JP2019091202A - Segment display reader and method - Google Patents

Abstract

To obtain appropriate recognition accuracy even when display data does not increase or decrease in a fixed direction or rate.SOLUTION: A storage unit 13 stores, as misrecognition pattern data 13A, a misrecognition pattern P formed of a combination of a front character m and a rear character n which are switched and displayed back and forth by a segment display and a switching character e which can be generated at their switching display time, a probability information generation unit 14B obtains a target recognition result D(i), an immediately preceding recognition result D(i-1), and an immediately subsequent recognition result D(i+1) from time series data 13B, extracts, for each digit position j from these results, a character located at the same digit as a target character x(i, j), an immediately preceding character x(i-1, j), and an immediately subsequent character x(i+1, j), and when a matching false recognition pattern P having e, m, and n matching these characters is registered in the false recognition pattern data 13A, generates accuracy information Q(i, j) of the target character x(i, j) based on a candidate character R(i, j) estimated from the corresponding pattern P.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a segment display reading technique for reading digital data displayed and output from an image obtained by capturing a segment display.

In the past, as a technology for measuring the usage of gas, electricity, water, etc., a segment display reading device is mounted on the integrated quantity meter that measures the integrated quantity of gas, electricity, water, etc. There has been proposed a technique for automatically reading the integrated amount output for display (see, for example, Patent Document 1).
This segment display reader reads numbers displayed and output as digital data by photographing the numbers displayed and output by LEDs and LCDs with a segment display with a camera and performing character recognition processing on the obtained image. is there.

Japanese Patent Application Laid-Open No. 2000-200322

  In general, the segment display switches segments to be lit in response to changes in display data. For this reason, depending on the imaging timing of the image, a segment that transitions from on to off by switching may be imaged overlapping with a segment that is on after switching. For example, when a certain digit is switched from "5" to "6", these "5" and "6" segments overlap and are photographed as "8". In the case of character recognition of such an overlapping image, a recognition error occurs.

In the prior art described above, the predicted value of the integrated amount is calculated in advance from the speed at which the integrated amount increases, and the recognition accuracy of the read result is determined by comparing the read result obtained by character recognition with the predicted value. doing.
Thus, when the deviation between the read result and the predicted value is small, the recognition accuracy of the read result is high, and when the deviation is large, it is determined that the recognition accuracy of the read result is low.

However, since the predicted value in the prior art is calculated based on the predicted amount × time on the assumption that the display data is increased or decreased at a fixed rate in a fixed direction, digital data such as an integration amount may be used. For example, although appropriate recognition accuracy can be obtained, there is a problem that when the display data changes irregularly, appropriate recognition accuracy can not be obtained.
For example, in a thermometer having a segment indicator, display data may be irregularly increased or decreased in order to indicate the measured temperature on the segment indicator. Moreover, not only the temperature but also the pressure and the flow rate may increase or decrease irregularly, and may not increase or decrease at a constant rate in a constant direction. Therefore, according to the prior art, an appropriate recognition accuracy can not be obtained in such a case.

  The present invention is intended to solve such problems, and to provide a segment display reading technology capable of obtaining an appropriate recognition accuracy even when display data does not increase or decrease in a fixed direction at a fixed rate. The purpose is.

  In order to achieve such an object, a segment display reading device according to the present invention is a camera which picks up a display output of a segment display which switches and displays a plurality of types of characters by turning on and off a plurality of segments. A character recognition unit configured to perform character recognition processing on an imaging output of the camera; a storage unit configured to store recognition results obtained by the character recognition unit as time series data; and the time series data And a certainty information generation unit configured to generate certainty information indicating the certainty of the subject recognition result for the subject recognition result selected from among A combination of any two types of preceding and following characters, and switching characters that may occur when these preceding and subsequent characters are switched back and forth Storing the plurality of erroneous recognition pattern data in which the plurality of erroneous recognition patterns are registered in advance, and the accuracy information generation unit is configured to calculate the target recognition result from the time series data and the target recognition result immediately before and in time series order. The immediately preceding recognition result and the immediately subsequent recognition result located immediately after are acquired, and from the object recognition result, the immediately preceding recognition result and the immediately subsequent recognition result, the character located at the same digit for each digit position is the target character, immediately preceding character And the immediately following character are extracted, and a matching false recognition pattern having a combination of the switching character, the previous character and the subsequent character that matches the combination of the target character, the immediately preceding character, and the immediately subsequent character is registered in the false recognition pattern data. If it does, generation of accuracy information indicating the recognition accuracy of the target character is generated based on the candidate character estimated from the matching false recognition pattern In which was to so that.

