention is to regard SHAPE DETECTION problem as an inverse Hough transform problem, searching being capable of table Show the best of breed of the parameter field element of view data.If each unit in parameter field is corresponding with a picture shape, For this process inverted equivalent to the dictionary of one possible shape of generation, image can most preferably be represented by then being found with base tracking The set of dictionary element.Compressed sensing can be seen as a kind of tracking of base, and it does not handle view data directly, but with non-traditional Sampling, accidental projection as measurement；Compressed sensing shows that the information in image can obtain from less random measurement, right For SHAPE DETECTION, its purpose is not reconstructed image, but some features in detection image；If Hough transformation comprises only A small number of peak values, then image is compressible or perhaps sparse in the sense that SHAPE DETECTION；, can be with compressed sensing The sparse peak set of Hough transformation is found by solving a linear programming problem, the position correspondence of these peak values is in detection To image in form parameter.

The pointer-type water meter recognition methods of the present invention, based on above-mentioned theory, implements according to following steps：

Step 1, image preprocessing

1.1) gray processing is carried out to image

RGB image is converted into gray level image, due to the shape only having in image for needing to extract to pointer instrument identification Shape characteristic information, so the color information of RGB color pattern reaction has many redundancies；And gray level image not only remains coloured silk Shape facility information in color image, and substantial amounts of memory space can be saved.

Pointer-type water meter original image shown in Fig. 1 is converted into the gray level image shown in Fig. 2.

1.2) binaryzation is carried out to image

Due to needing the shape facility information only having in image extracted, and the shape facility information required for it and the back of the body The intensity profile of scape is different, so needing to carry out gray level image binary conversion treatment, further saves memory space and improves and scheme As the efficiency of processing；Binarization method in this step uses sobel operators, and the threshold value of the operator is obtained by adaptive approach .

Gray level image shown in Fig. 2 is converted into the binary image shown in Fig. 3.

1.3) image is corrected

According to original image (Fig. 1) and with m³For reference, because the obtained width original image is tilted, therefore need Binary image is corrected, larger error when otherwise finally being identified to water meter be present；Using MATLAB's in this step Imrotate functions are corrected, and the angle of rotation is obtained using adaptive approach.

Binary image shown in Fig. 3 is corrected to obtain the correcting image shown in Fig. 4.

Step 2, construction dictionary D

2.1) dictionary D atom is built

A possible parameterized shape in each atom representative graph image field in dictionary D, i.e. each row of dictionary equivalent to Shape template, shown according to the numerical digit of original image (Fig. 1) pointer-type water meter, each is circular complete with 64 × 64 matrix Be included, it is contemplated that the authenticity of dictionary size, each round-shaped is transformed into 64 × 64 matrix is located Reason, it is 32 or so to obtain each circular radius, then the scope for finally taking radius is [30,34], and dictionary D size is [64* 64,64*64*5]。

Corresponding parameter vector is obtained according to priori, for each parameter vector, establishes a width size as 64 × 64 The image of pixel, the gray value for the pixel that the corresponding circle of this parameter vector in image passes through is arranged to 255, before representing that it is Scape, the gray value of other pixels are arranged to 0, and it is background to represent it；Obtain by this method corresponding to this parameter vector Parametric image, the gray value of each pixel of this parametric image is formed into a gray matrix, by the row of this gray matrix Vector links together in an orderly manner forms a length as 64 × 64 one-dimensional vector, and the one-dimensional vector is exactly the original in dictionary D Son.

2.2) dictionary D is constructed

For all parameter vectors, said process is performed both by, obtains the one-dimensional vector that 64*64*5 length is 64 × 64, These one-dimensional vectors are combined to obtain the matrix that a size is [64*64,64*64*5], this matrix is exactly dictionary D；Although Dictionary D is without the shape justified including all possibility, but the circle all wraps the shape for the circle being likely to occur in original image Containing inside, so dictionary D is complete for original image.

Step 3, the parameter information for obtaining circle

Parameter based on compressed sensing and Hough transformation test pattern, i.e., using formula f=D*P, wherein f is by 64 × 64 Image be converted into length be 64 × 64 one-dimensional vector, P be using parameter space indicate weighing vector, D is dictionary；P is bigger Represent that the correlation of atom corresponding to dictionary D is bigger.

In embodiment, according to the characteristic of structure dictionary, by each in pointer instrument shown in correcting image (Fig. 4) Pointer is transformed into the matrix of 64*64 sizes, and it is as shown in Figure 5 to obtain single pointer image.It can be seen from Fig. 5, the single pointer There was only a circle in image, then need to only find the maximum in parameter space instruction weighing vector P and correspond to atom in dictionary D i.e. For the required parametric shape obtained, you can obtain the corresponding parameter vector in step 2.According to the p value of acquisition, detection obtains list Accordingly round shape obtains single circular image as shown in fig. 6, the center of circle is (113,198) to individual pointer image (Fig. 5), radius 31.

Step 4, extraction pointer feature

The center of circle obtained according to step 3 and radius parameter, the data in circle are handled, retain the edge letter of pointer Breath；The expanding image for obtaining a pointer to single circular image (Fig. 6) marginal information progress expansion process again is as shown in Figure 7；Again Edge extracting is carried out to the expanding image (Fig. 7), the edge image for obtaining the pointer is as shown in Figure 8；

The circle in correcting image (Fig. 4) is detected successively, finally gives edge image such as Fig. 9 institutes of all pointers Show, "+" therein represents the position in the center of circle.

