JPH0729643U - Figure recognition device - Google Patents

Abstract

(57) [Summary] [Purpose] Highly accurate recognition is possible even for figures engraved on curved surfaces. [Structure] A CCD line sensor camera 3 is configured to detect the characteristics of reflected light from a figure P engraved on the curved surface of the target object 1 as one-dimensional image data, and the rotating means 4 moves the target object relative to the camera. While one-dimensionally rotating, one-dimensional image data obtained from the camera is temporally accumulated according to the rotation speed. This data is cut out as two-dimensional image data, the comparison means 7 extracts a figure, and then the figure extracted by the verification means 19 and the learning data accumulated in the learning means 18 are compared and verified to obtain a curved surface image. It is processed as a flat image without distortion and recognized with high accuracy.

Description

[Detailed description of the device]

[0001]

[Industrial applications]

 The present invention relates to a recognition device for letters, numbers or symbols (collectively referred to as “graphics” hereinafter), and in particular, a graphic engraved on a curved surface of an object is read using an imaging means and the recognition is performed using a learning method. Regarding the recognition device.

[0002]

[Prior art]

 The outer shell of the accumulator used for the hydraulic system of automobiles is engraved with figures such as manufacturing numbers and trademarks. Such engraved figures are widely used as a medium to provide important information for production instruction / production control in the assembly process and inspection process, or quality control in the market, because the readability is not lost even in an environment that is highly susceptible to aging and easily polluted. It's being used. For example, in the inspection process of the accumulator, the inspection operator reads the engraved manufacturing number and records the inspection data on the recording sheet corresponding to the manufacturing number. Further, as a figure recognition device for automatically recognizing such a figure, those disclosed in JP-A-63-66,684 and JP-A-64-36,387 are known. .

[0003]

[Problems to be solved by the device]

 However, if the operator reads the figure as in the past, even if the inspection data can be acquired automatically, the inspection record cannot be completely automated. However, although the automation of inspection records can be achieved by using the conventional pattern recognition device, no device has been developed that can accurately read even the pattern imprinted on the curved surface of the object. When the figure recognition for such a curved surface is processed by the two-dimensional image data, the image data is distorted and the recognition accuracy is lowered, and the illumination method for the curved surface also has various problems. Moreover, although the conventional figure recognition device discloses that the illumination means is used when the figure is read by the imaging means, the specific illumination method has not been examined, and therefore, the object is illuminated. At that time, the actual situation was to search for the lighting method each time so that the image captured by the imaging means would be the optimum conditions. Therefore, it is necessary to adjust the illumination each time the object changes, and if this is not done, there is a problem that recognition accuracy and recognition time are adversely affected.

As described above, in the figure recognition, the high accuracy of the figure recognition, the reduction of the recognition time, the ease of the recognition method, and the simplicity of the device configuration for realizing the method are the problems to be solved. In particular, when applying figure recognition technology to figure automatic identification in machine tools, industrial robots, etc., the real-time property required for recognition is realized while ensuring practical figure recognition accuracy. It is important to provide a figure recognition device with a simpler structure.

The present invention has been made in view of the above problems of the prior art, and it is an object of the present invention to provide a figure recognition device capable of recognizing a figure engraved on a curved surface with high accuracy. To do.

[0006]

[Means for Solving the Problems]

 In order to achieve the above-mentioned object, the figure recognition device of the present invention is an illumination means for irradiating light from at least two different directions to a target object in which a figure such as a character, a number or a symbol is engraved on a curved surface, and the target object. An image pickup means for picking up an image of a figure imprinted on the object to detect the characteristic of the reflected light by the illumination means as one-dimensional image data; and a target object for at least the figure so as to be picked up by the image pickup means. Rotating means for rotating relative to the image capturing means, feature storing means for storing characteristic items of a standard graphic pattern corresponding to the graphic, and one-dimensional image data captured by the image capturing means for the rotating means. Image storage means for storing the two-dimensional image data according to the rotation speed of the, and extracting an actual figure pattern from the two-dimensional image data stored in the image storage means. Comparing means for comparing and verifying the extracted actual figure pattern and the characteristic items of the standard figure pattern previously stored in the characteristic storing means, and comparing means for extracting the figure corresponding to the standard figure pattern having the strongest correlation; Learning means for accumulating the characteristic value of each characteristic item for each figure extracted by the comparing means; and the figure extracted by the comparing means and the learning data of the figure accumulated in the learning means. It is characterized by having verification means for comparative verification.