  In one configuration example of the segment display reading device according to the present invention, in the case where the accuracy information generation unit registers the false recognition pattern in the false recognition pattern data, a character following the false recognition pattern Is selected as the candidate character.

  Further, according to one configuration example of the segment display reading device according to the present invention, when the accuracy information generation unit does not register the false recognition pattern in the false recognition pattern data, switching among the false recognition pattern data is performed. The character selected after the misrecognition pattern is selected as the candidate character from among all the misrecognition patterns in which the hour character and the preceding character match the target character and the immediately preceding character.

  Further, according to one configuration example of the segment display reading device according to the present invention, when the accuracy information generation unit does not register the false recognition pattern in the false recognition pattern data, switching among the false recognition pattern data is performed. The character after the false recognition pattern is selected as the candidate character from among all false recognition patterns in which the hour letter matches the target character.

  Further, in one configuration example of the segment display reading device according to the present invention, when the target character is a numeral, the probability information generation unit adds the target character to the candidate character.

  A segment display reading method according to the present invention is a segment display reading method used in a segment display reading device for reading display output of a segment display, and a plurality of types of cameras are controlled by turning on and off a plurality of segments. An imaging step for imaging the display output of the segment display for switching and displaying the characters; a character recognition step in which the character recognition unit is configured to perform character recognition processing for the imaging output of the camera; The storage step of sequentially fetching the recognition result obtained in the character recognition step and storing it as time series data, and the certainty information generation unit indicates the certainty of the target recognition result for the target recognition result selected from the time series data. Generating probability information configured to generate probability information, said storing step It consists of a combination of any two types of preceding and following characters that are switched and displayed before and after on the segment display, and switching characters that can occur when these before and after characters are switched and displayed before and after. Storing a plurality of erroneous recognition pattern data in which a plurality of erroneous recognition patterns are registered in advance, and the accuracy information generation step is performed in chronological order with respect to the target recognition result and the target recognition result from the time series data The immediately preceding recognition result and the immediately subsequent recognition result located immediately before and immediately after are acquired, and from the target recognition result, the immediately preceding recognition result, and the immediately subsequent recognition result, the character located at the same digit for each digit position is the target character A character before switching, a character before, and a character that is extracted as the character before and the character after, and that matches the combination of the target character, the character before, and the character after, When a matching false recognition pattern having a combination of characters is registered in the false recognition pattern data, the probability information indicating the recognition probability of the target character is generated based on the candidate characters estimated from the matching false recognition pattern. Contains steps.

  According to the present invention, it is possible to obtain an appropriate recognition accuracy even when display data does not increase or decrease at a constant rate in a constant direction. Therefore, also for devices other than integrated quantity meters that measure integrated quantities of gas, electricity, and water, for example, various devices such as thermometers, pressure gauges, flow meters, etc. read display data displayed and output by their segment displays. In addition, appropriate recognition accuracy can be provided.

It is a block diagram which shows the structure of a segment display reader. It is a display example of a 7-segment display. It is an example of misrecognition occurrence in a 7-segment display. It is a structural example of misrecognition pattern data. It is a flowchart which shows character recognition processing. It is a flowchart which shows a probability information generation process. It is an explanatory view showing an example of accuracy information generation operation. It is an explanatory view of outlier determination. It is an example of outlier determination using a difference root mean square.

Next, an embodiment of the present invention will be described with reference to the drawings.
[Segment display reader]
First, the segment display reading device 10 according to the present embodiment will be described with reference to FIG. FIG. 1 is a block diagram showing the configuration of a segment display reader.

  The segment display reader 10 generally comprises an information processor such as a personal computer and an industrial controller, and in addition to an integrated quantity meter for measuring the integrated quantity of gas, electricity and water, a thermometer, a pressure gauge, a flow meter, etc. It is an apparatus which reads the display output of the segment display 20 mounted in measuring devices, such as various ,, and outputs digital data.

