KR20040077091A - Image processing Method of auto meter reading system - Google Patents

Abstract

PURPOSE: An image processing method for telemetry system is provided to improve the image processing speed and recognize correctly numbers by processing an image of a number display of a meter. CONSTITUTION: An image capture process is performed to capture a number display part of a meter as a gray scale image. The gray scale image of the number display part of the meter is converted to a binary image. A number part is set up as a search region. Each outline of models is tracked within the search region of the binary image. The tracking data are stored. A desired model is selected by removing the noise from the search region of the binary image. A model recognition process is performed to compare spectrums of the selected model with the practical spectrums of the corresponding model and determining a high correlation number as a model number.

Description

Image processing method of auto meter reading system

The present invention relates to an image processing method used to remotely read various meters such as an electric meter, a gas meter, a water meter, and the like, so that the number can be automatically recognized by taking a picture of the number display unit of the meter and automatically recognizing the corresponding number. It relates to an image processing method of a remote meter reading.

As is well known, the conventional meter reading method is a method that the meter operator directly visits the site and checks the numerical display of the meter visually and records it. In this case, the meter operator must visit the site every time. Personnel should be put into the metering operation, and if the resident or the building manager is absent while the meter is installed inside the building, there is a problem of repeatedly visiting the meter in order to read the meter.

Therefore, in recent years, a remote meter reading method for automatically recognizing the number displayed on the numeric display by image processing after the numerical display of the meter image is being researched and developed.

Figure 1 shows a remote meter reading system that is currently applied, with reference to this will be described in the conventional remote meter reading method.

First, when the operator operates the remote meter reading system by operating the input unit 70, the second control unit 60 is operated and controlled by the output signal from the input unit 70, so that the image photographing signal is transmitted to the second transmitting and receiving unit 50. Is transmitted wirelessly). The image photographing signal from the second transmitting / receiving unit 50 is then wirelessly received through the first transmitting / receiving unit 40 and then input to the first control unit 30, and the first control unit 30 is the first transmitting / receiving unit. The operation is controlled by the image photographing signal from the 40 to output the corresponding control signal to the image photographing unit 20, and the image photographing unit 20 is operationally controlled by the control signal of the first control unit 30. The number display unit 11 of the meter 10 is imaged. The image signal of the image photographing unit 20 is wirelessly transmitted through the first transmission / reception unit 40 after being input to the first control unit 30, and the image signal from the first transmission / reception unit 40 is the second transmission / reception unit 40. After wireless reception at 50, it is input to the second control unit 60. The video signal inputted to the second control unit 60 is image-processed by the second control unit 60 and recognized as an arbitrary number. After being recognized as an arbitrary number, the video signal is automatically calculated and outputted to the output unit 80. Will be outputted in the appropriate format. Here, the result value refers to things such as electricity bill, water bill, gas bill.

However, the image processing method used in the conventional remote meter reading method has not seriously organized the image processing method, the image processing speed is considerably slow, the problem that a lot of errors occur in recognizing the number has been seriously raised.

In addition, this type of problem requires a large capacity of remote meter reading system components, resulting in an increase in product cost.

Therefore, the present invention has been invented to solve the above problems, and the object of the present invention is to provide an image processing method of a remote meter reading that allows the image processing speed to proceed faster and more accurate number recognition.

1 is a block diagram showing a remote meter reading system;

2 is a flowchart illustrating an image processing method according to the present invention;

3 is a view for explaining an image photographing step;

4A and 4B are views for explaining a search region data conversion step;

5 to 10b are views for explaining a recognition model selection step;

11 to 12 are diagrams for explaining the model recognition step.

-Explanation of terms for the main parts of the accompanying drawings-

10; Meter, 11; Number Display,

20; Imaging unit, 30; First control unit,

40; A first transmitting and receiving unit, 50; Second transmission and reception unit,

60; Second control unit, 70; Input unit,

80; Output unit.

