JP4493040B2 - Founded character recognition device - Google Patents

Description

  The present invention relates to a technology for recognizing a cast character on the surface of a casting part such as a cylinder block of an internal combustion engine.

  Cast parts such as cylinder blocks of automobiles have various information such as production lot information and mold combination man-hours, etc. (in this specification, cast characters are not only letters but also numbers, symbols, codes, marks) It is used in the meaning of including.) These are transcribed from the sand mold, and the numbers will appear in the product. The cast lot production lot information is recorded from the viewpoint of traceability, but it has been written manually.

  Conventionally, a cast character cast into a cast part is imaged by a CCD camera (electrostatic coupling device camera), the imaged cast character is patterned by an image processing means, and compared with a pre-registered pattern ( A device that recognizes a character by performing pattern matching processing and determines a model to be determined based on the recognized character is known. (For example, see Patent Document 1.)

  However, when such a pattern matching processing method is used, recognition errors of cast characters such as cast iron cylinder blocks are particularly large, and cannot be used in an actual production line. The reason for this is that in the case of cast characters such as cast iron cylinder blocks, it is painted black so as not to rust. This is because it does not appear and it cannot be applied to the pattern matching processing based on the density processing of the captured image by the CCD camera.

In addition, when the cast character represents the date, it is necessary to change the mold every day, and since the inset resin is used, the character is largely crushed and deformed at the stage of transfer to the sand mold, and the shape and shape are molded. The accuracy becomes indefinite, and the variation in the difference between the master data of the pattern matching process and the individual measurement data increases.
In addition, since the shape and molding accuracy of cast characters are indefinite, some feature values (for example, the position of the center of gravity of the binary image, area, moment, etc.) are extracted from the characters captured by the CCD camera, and their features are extracted. It can be expected that the recognition error will increase even if the character is recognized by a method such as clustering in the space.

  Furthermore, it is practically difficult to maintain quantitative illuminance on the object in-line, and by performing feature extraction processing by binary image processing of the image captured by the CCD camera, the cast iron cylinder block, etc. It is also very difficult to recognize cast characters.

Japanese Patent Laid-Open No. 10-198804

  In order to cope with the above-described problems, the main object of the cast character recognition apparatus of the present invention is to increase the recognition accuracy of cast characters such as a cast iron cylinder block. Further, the cast character recognition device of the present invention is capable of performing measurement within an allowable time by increasing cast character recognition accuracy, and can be used in an actual production line (inline). Purpose.

  In addition, the cast character recognition device of the present invention does not use general image processing (shading processing or binary processing) such as pattern matching processing or feature amount extraction processing based on a photographed image, The object is to recognize characters by identifying irregularities and converting them into three-dimensional data and reproducing the cast characters on a computer in a pseudo manner.

  Furthermore, it is an object of the cast character recognition apparatus of the present invention to recognize a cast character by a method close to human character recognition logic.

In order to achieve the above object, a cast character recognition apparatus according to the present invention includes a three-dimensional measurement unit that measures the uneven shape of a character / numeric region by casting a casting, and a three-dimensional measurement obtained by the three-dimensional measurement unit. Image data conversion means for performing binarization processing after converting the data in the height direction of the data to gray scale processing and converting to two-dimensional image data, and information obtained by previously learning the image data obtained by the image data conversion means The three-dimensional measuring unit projects linear light onto the character / numeral area, and the image formed by the linear light drawn on the surface of the character / numeral area. In order to detect the partition position of each letter and numeral from the letter and numeral area, using a light cutting method to obtain the three-dimensional shape data of the letter and numeral area by imaging
a) Calculate the peak value of the displacement data obtained by the light cutting method,
b) Obtain a profile in which the calculated peak value of the displacement data is plotted in the direction of the character string,
c) calculating the barycentric position of each letter and numeral at a position where the area obtained by sequentially moving the area corresponding to the letter and numeral width from the profile in the letter and numeral string direction becomes a peak value;
d) It is characterized in that a valley position of the displacement data existing at a substantially middle point with respect to the gravity center position of two adjacent letters and numbers is calculated .