[Principle of the invention]
The segment display device 20 is a display device which switches and displays a plurality of types of characters by performing on / off control of a plurality of segments including display elements such as LEDs and LCDs. FIG. 2 is a display example of a 7-segment display. Here, an example is shown in which numbers are switched and displayed using seven segments.
When such a segment display unit 20 is imaged by a camera, a segment that transitions from on to off upon switching may be imaged overlapping with a segment to be lit after switching depending on the timing of image capture. When such an image is subjected to character recognition processing, erroneous recognition may occur as a result.

  FIG. 3 shows an example of erroneous recognition occurring in a 7-segment display. Here, an example of the switching time letter e generated when switching display from the previous letter m to the subsequent letter n is shown. For example, when the number "4" is switched to the number "7", an image in the middle of the switching may be recognized as an alphabetic character "q". In addition, when the display is switched from the number "2" to the number "3", an erroneous recognition of the alphabet character "a" may occur from the image during the switching. In addition, when the display is switched from the number “0” to the number “2”, an erroneous recognition of the number “8” may occur from the image during the switching. Further, when the number "4" is switched to the number "5", an erroneous recognition of the number "9" may occur from the image during the switching.

When all such false recognitions are verified, it is understood that the following relationship exists between the front character and the rear character which are displayed in succession before and after and the switching character which is misrecognized. The
The number of switching characters that can be generated is limited to several types (four types of “q”, “a”, “8”, and “9” in the case of a 7-segment display).
The combination of the switching character with the immediately preceding character and the immediately following character is limited to a certain extent for each switching character.
In each combination, the actual display character corresponding to the switching character is highly likely to be a character immediately after the combination.
If the switching character is a number, the switching character itself may be a real display character.
If there is no combination including the front character and the rear character as the combination of switching characters, the immediately following characters constituting all the combinations of switching characters may be the actual display characters.

  In the present invention, of the two characters switched and displayed before and after, the character displayed before switching is referred to as the front character, the character displayed after switching is referred to as the back character, and erroneous recognition is made during the switching. These characters are called switching characters. Also, the characters actually displayed are called real display characters.

  The present invention pays attention to the fact that the relationship as described above exists between the preceding character and the following character that are erroneously recognized as the following characters that are switched and displayed before and after. A combination of the preceding character and the following character is registered in advance as a misrecognition pattern, and among these misrecognition patterns, the target character actually recognized and the combination of the immediately preceding character and the immediately following character displayed switched back and forth are matched The candidate characters corresponding to the actual display character are estimated based on the erroneous recognition pattern, and the probability information indicating the recognition accuracy of the target character is generated based on the estimated candidate character.

  The target character, the immediately preceding character, and the immediately subsequent character correspond to the characters switched and displayed in chronological order for each of a plurality of characters constituting the recognition result. For this reason, when the recognition results actually obtained are sequentially stored as time-series data and accuracy information regarding an arbitrary recognition result is generated, the immediately preceding recognition result before and after the target recognition result to be an object From the recognition result immediately after, for each digit position, characters located at the same digit are extracted as a target character, a preceding character, and a subsequent character.

[Configuration of the present embodiment]
Next, the configuration of the segment display reading apparatus 10 according to the present embodiment will be described in detail with reference to FIG.
As shown in FIG. 1, the segment display reading device 10 according to the present embodiment includes a communication I / F unit 11, a camera 12, a storage unit 13, and an arithmetic processing unit 14 as main functional units.

The communication I / F unit 11 has a function of performing data communication with an external device such as the host device 30 via the communication line L.
The camera 12 is a general imaging device such as a CCD camera, and has a function of imaging the display output of the segment display 20.

The storage unit 13 is formed of a storage device such as a semiconductor memory or a hard disk, and has a function of storing various processing data and programs 13P used for processing in the arithmetic processing unit 14.
The program 13 P is a program that realizes various processes in cooperation with the CPU by being read and executed by the CPU of the arithmetic processing unit 14. For example, the program 13 P is an external connected via the communication I / F unit 11 It is read in advance from an apparatus or a recording medium (both not shown) and stored in the storage unit 13.

As main processing data stored in the storage unit 13, there are erroneous recognition pattern data 13A, time series data 13B, and reading result data 13C.
The misrecognition pattern data 13A may occur when any two types of front and back characters switched and displayed before and after on the segment display 20 and these front and back characters are switched and displayed back and forth. A plurality of false recognition patterns consisting of combinations with switching characters are data registered in advance.