The present invention for achieving the above object, the image taking step of photographing the numeric display portion of the meter as a grayscale image; A search region data conversion step of converting a grayscale image into a binary image and setting a portion where a number may exist as a search region; Recognition model selection step of tracing the outline of the models existing in the search region of the binary image, storing them, removing the miscellaneous existing in the search region of the binary image, and selecting a model to be recognized among these models; Based on one pixel, the horizontal component value, the vertical component value and the cumulative value of the angle of the selected model are spectralized, and the actual horizontal component spectrum, vertical component spectrum, angle accumulation spectrum, and numerical Comparing the reference spectra with each other, the model recognition step of taking a high number of correlation between them as a numerical value of the model is characterized in that the method characterized in that.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

According to FIG. 2, the image processing method of the remote meter reading according to the present invention includes: an image photographing step of photographing the numeric display unit 11 of the meter 10 as a grayscale image; A search region data conversion step of converting a grayscale image into a binary image and setting a portion where a number may exist as a search region; Recognition model selection step of tracing the outline of the models existing in the search region of the binary image, storing them, removing the miscellaneous existing in the search region of the binary image, and selecting a model to be recognized among these models; Based on one pixel, the horizontal component value, the vertical component value and the cumulative value of the angle of the selected model are spectralized, and the actual horizontal component spectrum, vertical component spectrum, angle accumulation spectrum, and numerical Comparing the reference spectra, a model recognition step of taking as a numerical value of the model a number of high correlation between them.

Each step will be explained in more detail as follows.

First, the image photographed by the image capturing unit 20 is generally composed of an 8-bit grayscale image as shown in FIG. 3. The image photographing unit 20 is installed in the meter 10 and is disposed at a position capable of accurately photographing the numeric display unit 11 of the meter 10.

In the case of the 8-bit grayscale image, the image information value is composed of RGB values. When the image information value is used as the basic image data, the image processing speed is greatly reduced, and the hardware becomes large. Convert to binary image as shown. In addition, since a portion of the entire image where a number may be present is limited and there is no need for image processing except for this portion, only a portion where a number may exist is set as a search region, as shown in FIG. 4B. Is set to the appropriate fixed value. On the other hand, in the present embodiment, in order to block the connection of the miscellaneous that may occur in any case, 2 pixels arranged outward from the search area are unconditionally converted to white.

Thereafter, the outlines of the models existing in the search region of the binary image are tracked and stored, which will be described in more detail with reference to FIGS. 5 to 10B.

First, a first step of sequentially checking data values of inspection area pixels in an inspection order is performed. The inspection sequence is from left to right and top to bottom.

Thereafter, it is determined whether the pixel satisfies the starting condition for the outline tracking, and if the starting condition is satisfied, the third step is performed, and if the starting condition is not satisfied, the second step is performed. In the present embodiment, the data value of the pixel with white color is "0", the data value of the pixel with black color is "4", and the initial reference pixels P1, P2, and P3 satisfying the starting condition are shown in FIG. As shown in Fig. 6, the data value of the corresponding pixel is " 4 ", and the data value of the left pixel in contact with the pixel is " 0 ".

As such, when a pixel that satisfies the starting condition for outline tracking is detected, a third step of storing the coordinate value of the pixel as the initial reference pixel is performed.

Subsequently, a fourth step of tracing an outline is performed by sequentially checking data values of pixels adjacent to the reference pixels P1, P2, and P3 according to an inspection order. In the present exemplary embodiment, the inspection order of the pixels adjacent to the reference pixels P1, P2, and P3 is sequentially set with the upper left as the first centering around the reference pixels P1 and P2 as shown in FIGS. 7A and 7B. Proceed clockwise, but proceeds to the starting sequence according to the following equations (1) and (2).

Here, d1 is the inspection order of the reference pixel, d2 is the position information value (inspection order) of the pixel relative to the reference pixel, d is the start order, and% 8 is (d1 + 1) or (d2 + 4) as 8 The remaining value is shown.

Thereafter, when the data value of the corresponding pixel satisfies "4", a fifth step of storing the positional information value of the corresponding pixel relative to the reference pixel in the reference pixel P1 is performed.

Thereafter, it is determined whether the corresponding pixel is the first reference pixel, and if it is not the first reference pixel, the seventh step is performed, and if the first reference pixel, the sixth step is performed. Here, if the pixel is the initial reference pixel, it means that the current reference pixel corresponds to the end point of the outline tracking, and if the pixel is not the first reference pixel, the pixel to be tracked as the outline still exists.