Here, the reason why the three-dimensional measuring means for measuring the uneven shape of the letters and numbers by casting is as follows.
As a general character recognition method, two-dimensional image processing using a CCD camera or the like is used, but the shadow of the character cast on cast iron has the same color as the background and the shape is irregular. This is because the recognition accuracy using only two-dimensional information does not reach the operational level.
In addition, even if you try to detect cast characters with a CCD camera, perform image processing, and discriminate by pattern matching, disturbance light will enter the measurement target range, and light reflection will not be constant, and images that can be pattern matched It was because he knew from experience that he could not get the

Next, the reason why the image data conversion means for converting the data in the height direction of the three-dimensional measurement data to gray scale processing and converting it into two-dimensional image data will be described.
When the cast character represents a date, it is necessary to change the mold every day, and since the inset resin is used, as described above, at the stage of transfer to the sand mold, the character is greatly crushed and deformed, and the shape and Molding accuracy is indefinite. Further, the cast character itself is not necessarily formed at a right angle to the surface. For this reason, the outline of characters is very varied at this casting stage.
In other words, a cast character is formed at a right angle, and the unevenness of the character is uniform throughout the character and the outline of the character is clear, while a cast character is formed with an inclination, and the unevenness of the character is uneven across the character. In some cases, the character outline is unclear. Accordingly, the three-dimensional measurement data of the cast characters are also very varied data, and cannot be said to be data suitable for character recognition, and reprocessing for the data is necessary.
Therefore, by converting the data in the height direction of the 3D measurement data to grayscale processing and converting it to 2D image data, the clarity of the character outline due to this variation in casting accuracy is converted to grayscale brightness information. We decided to do data processing.

  Then, it was the characters cast on the surface of cast iron having the above-mentioned characteristics that decided to distinguish the letters and numbers based on information learned in advance from the two-dimensional image data converted from the three-dimensional measurement data. However, since most of these characters can be identified by the human eye (brain), recognition accuracy is improved by discriminating characters by a method that approximates the learning decision logic that is human character recognition logic. This is because it was thought that it could be increased. In fact, as will be described later in the embodiments, the character recognition accuracy using the cast character recognition device of the present invention is very superior to the character recognition accuracy by the conventional pattern matching process. It was.

Here, the three-dimensional measuring means for measuring the uneven shape of the letters and numerals by casting the casting projects linear light on the letter and numeral area to be measured, and forms the linear light drawn on the surface of the letter and numeral area. It is preferable to use an optical cutting method that obtains the three-dimensional shape data of the character / numeric region by capturing an image.
As described above, the cast character is a character whose surface is black, glossy or non-glossy, the background is the same color, and the shape of the character is unstable due to casting or crushing. is there. For this reason, the light cutting method which can perform stable sensing without being affected by the surface color and state of the measurement object is preferable.
In addition, the processing algorithm of the light section method is simple and real-time measurement can be secured, so that it can be used in-line and it is robust and suitable for recognition of cast characters for this measurement. Because. Furthermore, it is also advantageous that an optical system that is a component of the apparatus can be constructed at low cost.
Note that a 3D scanner that can capture the state of the shape surface of the three-dimensional object as continuous numerical information such as inclination, depth, and distance may be used as the three-dimensional measurement unit. However, it should be noted that when a 3D scanner is used, a certain amount of time is required for measurement.

Further, the three-dimensional measuring means is characterized in that the projection area of the casting is used to detect the position of the character / numeric area to be measured. Here, the projection region of the casting corresponds to, for example, a processing reference boss.
In the three-dimensional measurement using the above-described optical cutting method, the projection area such as the casting processing reference boss is used so that the position of the character / numeric area can be estimated within the scanning range of the laser beam for optical cutting. I decided to do it.

Further, in the three-dimensional measuring means, in order to detect the partition position of each alphanumeric character from the alphanumeric character region to be measured,
a) Calculate the peak value of the displacement data obtained by the light cutting method,
b) Obtain a profile in which the calculated peak value of the displacement data is plotted in the direction of the character string,
c) calculating the barycentric position of each letter and numeral at a position where the area obtained by sequentially moving the area corresponding to the letter and numeral width from the profile in the letter and numeral string direction becomes a peak value ;
d) calculating the root position of the displacement data that exists in a substantially intermediate point relative to the center of gravity of two adjacent alphanumeric,
The reason is that the height information of the three-dimensional measurement data is used in consideration of the characteristics of the cast characters . This will be described in detail in an embodiment described later.

  In the three-dimensional measuring means, the line width of the linear light projected onto the character / numerical area to be measured by the light cutting method is 100 μm to 500 μm. Laser light using the light cutting method is condensed into a linear shape by a lens to form linear light. By narrowing the line width of this linear light, it is possible to accurately measure the step of cast characters. Can do it. When the line width of the linear light is larger than 500 μm, it is difficult to measure the level difference of the cast character cast into the cast iron. Moreover, it can be said that making the line width of the linear light smaller than 100 μm is not practical because it is difficult to adjust the optical system.