FIG. 4 is a configuration example of false recognition pattern data. Here, misrecognition patterns consisting of a combination of the front character m and the back character n and the switching character e (m, n) are registered in the vertical and horizontal matrix format, the vertical direction indicates the front character m, and the horizontal direction is The back letter n is shown, and the switching time letter e (m, n) is arranged at the intersection position.
For example, switching character e (4, 7) which may occur when switching display from “4” to “7” at the intersection position of the front character m = “4” and the rear character n = “7” = "Q" is arranged. Further, the intersection position of the front character m = “4” and the rear character n = “8” is e (4, 8) = blank, which indicates that no erroneous recognition occurs.

The time-series data 13 B is data in which recognition results obtained by character recognition processing of image data which is an imaging output of the camera 12 by the arithmetic processing unit 14 are sequentially fetched and recorded.
The reading result data 13C is data obtained by adding probability information indicating the certainty to the recognition result recorded in the time series data 13B.

The arithmetic processing unit 14 has a CPU and its peripheral circuits, has a function of realizing various processing units by causing the hardware and software to cooperate with each other by reading and executing the program 13P of the storage unit 13. doing.
As main processing units realized by the arithmetic processing unit 14, there are a character recognition unit 14A, a probability information generation unit 14B, and a recording unit 14C.

The character recognition unit 14A has a function of performing character recognition processing on image data that is an imaging output of the camera 12. A known method may be used for the character recognition process.
The certainty information generation unit 14B has a function of generating certainty information Q indicating the certainty of the target recognition result D with respect to the target recognition result D selected from the time-series data 13B of the storage unit 13.
The recording unit 14C has a function of recording the certainty information Q generated by the certainty information generation unit 14B in the reading result data 13C of the storage unit 13 in association with the target recognition result D.

[Operation of First Embodiment]
Next, the operation of the segment display reading device 10 according to the present embodiment will be described.
First, the character recognition operation of the segment display reading device 10 will be described with reference to FIG. FIG. 5 is a flowchart showing the character recognition process.
The character recognition unit 14A repeatedly executes the character recognition process of FIG. 5 at a constant cycle.

First, the character recognition unit 14A picks up the display output of the segment display 20 with the camera 12 (step 100), and performs character recognition processing on the obtained image data (step 101). Thereby, recognition results D regarding a plurality of characters displayed and output are generated.
Thereafter, the character recognition unit 14A records the obtained recognition result D in the time series data 13B of the storage unit 13 (step 102), and ends the series of character recognition processing.

Next, with reference to FIG. 6, the operation of generating the likelihood information of the segment display reading device 10 will be described. FIG. 6 is a flowchart showing the probability information generation process.
For example, when a predetermined time arrives, when a certain number of recognition results D are recorded in the time-series data 13B, the probability information generation unit 14B is notified of a read result transmission request from the higher-level device 30. The probability information generation process shown in FIG. 6 is executed at a time such as when it is necessary to generate the probability information.

  First, from the time-series data 13B of the storage unit 13, the probability information generation unit 14B generates the target recognition result D (i) at time i that is the target of the probability information generation process and the target recognition result D (i). The immediately preceding recognition result D (i-1) and the immediately subsequent recognition result D (i + 1) located immediately before and immediately after the sequential order are acquired (step 110).

  Next, from these D (i), D (i-1), D (i + 1), the likelihood information generation unit 14B extracts the target character x (i, j) of the unextracted digit j and the immediately preceding character x (i−). 1.j) and the next character x (i + 1, j) are extracted (step 111), and it is confirmed whether D (i) is other than “8”, “9”, “a”, “q” (step 112), if D (i) is other than “8”, “9”, “a”, “q” (step 112: YES), the process proceeds to step 121 described later.

  On the other hand, when D (i) is any of “8”, “9”, “a”, and “q” (step 112: NO), the likelihood information generation unit 14B determines the erroneous recognition pattern data of the storage unit 13 13A, it has a combination of switching character e, previous character m, and subsequent character n that matches the combination of x (i, j), x (i-1, j), x (i + 1, j) Whether or not the misrecognition pattern P is registered is confirmed (step 113).