The seventh step is to return to the fourth step after storing the coordinate value of the corresponding pixel as a new reference pixel.

The eighth step is to determine whether the next pixel of the corresponding pixel exists, to return to the first step if the next pixel to be inspected exists, and to end all operations if the next pixel to be inspected does not exist.

Referring to FIG. 7A as an example, an outline tracking operation is described in detail. A pixel having a data value of "4" and a data value of "0" of a left pixel adjacent to the pixel is taken as an initial reference pixel P1, and thus the object is not shown. After storing the coordinate value, the starting order is taken using Equation 1, and the outline (pixel having a data value of "4") is traced from this starting order according to the checking order. Here, in the case of the first reference pixel P1, since the inspection order d1 of the reference pixel is "0" (in the case of the first reference pixel, d1 is "0" unconditionally), so that the starting sequence d is "1". Then, adjacent pixels are searched for the pixels having a data value of "4" by comparing the data values in the clockwise direction from the starting order "1" around the reference pixel P1 according to the inspection order. In the case of FIG. 7A, since the inspection order “4” corresponds to this, the relative position information value of the corresponding pixel with respect to the reference pixel becomes “4”, and the corresponding pixel becomes a new reference pixel Px. Thereafter, as shown in FIG. 7B, the start order d is determined by sequentially applying Equation 2 and Equation 1 based on the new reference criterion Px. As mentioned above, since d2 of the new reference pixel Px is "4", when it is applied to Equation 2, d1 is "0", and when it is applied to Equation 1 again, d is "1". . Thereafter, the above operation is repeated to track the outline. In the case of "1, 2, 3, 5, 7", the outline is one, but in the case of "0, 4, 6, 9", the outside and the inside are There are two outlines, one each in the case of "8" as in the present embodiment, and one outer line and two inner lines, that is, three outer lines, exist on the outside. In the case of the outlines outside these numbers, the traces of the outlines are tracked in the clockwise direction with respect to the initial reference pixel as shown in Figs. 7A and 7B.

8A and 8B are diagrams for explaining an operation of tracking an inner outline, and the tracking operation thereof is also the same as that of FIGS. 7A and 7B, and thus a detailed description thereof will be omitted. However, in the case of the inner outline, the advancing direction of the trace tracking proceeds in the counterclockwise direction based on the initial reference pixel P2 (P3 is the same as P2).

When the tracking of the outline is completed in this way, in the case of "8", three models are traced to the outline as shown in FIG. 9.

Thereafter, as shown in FIG. 10, in the state where all models in the search region are tracked, the operation for removing miscellaneous images and selecting a model to be recognized as a number is performed.

In the present embodiment, when the number of pixels corresponding to each model is smaller than the reference value, it is recognized as a ghost and removed. In this state in which the debris is removed (see FIG. 10A), two models having the largest square area formed by the maximum X and Y coordinate values and the minimum X and Y coordinate values of each model are taken, Comparing the lengths, if the ratio of the vertical length is within the reference range, the two models are judged to be half-finished, and both models are taken as recognition models, and if the ratio of the vertical length exceeds or falls short of the reference range, Only one model having a long vertical length among the three models was taken as a recognition model (see FIG. 10B). Here, when an arbitrary model (inner outline) exists in the area of another arbitrary model (outer outline), these are recognized as one model.

Then, based on any one pixel, the horizontal component value, the vertical component value, and the cumulative value of the angle of the selected model are spectrified.

Referring to the operation of selecting the pixel as the reference of the spectrum, as shown in Fig. 11, taking the reference line (L) horizontally transverse to the center of the longitudinal length of the recognition model, and among the pixels forming the outline of the model If there are two pixels P 'and P "which meet the reference line L, the left pixel P' is defined as the reference pixel P 'for spectrumization among the pixels. If there are three or more pixels that meet (L), the reference line (L) is sequentially moved downward from the longitudinal center of the recognition model until there are two pixels (P ', P ") that meet the reference line (L). After that, the left pixel P 'is defined as the reference pixel P' for spectra.

As such, when the reference pixel P 'for spectralization is determined, the horizontal component spectrum, the vertical component spectrum, and the angular cumulative spectrum are obtained, and as shown in FIG. 12, the actual horizontal component spectrum and the vertical component spectrum of the corresponding model. And the angular cumulative spectrum and the reference spectrum for each number are compared with each other, and a number having a high correlation between them is taken as the numerical value of the model.