In the three-dimensional measuring means, the wavelength of the linear light projected onto the character / numerical region to be measured by the light cutting method is 650 nm to 900 nm.
By using a red laser whose wavelength of linear light projected on the character / numeric area to be measured by the light cutting method is 650 nm to 900 nm, it is possible to construct an environment that is not affected by diffuse reflection due to ambient light or changes in reflectance. is there.

Further, the character / numeral discrimination means uses a learning method based on a neural network. In particular, the neural network is preferably a multi-layer perceptron. Here, the multi-layer perceptron is an algorithm that is also used for handwritten character recognition by a neural network, and effectively acts on data learning with variations obtained by measuring cast characters.
Recognition accuracy could be improved by character recognition using a neural network learning method approximated to learning decision logic, which is human character recognition logic.

  According to the cast character recognition device of the present invention, the recognition accuracy of cast characters such as cast iron cylinder blocks is high, and can be measured within the allowable measurement time on the line and used in an actual production line (inline). There is an effect of having an operational level to obtain.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. However, the scope of the present invention is not limited to the illustrated examples.

FIG. 1 shows a block diagram of a device configuration of a cast character recognition device according to the first embodiment.
In FIG. 1, characters are cast on the surface of a cylinder block 2 made of cast iron, and the cylinder block 2 is sequentially conveyed by a belt conveyor 3 in the direction of the arrow in the figure.
In the first embodiment, an optical cutting method is used as a three-dimensional measuring means for measuring the uneven shape of letters and numbers by casting a casting, and the three-dimensional shape is measured by the laser scanning sensor 8. A line beam is irradiated from the laser scanning sensor 8 to the character region of the cylinder block 2 to be measured, and the reflected light is detected to measure the three-dimensional shape. The actuator 9 is a driving device for scanning laser light.
In order to increase the measurement accuracy of the three-dimensional shape by the laser scanning sensor 8, the belt conveyor 3 is provided with a stopper 6 and pushers (7a, 7b), and the position of the conveyed cylinder block 2 during measurement is substantially constant. It is in a state. In addition, position detection sensors A and B (4, 5) are provided to detect the position of the cylinder block 2 conveyed by the belt conveyor 3.

  The motor control of the belt conveyor 3, the signal processing from the position detection sensors A and B, and the operation control of the pushers (7 a and 7 b) are performed by a control device (programmable controller; PLC) 10. A control device (programmable controller; PLC) 10 is connected to a computer 13 via a communication cable, and the computer 13 monitors and maintains the control device (programmable controller; PLC) 10. Here, a specific processing operation flow (operation flow) is not a point of the present invention, and will be omitted.

The laser scanning sensor 8 is distributed and controlled by a laser scanning sensor controller 11, and the actuator 9 is distributed and controlled by an actuator controller 12. The laser scanning sensor 8 and the laser scanning sensor controller 11 are connected by a dedicated cable 24, and the three-dimensional data acquired by laser scanning is sent to a computer (PC) 13 via a USB (Universal Serial Bus) cable 21. ing. The actuator 9 and the actuator controller 12 are also connected by a dedicated cable, and the actuator controller 12 is controlled by the computer (PC) 13 via the RS-232C cable 22.
When the laser scanning sensor 8 captures one line of data, the data is sent from the laser scanning sensor controller 11 to the computer 13, a movement signal is output from the computer 13 to the actuator controller 12, the actuator 9 moves, and the actuator A trigger signal is output from the controller 12 to the laser scanning sensor controller 11 instructing the timing for taking in one line data.
The actuator 9 performs a pitch movement at a preset movement pitch, and every time one pitch movement is completed, a trigger that instructs the laser scanning sensor controller 11 from the actuator controller 12 to capture one line data. A signal is output. When the capture of all line data is completed, the computer 13 receives all data from the laser scanning sensor controller 11.

  Next, image data to be converted into two-dimensional image data by performing gray scale processing (binarization processing) on the data in the height direction of the three-dimensional measurement data obtained by the three-dimensional measuring means using this light cutting method As the conversion means, processing is performed by a program in the computer 13.