Here, the erroneous recognition pattern P having a combination of e, m and n that matches the combination of x (i, j), x (i-1, j) and x (i + 1, j) is erroneous recognition pattern data If it is registered in 13A (step 113: YES), the likelihood information generation unit 14B selects the subsequent character n of the misidentification recognition pattern P as a candidate character R (i, j) of x (i, j) (Step 116).
Next, if x (i, j) is a numerical value, the probability information generator 14 B adds x (i, j) to R (i, j) (step 117).

  After this, based on R (i, j) thus obtained, the probability information generation unit 14 B determines that the probability that x (i, j) is R (i, j) is high ( Step 118), based on the determination result, the probability information Q (i, j) on x (i, j) indicating that x (i, j) is likely to be R (i, j) Generate (step 119). The recording unit 14C associates the obtained probability information Q (i, j) with x (i, j) of D (i), and records it in the reading result data 13C of the storage unit 13 (step 120).

Next, the probability information generation unit 14B confirms whether extraction of all digits of D (i) is completed (step 120), and if there is an unextracted digit (step 121: NO), the process proceeds to step 111. Return.
On the other hand, when extraction of all digits of D (i) is completed (step 121: YES), accuracy information generating unit 14B transmits accuracy information Q (i) to accuracy information Q (i) of Q (i, j) of each digit j. ) Is generated (step 122). The recording unit 14C associates the obtained probability information Q (i) with D (i), and records the obtained probability information Q (i) in the reading result data 13C of the storage unit 13 (step 123), and ends the series of probability information generation processing.

  Further, in step 113, the misidentification recognition pattern P having a combination of e, m and n that matches the combination of x (i, j), x (i-1, j) and x (i + 1, j) is erroneous. When not registered in the recognition pattern data 13A (step 113: NO), the probability information generation unit 14B combines the combinations of e and m that match the combination of x (i, j) and x (i-1, j). The registration presence or absence of the matching false recognition pattern P is confirmed (step 114).

  Here, a case in which a mismatch recognition pattern P having a combination of e and m that matches a combination of x (i, j) and x (i-1, j) is registered in the misrecognition pattern data 13A (step 114: YES), after moving to the above-described step 116, after all matching false recognition patterns P having combinations of e and m that match the combination of x (i, j) and x (i-1, j) The letter n is selected as a candidate letter R (i, j) (step 116).

  Further, in step 114, the matching false recognition pattern P having the combination of e and m that matches the combination of x (i, j) and x (i-1, j) is not registered in the false recognition pattern data 13A. In the case (step 114: NO), the probability information generation unit 14B confirms the registration presence or absence of the misidentification recognition pattern P having e that matches x (i, j) (step 115).

  Here, when the matching false recognition pattern P having e matching the x (i, j) is registered in the false recognition pattern data 13A (step 115: YES), the process proceeds to step 116 described above, and x A letter n after all matching false recognition patterns P having e matching (i, j) is selected as a candidate letter R (i, j) (step 116).

In addition, in step 115, when the matching false recognition pattern P having e matching the x (i, j) is not registered in the false recognition pattern data 13A (step 115: NO), the process proceeds to step 121 described above. .
In this case, since the recognition result of the target result x (i, j) extracted from the target recognition result D (i) is correct, it is not necessary to generate the probability information Q (i, j). It should be noted that the process may move to step 120 after the probability information Q (i, j) indicating that the recognition result of the target result x (i, j) is correct (no error).

[Operation example]
Next, with reference to FIG. 7, an operation example of the accuracy information generation process according to the present embodiment will be described. FIG. 7 is an explanatory view showing an example of the accuracy information generation operation.
Here, the target recognition result D (i) at time Ti is “a8”, and the immediately preceding recognition result D (i + 1) at time Ti−1 immediately before that is “30”, and the recognition result D immediately after time Ti + 1 A case where (i + 1) is “29” and each of which is composed of two-digit characters will be described as an example. It is assumed that the content shown in FIG. 4 is registered in advance in the erroneous recognition pattern data 13A of the storage unit 13.

  First, in the probability information generation process for the first digit (j = 1), the target character x (i, 1) = “a”, the immediately preceding character x (from D (i), D (i + 1), D (i + 1) i-1, 1) = "3", the immediately following letter x (i + 1, 1) = "2" is extracted. Next, an erroneous recognition pattern having a combination of the switching character e, the previous character m, and the subsequent character n which matches the combination of “a”, “3” and “2” is registered in the erroneous recognition pattern data 13A Are checked.