[0007]

[Action]

 When a figure such as letters, numbers or symbols is imprinted on the curved surface of the target object, if this figure is directly captured as a two-dimensional image, the image data will be distorted, and the recognition accuracy will be reduced depending on the lighting arrangement. There is a full possibility of fear. However, if the curved surface image is regarded as a temporal set of one-dimensional images, that is, as in the present invention, the image pickup means is configured to detect the characteristic of the reflected light from the figure imprinted on the target object as one-dimensional image data. By rotating the target object relative to the image pickup means and temporally accumulating the one-dimensional image data obtained from the image pickup means in accordance with the rotation speed, the curved surface image can be transformed into a flat image (two-dimensional image). ) Can be treated as Moreover, the two-dimensional image data thus obtained is compared and verified with the characteristic items of the standard graphic pattern stored in advance in the characteristic storage means, and then with the accumulated learning data. Since it is performed, the figure corresponding to the standard figure pattern having the strongest correlation can be recognized with high accuracy as the desired figure.

[0008]

【Example】

 An embodiment of the present invention will be described below with reference to the drawings. 1 is a block diagram showing a figure recognition device according to an embodiment of the present invention, FIG. 2 is a partially sectional front view showing a rotating means according to the embodiment, and FIG. 3 is a recognition target according to the embodiment. FIGS. 4 and 5 are diagrams showing specific examples of characteristic items according to the same embodiment, and FIG. 6 is a learning means according to the same embodiment. FIG. 7 is a diagram showing accumulated data in FIG. 7, and FIG. 7 is a flowchart showing an information processing procedure of the same embodiment.

The figure recognition device shown in FIG. 1 is composed of a lighting unit 2 including a light source 2a and an optical fiber cable 2b, a line sensor camera (imaging unit) 3 using a CCD, an analog / digital conversion unit 8, a RAM and the like. Image storage means 6, a comparison means 7 composed of a personal computer or the like, a verification means 19 also composed of a personal computer or the like, a feature storage means 5 composed of a ROM or the like, a learning means composed of a RAM or the like. 18 and output means 9 composed of a display, a printer and the like.

In this figure recognition device, the line sensor camera 3 is provided with a string of numbers “0” to “9” and a string of letters “X” to “Z” stamped on the surface of the target object 1. So that the captured analog image data is converted into digital image data by the analog / digital converter 8 and the obtained one-dimensional image data is stored in the image storage means 6 in a time sequence. It is configured.

Further, in order to apply good illumination to the stamped numeral string and the stamped alphabetic character string, the lighting unit 2 including a halogen lamp or the like having a frequency corresponding to the reading frequency of the line sensor camera 3 is provided. In the present embodiment, two optical fiber cables 2b, 2b are arranged in consideration of the reading accuracy of the line sensor camera, and their optical axes are set so that the curved surface of the target object is irradiated with light from different directions. .

In this embodiment, a rotating means 4 for rotating the target object 1 is provided so that the line sensor camera 3 can capture an image of a number string or a character string engraved on a curved surface such as the outer shell of the accumulator 1. The rotating means 4 is configured, for example, as shown in FIG. That is, as shown in the figure, a rotating plate 12 is rotatably provided on a base 10 via a bearing 11, and the accumulator 1 is placed in an opening 13 formed in the rotating body 12. . On the other hand, a motor 14 is fixed to the base 10, and a gear 15 that meshes with a gear 16 attached to the rotating body 12 is attached to a rotation shaft 14 a of the motor 14. As a result, when the motor 14 is driven, the rotation of the rotating shaft 14a is transmitted to the rotating body 12 via the gears 15 and 16, and the accumulator 1 mounted on the rotating body 12 rotates at a constant speed. . By replacing the plate 17 in which the opening 13 for accumulating the accumulator 1 is formed, the accumulator 1 can be used for various accumulators.