Hereinafter, a method for extracting a character area cast on cast iron in the three-dimensional measurement process by the optical cutting method of the cast character will be described.
FIG. 2 is a diagram for explaining a scan range on cast iron by a light cutting method. FIG. 3 is a diagram for explaining how to estimate a character area.
In FIG. 2, the laser light emitted from the laser scanning sensor is indicated by a line beam (linear light) 31. The line length of the line beam (linear light) 31 is about 100 mm. By scanning the line beam (linear light) 31, the three-dimensional shape of the cast character in the scan range (shown by a dotted line) illustrated in FIG. 2 is measured.
In the present embodiment, the casting characters to be measured are three characters in each character area (rectangle) 33.
Further, in order to estimate an approximate rectangular area where each character area (rectangle) 33 exists, a machining reference boss 32 on the surface of the cylinder block is provided as a measurement target.
Next, a method of estimating the character area will be described with reference to FIG. First, a minimum rectangle including the processing reference boss 32 is obtained, and the center thereof is set as the center of the processing reference boss 32. Next, an approximate rectangular area including three character areas (rectangles) 33 is obtained from a straight line passing through the center of the machining reference boss 32 and parallel to the y-axis, and a relative position from the center of the machining reference boss 32. .

Next, a method for acquiring the three-dimensional shape data of three cast characters from the character region will be described in order with reference to FIGS.
FIG. 4 shows a displacement data string in which the center of the character “C” is captured by the laser scanning sensor for the character “C” positioned at the center of the character area. The horizontal axis z-axis of the graph of FIG. 4 is the height direction of the unevenness, and the vertical axis y-axis indicates the vertical direction of the characters. As shown in FIG. 4, the center of the letter “C” has a portion projecting up and down, and the center is depressed, which is shown in the plotted data.
First, the peak values of all lines in the y direction, which is the vertical direction of the character of this displacement data, are calculated. However, the line peak value is obtained by sorting the line data and excluding a predetermined number from the top as noise, and then setting the average value for the number specified in advance from the top as the peak value.
FIG. 5 shows a graph in which the obtained peak values are plotted in the x direction, which is the character string direction. In order to make the graph plot smooth, noise removal processing is performed.
Next, as shown in FIG. 6, an area S of B width (corresponding to character width) is obtained by sequentially moving in the x direction with respect to the obtained profile after noise removal in the x direction, and S is a peak value. The position to take is determined as the center of gravity position of the character area.
Then, as shown in FIG. 7, for the center of gravity position (G1, G2 in FIG. 7) of the obtained two character regions, the valley bottom (V1) is searched near the middle point, and V1 is the left character. And the middle character area boundary point. The boundary point (V2) between the middle character and the right character region is obtained by searching a valley bottom in the vicinity of 1/2 of the region width of the middle character from G2. Further, the left boundary V0 of the left character area is obtained as a position to the left of V1 by the area width of the left character (D1; the setting value is assumed to be +2.0 mm). Similarly, the right boundary V4 of the right character area is obtained as a position to the right of the right character area width (D3; the set value is +2.0 mm) from V2.
With the above method, the boundary in the x direction of the three character areas (rectangles) is detected.

Next, a method for detecting the boundary in the y direction of three character areas (rectangles) will be described. First, the peak values of all the lines in the x direction of the displacement data are calculated using the same method as the boundary detection method in the x direction.
Then, as shown in FIG. 8, in the graph in which the obtained peak values are plotted, a position (peak point) where the area of the concave portion is larger than a predetermined value is searched from the left end. A point (minimum point) having the minimum height is retrieved from the retrieved peak points for the number of data designated in advance, and set as the lower boundary of the three character regions (rectangles). The position corresponding to the height of the character area (rectangle) from the obtained lower boundary is set as the upper boundary of the three character areas (rectangles). In this way, the boundary in the y direction of the three character areas (rectangles) is detected.