  Here, in the case where a misidentification recognition pattern P having a combination of e = “a”, m = “3”, n = “2” is registered, n = “2” of P is x (i, 1) The probability information Q (i, 1) is generated which is selected as the candidate character R (i, 1) of x and that x (i, 1) is likely to be “2”.

  Also, in the probability information generation process for the second digit (j = 2), from D (i), D (i + 1), D (i + 1), target character x (i, 2) = “8”, immediately preceding character x ( i-1, 2) = "0", and the immediately following character x (i + 1, 2) = "9" is extracted. Next, a misrecognition pattern having a combination of the switching letter e, the previous letter m, and the subsequent letter n that matches the combination of “8”, “0”, and “9” is registered in the misrecognition pattern data 13A. Are checked.

  Here, when a false recognition pattern P having a combination of e = “8”, m = “0”, n = “9” is registered, n = “9” of P is x (i, 2) Are selected as candidate characters R (i, 2) of. Also, since x (i, 2) is a number, x (i, 2) is added as a candidate character R (i, 2). As a result, probability information Q (i, 2) indicating that x (i, 2) is likely to be “8-9” (8 or 9) is generated.

  Thereafter, candidate character R (i, 1) of x (i, 1) and candidate character R (i, 2) of x (i, 2) are synthesized, and candidate character R (D) of D (i) is synthesized. i) = "22, 28-29" is generated. As a result, probability information Q (i) indicating that D (i) is likely to be “22, 28-29” (22, 28 or 29) is generated.

[Effect of this embodiment]
As described above, according to the present embodiment, the storage unit 13 displays the front character m and the rear character n that are switched and displayed before and after the segment display 20 and the switching character e that can be generated at the time of the switching display. The misrecognition pattern P consisting of a combination is stored as misrecognition pattern data 13A, and the probability information generation unit 14B extracts the target recognition result D (i), the immediately preceding recognition result D (i-1), and the immediately subsequent recognition result from the time series data 13B. D (i + 1) is obtained, and from these, for each digit position j, the character located at the same digit is the target character x (i, j), the immediately preceding character x (i-1, j), and the immediately following character x (i + 1) , J), and a matching misrecognition pattern P having e, m, n matching them is registered in the misrecognition pattern data 13A, candidate character R (i, j) estimated from the relevant p Based on x ( , In which so as to generate the accuracy information j) Q (i, j).

  As a result, even when the display data displayed and output by the segment display unit 20 does not increase or decrease at a constant rate in a constant direction, it is possible to obtain an appropriate recognition accuracy. Therefore, the display data output by the segment display unit 20 is also used for devices other than integrated quantity meters that measure integrated quantities of gas, electricity, and water, for example, various apparatuses such as thermometers, pressure gauges, and flow meters. While reading, it is possible to provide appropriate recognition accuracy.

Further, in the present embodiment, when the matching false recognition pattern P is registered in the false recognition pattern data 13A, the probability information generation unit 14B selects a character after the matching false recognition pattern P as a candidate character R (i, It may be selected as j).
As a result, it is possible to accurately estimate the target character x (i, j) based on the subsequent character n of the matching error recognition pattern P registered in the error recognition pattern data 13A.

Further, in the present embodiment, when the matching false recognition pattern P is not registered in the false recognition pattern data 13A, the probability information generation unit 14B targets the switching-time letter e and the preceding letter m in the false recognition pattern data 13A. From all the misrecognition patterns that match the character x (i, j) and the immediately preceding character x (i-1, j), the character n after the misrecognition pattern is selected as a candidate character R (i, j) May be
As a result, even if the matching misrecognition pattern P is not registered in the misrecognition pattern data 13A, a similar misrecognition pattern P that matches the target character x (i, j) and the immediately preceding character x (i-1, j) The target character x (i, j) can be accurately estimated on the basis of the letter n after.

Further, in the present embodiment, when the matching false recognition pattern P is not registered in the false recognition pattern data 13A, the probability information generation unit 14B determines that the switching time letter e in the false recognition pattern data 13A is the target character x (i The character n after the misrecognition pattern may be selected as a candidate character R (i, j) from all the misrecognition patterns matching the j, j).
As a result, even if the matching misrecognition pattern P is not registered in the misrecognition pattern data 13A, the target character x is obtained based on the subsequent character n of the similar misrecognition pattern P that matches the target character x (i, j). (I, j) can be estimated accurately.