Meanwhile, the following equations (3) to (5) are used to generate the model.

Where fv _i is the feature vector at the i th sample point, x _c is the horizontal component of the recognition character, y _c is the vertical component of the recognition character, θ _c is the angular component of the recognition character, and m is the feature vector of the arbitrary character. The number of character images used to represent S denotes the number of samplings.

In addition, by comparing the spectrum of the actual model and the reference spectrum for each number using the following equations (6) to (9), after calculating the degree of association between these spectra, the most similar one is taken as the number of the model.

Here, Dc is the distance difference between the feature vector of the reference character and the feature vector of the recognition model (c), Dx _c is the distance difference between the feature vector horizontal components between the reference character and the recognition model (c), and Dy _c is the reference character and the recognition model. (c) Difference in distance between the feature vector vertical components, Dθ _c is the difference in distance between the feature vector angle accumulation values between the reference character and the recognition model (c), x is the horizontal component of the reference character, y is the vertical component of the reference character, θ Is the angle cumulative value of the reference character, w _x is the weight of the horizontal component, w _y is the weight of the vertical component, w _θ is the weight of the angle component. These weights (w _x , w _y , w _θ ) can be appropriately changed according to the type of the character, and the determination is obtained by calculating the most appropriate values through several experiments.

The feature vector of the reference character and the feature vector of the recognition model are compared with each other, and after determining the reference number having a small distance difference Dc between the feature vectors based on the detected recognition model as the number of the recognition model, this is taken.

On the other hand, the spectrum of the reference character may be provided with various types of reference spectrum according to the form of the externally exposed (according to the vertical length of the character exposed to the outside), the distortion of the image by the miscellaneous frequency may be frequently generated There are several types of spectra available. The number of the reference spectrum is provided in a variety of forms can reduce the number recognition error, but as the number of reference spectrum increases the number recognition work rate is lowered, the appropriate number spectrum in consideration of the work speed is set as the reference spectrum.

The present invention is not limited to the above embodiments, and of course, various modifications can be made without departing from the scope of the following claims.

For example, in the present embodiment, the search area is set after converting a grayscale image to a binary image and converting it to a binary image. However, if necessary, the search region is first set from a grayscale image and then converted to a binary image. It may be.

In addition, in the present embodiment, after extracting and extracting the outline, the recognition model is selected by removing the ghost, but of course, the ghost may be removed at the same time while tracking the outline.

In addition, in the present embodiment, when tracking the outline, the data value of the white pixel outside the outline is converted to "7", and the data value of the pixel forming the outline is converted to "2". The starting condition for tracking alone has the advantage that it is possible to track separate outlines easily and accurately without errors. If the data values of the pixels forming the outline and the white pixel outside the outline are not converted, the data value of the corresponding pixel is "4" even if it is not the first reference pixels (P1, P2, P3), and the left pixel adjacent thereto is not converted. If the data value of "0" is compared to the new condition every time, it is necessary to determine whether the pixel is the first reference pixel (P1, P2, P3) or the pixel belonging to the traced outline. The disadvantage of greatly lowering may occur.

Meanwhile, conventionally, the first control unit 30 that controls the operation of the image capturing unit 20 receives a grayscale image from the image capturing unit 20 and outputs it to the outside through the first transmission / reception unit 40. When the output, the second control unit 60 receives the grayscale image data through the second transmission and reception unit 40, and image processing the number and then take a way to output it through the output unit 80 have.

However, in this case, since the image data transmitted from the first transmitting / receiving unit 40 to the second transmitting / receiving unit 50 is in the form of grayscale image data, the amount of data information to be transmitted is large and the processing speed of the remote meter reading system is greatly reduced overall. In order to solve this problem, the processing capacity of the second control unit 60 must be sufficiently secured. However, even if such measures are taken, the processing of the remote meter reading system due to the reduction in the transmission speed in the data transmission process is essential. It is not possible to prevent a decrease in speed, and since a large amount of information must be provided for hardware, a problem arises in that facility cost is also greatly increased.