Further, the tip ends of the two optical fiber cables 2b, 2b are fixed below the accumulator 1 placed in the opening 13 so that the position of the reflected light is variable, as described above. It is incident on 3. Further, since the position of this line sensor camera 3 is also fixed variably, when the specification of the accumulator is changed, the position most suitable for incident reflected light can be finely adjusted together with the two optical fiber cables 2b, 2b. .

When the one-dimensional images obtained by the line sensor camera 3 are sequentially stored in the image storage means 6, the relationship between the rotation speed of the rotating body 12 and the reading speed by the line sensor camera 3 is confirmed. Need to be kept. Therefore, the rotation speed of the rotator 12 is set to a constant speed, or the rotation position of the rotator 12 is sequentially fetched and stored in the comparison means 7 via an interface or the like.

The comparison means 7 composed of a personal computer or the like reads the two-dimensional image data stored in the image storage means 6 character by character, processes the image data, and processes the embossed numeral string and the embossed alphabetic characters. Recognize columns. When recognizing a figure, the characteristic items of the target numeral string and alphabetic character string are listed in advance, and the reference value of the standard figure pattern is determined for each characteristic item. Then, the image data read out from the image storage means 6 is compared and verified for each characteristic item, and the standard graphic pattern having the highest matching rate is extracted.

Since the figures to be recognized in the present embodiment are the numerical strings “0” to “9” and the alphabetic characters “X” to “Z”, specifically as shown in FIG. The character width (see feature 19 in FIG. 5), the parallelism at the bottom of the character (see feature 2 in FIG. 5), or the parallelism at the top of the character (see feature 1 in FIG. 5), etc. The characteristic items are determined according to the figure to be recognized. Then, as shown in FIG. 5, 19 characteristic items are determined, and the reference range (MAX value, MIN value) of the standard graphic pattern for each characteristic item is determined as shown in FIG. More specifically, the feature 0 shown in FIG. 5 is set to “1” when the image is in the right half when scanning from the center to the outside with respect to the right half image, and is set to “0” in other cases. ". This feature item helps to distinguish between the letters "X" and "Y" as shown in FIG. Similarly, the characteristic item 18 is the number of dots in the vicinity of the center of the image, and this characteristic item helps to identify the numbers "0" and "8" as shown in FIG.

The contents of the characteristic items and the reference values stored in the characteristic storage means 5 are not limited to those in the above-described embodiment. This is appropriately determined depending on the type of figure to be recognized. When the data read from the image storage means 6 is applied to the reference value of the standard graphic pattern stored in the characteristic storage means 5 for comparison and verification, the standard graphic pattern with the highest matching rate is recognized as the target graphic. However, if the matching rate is less than a certain value, it is considered that the data stored in the image storage unit 6 is defective and re-verification is performed, or the reading by the line sensor camera 3 is defective. You may handle it.

In particular, in this embodiment, the learning means 18 for accumulating actual data relating to the characteristic items of the graphic extracted by the comparing means 7, and the graphic extracted by the comparing means 7 are really accumulated in the learning means 18. And a verification means 19 for verifying whether or not the learning data matches the existing learning data. The learning means records, for each characteristic item, actual feature values cut out from the image storage means 6 for a certain figure (figure recognized by the comparison means 7), for example. Accumulate values. For example, as shown in FIG. 6 (A), the actual characteristic value for a certain figure (for example, the number “1”) should vary with respect to the characteristic item 1. From this point of view, the relationship between the characteristic value and the frequency is accumulated for each characteristic item as shown in FIG. 6 (A), and the characteristic value (feature number) is accumulated as shown in FIG. 6 (B). ) Is recorded as a range of feature values for The learning data in the learning means 18 is accumulated every time the graphic is extracted by the comparison means 7, and suggests the allowable range of the characteristic value for a certain characteristic item as shown in FIG. 6B.