Further, in the present embodiment, in the case where the segment display 20 is a 7-segment display that switches and displays a plurality of types of numbers by turning on and off 7 segments, the target character x (i, j) is If it is a number, the probability information generation unit 14B may add x (i, j) to the candidate character R (i, j).
Thus, the probability information that the target character x (i, j) in the process of being switched and displayed from the immediately preceding character x (i−1, j) to the immediately subsequent character x (i + 1, j) is the recognition result, Can be included.

[Extension of the embodiment]
As mentioned above, although this invention was demonstrated with reference to embodiment, this invention is not limited to the said embodiment. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention. Moreover, about each embodiment, it can combine arbitrarily and can be implemented in the not conflicting range.

  In the above-described embodiment, the segment display 20 is described as an example of a 7-segment display in which numbers are switched and displayed using 7 segments, but the number of segments is not limited to 7. For example, the present invention is similarly applicable to segment displays having other numbers of segments, such as 9, 14, 16.

  Further, in FIG. 4 described above, although the case where each erroneous recognition pattern is registered in the vertical and horizontal matrix form in the erroneous recognition pattern data 13A has been described as an example, the present invention is not limited to this. For example, an erroneous recognition pattern consisting of a combination of the front character m and the rear character n and the switching character e (m, n) may be registered in another format such as a list format.

  Further, in the misrecognition pattern data 13A, candidate characters estimated from the respective misrecognition patterns are registered in advance in the misrecognition pattern data 13A in association with the respective misrecognition patterns, for example, in step 115 of FIG. The probability information generation unit 14B may extract the candidate character R (i, j) from the corresponding false recognition pattern P.

Further, in the embodiment described above, outlier determination may be performed on the object recognition result D (i), and when it is determined that the outlier is, for example, the probability information Q (i) of “low” may be generated.
FIG. 8 is an explanatory view of outlier determination. When the target recognition result D (i), the immediately preceding recognition result D (i-1), and the immediately subsequent recognition result D (i + 1) are obtained from the time-series data 13B of the storage unit 13, the probability information generation unit 14B The difference value ΔD = D (i) −D (i−1) or the difference ΔD = D (i) −D (i + 1) is determined, and the outlier determination is performed by comparing with the allowable range ε set in advance. You may

Here, when ΔD is outside the allowable range ε and ΔD is an outlier, the accuracy information generating unit 14B generates accuracy information Q (i) indicating that the recognition accuracy regarding D (i) is low.
At this time, for example, when it is confirmed in step 110 of FIG. 6 that ΔD is out of the allowable range ε, the process may proceed to step 122 after the probability information Q (i) is generated.
Note that the method of determining the outlier is not limited to the comparison between the difference ΔD and the tolerance range ε, and, for example, the ratio of D (i) to D (i-1) or D (i + 1) Known methods may be applied.

If it is determined that the object recognition result D (i) is an outlier, the immediately subsequent recognition result D (i + 1) may be adopted as the object recognition result D (i).
One of the methods of outlier determination is a method of using the root mean square of the difference between adjacent time series data. FIG. 9 shows an example of outlier determination using the difference root mean square. Here, an example in which the d6 outlier determination is performed based on d1 to d5 is shown, and assuming that the difference between d1 to d5 is 3, 2, -2, -3, respectively, the difference root mean square of these differences The square root r is
r = α × √ {(9 + 4 + 4 + 9) / 5} = 6.84
It becomes. Here, α represents a magnification, and in this example, α was set to 3.

On the other hand, if the difference between d5 and d6 is 10, then r <10, so d6 is determined to be an outlier. However, when d7 immediately after d6 is d6 <d7, d6 is not an outlier since d5 → d6 → d7 is increased in order.
On the other hand, as described above, when it is determined that the object recognition result D (i) is an outlier, if the immediately subsequent recognition result D (i + 1) is adopted as the object recognition result D (i) It is possible to avoid an erroneous determination when using such a difference root mean square.