Therefore, in the present embodiment, the first control unit 30, which is built into the meter 10 to control the operation of the image photographing unit 20, receives the grayscale image from the image photographing unit 20 and processes the image. After the number is recognized, only the model-recognized numerical value is wirelessly transmitted through the first transmission / reception unit 40. According to the present embodiment, since the first control unit 30 provided in the meter 10 carries out the image processing by itself, only the model-recognized numeric value is transmitted, thereby greatly reducing the amount of data to be transmitted, and thus, the remote meter reading system. The overall processing speed of the system is greatly improved, and a large amount of information processing is not required, so the cost of equipment is also greatly reduced. The first transmitting unit 40 and the second transmitting unit 50 may be wired or wirelessly connected, but it is preferable to connect in a wireless manner in consideration of various conditions.

According to the present invention as described above, after converting the grayscale image into a binary image, taking only the portion where the number can exist as a search region, and traces the outline of the models existing in the search region of the binary image and stores it, After removing the noise in the search region of the binary image, selecting the model to be recognized from these models, and then accumulating the horizontal component value, the vertical component value and the cumulative value of the selected model based on an arbitrary pixel Spectralizing and comparing the actual horizontal component spectrum, the vertical component spectrum and the angular cumulative spectrum of the model with the reference spectrum for each number, and take a high number of correlations between them as the numerical value of the model, It has the effect of being able to recognize numbers faster and more accurately.

In addition, since the hardware of the remote metering system can be configured more simply, its cost is reduced, and the processing speed of the remote metering system is expected to be faster due to the improvement of image processing speed.

Claims (5)

An image photographing step of photographing the numeric display unit 11 of the meter 10 as a grayscale image; A search region data conversion step of converting a grayscale image into a binary image and setting a portion where a number may exist as a search region; Recognition model selection step of tracing the outline of the models existing in the search region of the binary image, storing them, removing the miscellaneous existing in the search region of the binary image, and selecting a model to be recognized among these models; On the basis of any one pixel, the horizontal component value, the vertical component value, and the cumulative value of the angle of the selected model are spectralized, and the actual horizontal component spectrum, the vertical component spectrum, the angle accumulation spectrum, and the number of the corresponding model. And a model recognition step of comparing each reference spectrum with each other and taking a number having high correlation with each other as a numerical value of the model. According to claim 1, wherein the outline tracking operation in the recognition model selection step, A first step of sequentially checking data values of the inspection area pixels in an inspection order; A second step of determining whether the pixel satisfies the starting condition for outline tracking, performing a third step if the starting condition is satisfied, and performing an eighth step if the starting condition is not satisfied; A third step of storing the coordinate value of the pixel as the initial reference pixel; A fourth step of tracking an outline by sequentially checking data values of pixels adjacent to the reference pixel in an inspection order; A fifth step of storing the position information of the corresponding pixel with respect to the reference pixel in the reference pixel; Determining whether the pixel is an initial reference pixel, and performing a seventh step if it is not the first reference pixel, and performing an eighth step if the first reference pixel; A seventh step of returning to the fourth step after storing the coordinate value of the corresponding pixel as a new reference pixel; An eighth step of determining whether the next pixel of the pixel exists, returning to the first step if the next pixel to be checked exists, and ending all operations if the next pixel to be checked exists; Image processing method of a remote meter reading, characterized in that consisting of. The image processing method of claim 1, wherein in the recognition model selection step, the blurring operation is performed to recognize and remove the blurring when the number of pixels forming each model is smaller than the reference value. . The method of claim 1, wherein in the recognition model selection step, after taking two models having the largest square area formed by the maximum X and Y coordinate values and the minimum X and Y coordinate values of each model, Comparing the lengths, if the ratio of the vertical length is within the reference range, the two models are judged to be half-finished, and both models are taken as recognition models, and if the ratio of the vertical length exceeds or falls short of the reference range, A method of processing a remote meter reading, wherein only one model having a long vertical length is taken as a recognition model. The image processing according to claim 1, wherein the first control unit (30) installed in the meter (10) controls the operation of the image photographing unit (20) and receives a grayscale image from the image photographing unit (20). After recognizing a number by using the first transmission unit 40, only the image value of the model recognition method for remote meter reading, characterized in that for transmitting.