On the other hand, in the verification means 19, the learning accumulated in the learning means 18 described above is performed in order to improve whether or not the figure extracted by the comparison means 7 is a target figure, that is, the recognition accuracy. Compare and verify with the data. For example, if the character extracted by the comparison means 7 is "1", the verification means 19 sends to the learning means 18 a command signal for retrieving learning data relating to the character "1", and the retrieved learning data and The data extracted by the comparison means 7 is verified for each characteristic item. If there is an item having an extreme feature with respect to a particular figure among the plurality of feature items, the comparison and extraction of data by the comparison unit 7 is omitted, and the direct learning unit 18 and the verification unit 19 are accessed. You may. In addition, since the most characteristic item among a plurality of characteristic items differs depending on the figure, it is also effective to "weight" the characteristic item from this viewpoint.

Next, the operation will be described. Steps 1 to 3 A line sensor camera 3 as an image pickup means picks up an embossed numeral string and an embossed alphabetic character string of an accumulator which is mounted on a rotating means through a camera lens portion and rotates at a constant speed. The picked-up image data is, for example, black and white analog image data which is taken in at an interval of 1 MHz and is arranged in a one-dimensional manner and has light and shade (gradation). The analog / digital converter 8 converts the analog image data into digital image data and applies it to the image storage means 6 in a time sequence. In this case, the intensity (output voltage) of the one-dimensional image data captured by the line sensor camera 3 is observed as the accumulator 1 rotates, and writing to the image storage means 6 is performed from the time when the presence of the marking pattern is confirmed. . Similarly, the end of writing is also controlled based on the output voltage of the line sensor camera 3.

Step 4 The one-dimensional image data thus written in the image storage means 6 in the time sequence is cut out in units of the time sequence and taken in as the two-dimensional image data in the comparison means 7. The two-dimensional image data is not an actual two-dimensional image captured as it is, but is created by the image storage means 6 based on the one-dimensional image data. Even a curved surface can be processed as a flat image without distortion.

That is, when a figure such as a character, a number, or a symbol is imprinted on the curved surface of the target object, if the figure is directly captured as a two-dimensional image, the image data may be distorted, or depending on the lighting arrangement. Although there is a strong possibility that the recognition accuracy may decrease, the curved surface image is regarded as a temporal set of one-dimensional images, and the one-dimensional image data such as the line sensor camera 3 which is less influenced by the curved surface is used and the target object 1 By rotating this image with respect to the line sensor camera 3 and temporally accumulating this one-dimensional image data according to the rotation speed, it is possible to process a curved surface image as a plane image (two-dimensional image) without distortion. When outputting the image data stored in the image storage means 6 to the comparison means 7, pre-processing (correction) for removing noise components may be performed.

[0023] Upon completion of the rotation of the accumulator 1 according to step 5-6 the rotation means 4, cut out of the image data memorize the image storage unit 6. The following description describes the case of processing one digit (one figure) of cut image data. First, each of the characteristic items 0 to 19 (see FIGS. 4 and 5) stored in the characteristic storage means 5 is verified with respect to the one-digit image data cut out in this way. By this verification, verification values of the characteristic items 0 to 19 are obtained for the image data. Then, the verification value thus obtained is compared with the reference value (see FIG. 3) of the standard graphic pattern stored in the feature storage means 5 to determine the graphic having the highest matching rate. If the matching rate at this time is less than the predetermined value, the verification is performed again or it is processed as unrecognizable.

Steps 7 to 11 If the concordance rate is equal to or more than a predetermined value, it is checked whether the number of accumulated data in the learning means 18 is sufficient for effective verification. In the verification method using learning data, effective verification cannot be performed unless a sufficient amount of learning data is secured. Therefore, for example, it is also effective to accumulate the dummy data before performing the figure recognition on the actual accumulator. When the number of accumulated data in the learning means 18 is sufficient, the data including the characteristic value from the comparing means 7 is sent to the verifying means 19 and the learning data accumulated in the learning means 18 is verified. For example, if the character extracted by the comparison means 7 is "1", the verification means 19 sends to the learning means 18 a command signal for retrieving learning data relating to the character "1", as shown in FIG. The data extracted by the comparison means 7 is applied to the learned data retrieved as shown in each characteristic item. As a result of this verification, if they match, the feature value is stored in the learning means 18 as new learning data, and the recognized figure is output to the output means 9 such as a display or a printer.