  DESCRIPTION OF SYMBOLS 10 ... Segment display reading apparatus, 11 ... Communication I / F part, 12 ... Camera, 13 ... Storage part, 13A ... Misrecognition pattern data, 13B ... Time series data, 13C ... Reading result data, 13P ... Program, 14 ... Calculation Processing unit, 14A: character recognition unit, 14B: probability information generation unit, 14C: recording unit, 20: segment display unit, 30: high-order device, L: communication line, D (i): target recognition result, D (i- 1) ... immediately before recognition result, D (i + 1) ... immediately after recognition result, x (i, j) ... target character, x (i-1, j) ... immediately before character, x (i + 1, j) ... immediately after character, e ... Switching character, m ... previous character, n ... subsequent character, P ... matching false recognition pattern, R (i, j) ... candidate character, Q (i, j), Q (i) ... probability information.

Claims (6)

A camera which picks up a display output of a segment display unit which switches and displays a plurality of characters by turning on and off a plurality of segments;
A character recognition unit configured to perform character recognition processing on an imaging output of the camera;
A storage unit configured to store the recognition result obtained by the character recognition unit as time-series data;
A certainty information generation unit configured to generate certainty information indicating the certainty of the target recognition result for the target recognition result selected from the time-series data;
The storage unit is capable of being generated when switching is performed before and after any two types of front and rear characters switched and displayed by the segment display and before and after these characters. Storing misrecognition pattern data in which a plurality of misrecognition patterns consisting of combinations with characters are registered in advance;
The certainty information generation unit acquires, from the time series data, the object recognition result and immediately preceding and succeeding recognition results positioned immediately before and after the object recognition result with respect to the object recognition result, and these object recognition results The character located at the same position is extracted as the target character, the previous character, and the subsequent character for each digit position from the immediately preceding recognition result and the immediately subsequent recognition result, and the combination of the target character, the immediately preceding character, and the subsequent character In the case where a matching false recognition pattern having a combination of a switching character, a preceding character, and a trailing character that matches with is registered in the false recognition pattern data, the false recognition pattern data is based on candidate characters estimated from the matching false recognition pattern. What is claimed is: 1. A segment display reading device that generates probability information indicating a recognition probability of a target character. In the segment display reader according to claim 1,
The segment display reading apparatus according to claim 1, wherein the certainty information generation unit selects a character subsequent to the false recognition pattern as the candidate character when the false recognition pattern is registered in the false recognition pattern data. In the segment display reader according to claim 1 or 2,
The certainty information generation unit, when the false recognition pattern is not registered in the false recognition pattern data, all the false recognition pattern data in which the switching character and the preceding character match the target character and the immediately preceding character. A segment display reading device, wherein a character subsequent to the erroneous recognition pattern is selected as the candidate character from the erroneous recognition pattern. The segment display reader according to any one of claims 1 to 3.
The certainty information generation unit, when the matching false recognition pattern is not registered in the false recognition pattern data, from among the false recognition pattern data, all false recognition patterns in which the switching character matches the target character A segment display reader characterized by selecting a character after a misrecognition pattern as the candidate character. The segment display reading device according to any one of claims 1 to 4.
The segment display reading device characterized in that, when the target character is a number, the certainty information generator adds the target character to the candidate character. A segment display reading method for use in a segment display reader for reading display output of a segment display, comprising:
An imaging step of imaging a display output of the segment display device in which a camera switches and displays a plurality of types of characters by performing on / off control of a plurality of segments;
A character recognition step, wherein the character recognition unit is configured to perform character recognition processing on the imaging output of the camera;
A storage step of sequentially capturing the recognition result obtained in the character recognition step and storing the result as time-series data;
Providing a probability information generation step configured to generate, for the target recognition result selected from the time series data, the probability information indicating the certainty of the target recognition result;
The storage step may occur when any two types of front and rear characters switched and displayed before and after on the segment display and when these front and rear characters are switched and displayed before and after Including a step of storing false recognition pattern data in which a false recognition pattern consisting of a combination of characters is registered in advance;
The certainty information generation step acquires, from the time series data, the object recognition result and immediately preceding and succeeding recognition results positioned immediately before and after the object recognition result in time series order, and these object recognition results The character located at the same position is extracted as the target character, the previous character, and the subsequent character for each digit position from the immediately preceding recognition result and the immediately subsequent recognition result, and the combination of the target character, the immediately preceding character, and the subsequent character In the case where a matching false recognition pattern having a combination of a switching character, a preceding character, and a trailing character that matches with is registered in the false recognition pattern data, the false recognition pattern data is based on candidate characters estimated from the matching false recognition pattern. A segment display reading method comprising the step of generating accuracy information indicating the recognition accuracy of a target character.

Images