Step 5, judge reading

According to the result of step 4, it is contemplated that the center of pointer and the cylindrical center of circle should be in a places, therefore by cylindrical circle Center of the heart as pointer, using the point farthest from the center of circle as the needle point of pointer, the angle of pointer rotation is obtained, finds correction figure As the center of circle of all circles in (Fig. 4) and the needle point image of pointer are as shown in Figure 10；

The anglec of rotation of each pointer is obtained according to the needle point in the round center of circle and pointer, then with look-up table, by the anglec of rotation Degree brings table 1 into, obtains the reading of each pointer, finally gives meter reading

It is as shown in table 2 the final meter reading of embodiment 1 (Fig. 1), is that embodiment 2 (Figure 11) is final as shown in table 3 Meter reading.

The corresponding relation of table 1, the anglec of rotation and total indicator reading

Anglec of rotation θ	Total indicator reading
		θ>5.3407&θ<=5.9690	3
(θ>=0＆ θ<=0.3142) | (θ>5.9690&θ<=6.2832)	2
		θ>0.3142&θ<=0.9425	1
θ>0.9425&θ<=1.5708	0
		θ>1.5708&θ<=2.1991	9
θ>2.1991&θ<=2.8274	8
		θ>2.8274&θ<=3.4558	7
θ>3.4558&θ<=4.0841	6
		θ>4.0841&θ<=4.7124	5
θ>4.7124&θ<=5.3407	4

The reading of each pointer in table 2, Fig. 1

Above step is applied to (embodiment 2) as shown in figure 11, obtains the needle point for the center of circle and pointer justified in water meter such as Shown in Figure 12, it is as shown in table 3 to finally give meter reading.

The reading of each pointer in table 3, Figure 11

。

Claims (5)

1. a kind of pointer-type water meter recognition methods based on compressed sensing and Hough transformation, it is characterised in that according to following steps Implement：

Step 1, image preprocessing；

Step 2, construction dictionary D；

Step 3, the parameter information for obtaining circle；

Step 4, extraction pointer feature

The center of circle obtained according to step 3 and radius parameter, the data in circle are handled, retain the marginal information of pointer；Again Expansion process is carried out to marginal information to obtain the expanding image of a pointer；Edge extracting is carried out to the expanding image again, obtained To the edge image of the pointer；

The circle in correcting image is detected successively, finally gives the edge graph of all pointers；

Step 5, judge reading

Center using the cylindrical center of circle as pointer, using the point farthest from the center of circle as the needle point of pointer, obtain pointer rotation Angle, find the center of circle of all circles of correcting image and the needle point of pointer；

The anglec of rotation of each pointer is obtained according to the needle point in the round center of circle and pointer, then with look-up table, obtains each pointer Reading, finally give meter reading.

2. the pointer-type water meter recognition methods according to claim 1 based on compressed sensing and Hough transformation, its feature exist In in described step 1, specifically including：

1.1) gray processing is carried out to image

RGB image is converted into gray level image, pointer-type water meter original image is converted into gray level image；

1.2) binaryzation is carried out to image

Using sobel operators, the threshold value of the operator is obtained by adaptive approach, converts gray images into binary image；

1.3) image is corrected

Corrected using MATLAB imrotate functions, the angle of rotation is obtained using adaptive approach；By binary picture As being corrected to obtain correcting image.

3. the pointer-type water meter recognition methods according to claim 1 based on compressed sensing and Hough transformation, its feature exist In in described step 2, specifically including：

2.1) dictionary D atom is built

A possible parameterized shape in each atom representative graph image field, shows according to the numerical digit of pointer-type water meter in dictionary D Show, each circle be completely included with 64 × 64 matrix, by each it is round-shaped be transformed into 64 × 64 square Handled in battle array, then the scope for finally taking radius is [30,34], and dictionary D size is [64*64,64*64*5]；

Corresponding parameter vector is obtained according to priori, for each parameter vector, it is 64 × 64 pixels to establish a width size Image, the gray value for the pixel that the corresponding circle of this parameter vector in image passes through is arranged to 255, it is prospect to represent it, The gray value of other pixels is arranged to 0, and it is background to represent it；Parameter corresponding to this parameter vector is obtained by this method Image, the gray value of each pixel of this parametric image is formed into a gray matrix, by the column vector of this gray matrix Link together the one-dimensional vector for forming that a length is 64 × 64 in an orderly manner, and the one-dimensional vector is exactly the atom in dictionary D；

2.2) dictionary D is constructed

For all parameter vectors, said process is performed both by, the one-dimensional vector that 64*64*5 length is 64 × 64 is obtained, by this A little one-dimensional vectors combine to obtain the matrix that a size is [64*64,64*64*5], and this matrix is exactly dictionary D.

4. the pointer-type water meter recognition methods according to claim 1 based on compressed sensing and Hough transformation, its feature exist In in described step 3, detailed process is：

Parameter based on compressed sensing and Hough transformation test pattern, i.e., using f=D*P, wherein f is to turn 64 × 64 image Change the one-dimensional vector that length is 64 × 64 into, P is to indicate weighing vector using parameter space, and D is dictionary.

5. the pointer-type water meter recognition methods according to claim 1 based on compressed sensing and Hough transformation, its feature exist In, in described step 5, table look-up is using the corresponding relation of table 1,

The corresponding relation of table 1, the anglec of rotation and total indicator reading

。