FIG. 9 shows a state of three-dimensional measurement of cast characters by the light cutting method. (A) is a photograph of a state in which a letter “4A9” cast on the surface of an actual cast iron engine block cylinder block is being read by a laser scanning sensor. 9 shows a case where the line beam 41 of the laser scanning sensor is in the horizontal direction (x direction) and is scanned in the vertical direction (y direction) of the character area (rectangle). ing. In the above-described embodiment, the line beam is in the vertical direction (y direction), and the character beam (rectangular) is scanned in the horizontal direction (x direction). Is 90 ° different.
Further, (b) is an appearance photograph of the character “4A9” cast on the surface.
What measured this with the above-mentioned three-dimensional measuring means is shown to (A-1) and (B-1) of FIG. It can be understood that the irregularities of the cast characters are reproduced by computer graphics.
(A-2) and (A-2) in FIG. 10 are obtained by converting the data in the height direction of the three-dimensional measurement data of the unevenness of the cast character reproduced by this computer graphic into a two-dimensional image data by performing gray scale processing. B-2). The casting height of the characters cast on the cylinder block surface is expressed in brightness when converted to gray scale, and even when converted to two-dimensional image data, the outline of the cast characters can be understood and character recognition is performed. I can understand that I can do it enough.
In a test in which character numbers were determined from these image data based on information learned in advance by a neural network, the reading rate was close to 100% regardless of the character type.

  The cast character recognition device of the present invention is a cylinder block of an internal combustion engine for vehicles, ships, generators, construction machinery, etc., cast parts of vehicles such as engine camshafts, piping, knuckles, and manholes. It can recognize letters, numbers, and symbols such as manufacturing lots cast into raw materials and use them for traceability management of these parts.

The block diagram which showed the apparatus structure of the cast-out character recognition apparatus of Example 1. The figure explaining the scanning range on cast iron by the light cutting method Diagram explaining how to estimate the character area Example of plotting line data for each character area Example of plotting peak values of all lines in the vertical direction (y direction) of characters in the character area in the character string direction (x direction) The figure explaining how to calculate the gravity center position of each character area The figure explaining the method of calculating the boundary line of the character string direction (x direction) of each character area is shown. The figure explaining the method of calculating the boundary line of the vertical direction (y direction) of the character of each character area is shown. A state of three-dimensional measurement of cast characters by the light cutting method is shown. An example of grayscale converted 3D data and converted to 2D image data is shown.

Explanation of symbols

1 Founded character recognition device
2 Cylinder block 3 Belt conveyor 4 Object detection sensor A
5 Object detection sensor B
6 Stopper 7a, 7b Pusher 8 Laser scanning sensor 9 Actuator 10 Control device (programmable controller; PLC)
11 Controller for laser scanning sensor 12 Controller for actuator 13 Computer (PC)
21 USB (Universal Serial Bus) cable 22 RS-232C cable 23 Communication cable 24, 25 Dedicated cable
31 line beam (linear light)
32 Processing standard boss 33 Each character area (rectangular)
41 Line beam of laser scanning sensor

Claims (6)

Three-dimensional measuring means for measuring the uneven shape of the letters and numbers area by casting, and two-dimensional image data by gray-scale processing the height direction data of the three-dimensional measurement data obtained by the three-dimensional measuring means Image data conversion means for performing binarization processing after conversion into character data, and character / numeric discrimination means for discriminating letters and numbers based on information learned in advance from the image data obtained by the image data conversion means,
The three-dimensional measuring means projects linear light onto the alphanumeric character area, and captures the three-dimensional shape data of the alphanumeric character region by capturing an image formed by the linear light drawn on the surface of the alphanumeric character region. Use the light cutting method to get
In order to detect the partition position of each alphanumeric character from the alphanumeric character region,
a) Calculate the peak value of the displacement data obtained by the light cutting method,
b) Obtain a profile in which the calculated peak value of the displacement data is plotted in the direction of the character string,
c) calculating the barycentric position of each letter and numeral at a position where the area obtained by sequentially moving the area corresponding to the letter and numeral width from the profile in the letter and numeral string direction becomes a peak value;
d) calculating a valley bottom position of the displacement data existing at a substantially middle point with respect to the gravity center positions of two adjacent letters and numbers;
A cast character recognition device characterized by that.   The cast character recognition device according to claim 1, wherein the three-dimensional measuring means has a line width of 100 μm to 500 μm of linear light projected onto a character / numeric area to be measured by a light cutting method.   2. The cast character recognition apparatus according to claim 1, wherein the three-dimensional measuring means has a wavelength of linear light projected on a character / numerical region to be measured by a light cutting method in a range of 650 nm to 900 nm.   The cast character recognition apparatus according to claim 1, wherein the character / numeral discrimination means uses a learning method based on a neural network.   The cast character recognition apparatus according to claim 4, wherein the neural network is a multi-layer perceptron.   2. The cast character recognition apparatus according to claim 1, wherein the three-dimensional measuring means measures the uneven shape of the cast character numbers of the cylinder block of the internal combustion engine.