It should be noted that the embodiments described above are provided for facilitating the understanding of the present invention, and are not intended to limit the present invention. Therefore, each element disclosed in the above embodiments includes all design changes and equivalents within the technical scope of the present invention. For example, in the above embodiments, recognition was performed when numbers and letters were imprinted as figures, but the figure recognition device of the present invention is not limited to numbers and letters, and other figures, such as special marks. It can also be applied to the recognition such as. Further, the present invention is not limited to the stamped figure, and can be applied to a printed figure or a case where paint is applied to the stamped figure.

[0026]

[Effect of device]

 The figure recognition device of the present invention is configured such that the image pickup means detects the characteristics of the reflected light from the figure imprinted on the target object as one-dimensional image data, and the target object is rotated relative to the image pickup means. However, since the one-dimensional image data obtained from the image pickup means is temporally accumulated according to the rotation speed, the curved surface image can be processed as a plane image without distortion (two-dimensional image). As a result, it is possible to accurately recognize even a figure engraved on a curved surface. Further, since the illumination means illuminates from different directions, the amount of reflected light taken into the imaging means becomes constant, which also improves the recognition accuracy. In addition, since the figure recognition device of the present invention further compares and verifies the figure extracted by the comparison means with the learning data, it is possible to prevent erroneous recognition in the comparison means and improve the recognition accuracy.

[Brief description of drawings]

FIG. 1 is a block diagram showing a graphic recognition apparatus according to an embodiment of the present invention.

FIG. 2 is a partially sectional front view showing a rotating means according to the embodiment.

FIG. 3 is a diagram showing a specific example of a standard graphic pattern (character) to be recognized and respective characteristic items according to the embodiment.

FIG. 4 is a diagram showing a specific example of characteristic items according to the embodiment.

FIG. 5 is a diagram showing a specific example of characteristic items according to the same embodiment.

FIG. 6 is a diagram showing accumulated data in the learning means according to the embodiment.

FIG. 7 is a flowchart showing an information processing procedure of the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Target object 2 ... Illumination means 2a ... Light source 2b ... Optical fiber cable 3 ... CCD line sensor camera (imaging means) 4 ... Rotation means 5 ... Feature storage means 6 ... Image storage means 7 ... Comparison means 8 ... Analog / digital conversion Part 9 ... Output means 10 ... Base 11 ... Bearing 12 ... Rotating body 13 ... Opening portion 14 ... Motor 14a ... Rotating shaft 15 ... Gear 16 ... Gear 17 ... Plate 18 ... Learning means 19 ... Verification means P ... Graphic

Continuation of front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location G06K 9/66 9289-5L

Claims (1)

[Scope of utility model registration request] 1. Illuminating means (2) for irradiating a target object (1) having a curved surface with a figure (P) such as letters, numbers or symbols with at least two different directions, and the target object (1). An image pickup means (3) for picking up the image marked on 1) and detecting the characteristic of the reflected light by the illumination means (2) as one-dimensional image data; and at least the image pickup means for picking up the image by the image pickup means. A rotation means (4) for rotating the target object relative to the imaging means, a characteristic storage means (5) for storing characteristic items of a standard graphic pattern corresponding to the graphic, and the imaging means (3 ) Stores the one-dimensional image data imaged in (1) according to the rotation speed of the rotating means to obtain two-dimensional image data, and the two-dimensional image data stored in the image storage means. Shape pattern Pattern is extracted, the extracted actual graphic pattern is compared with the characteristic items of the standard graphic pattern stored in advance in the characteristic storage means, and the graphic corresponding to the standard graphic pattern having the strongest correlation is extracted. A comparing means (7) and a learning means (18) for accumulating the characteristic value for each characteristic item regarding each figure extracted by the comparing means (7).
And a verification means (19) for comparing and verifying the figure extracted by the comparing means (7) and the learning data of the figure accumulated in the learning means (18). apparatus.