JP3338115B2 - Pattern recognition apparatus and pattern recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[Contents] Industrial application field Conventional technology (FIGS. 10 and 11) Problems to be solved by the invention (FIG. 12) Means for solving the problems (FIGS. 1 and 2) 3-9) Effects of the invention

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern recognizing apparatus and a pattern recognizing method, and more particularly, to an apparatus for separating and recognizing a shape pattern in which circles, rectangles and the like overlap, and an improvement of the recognizing method. is there.
2. Description of the Related Art In recent years, in the field of image processing, an information processing apparatus that performs high-performance and high-speed data processing has been used. For example, a cafeteria clearing system that automatically discriminates a food and drink by identifying and measuring tableware is adopted.

In this system, a favorite type or a plurality of individual dishes are placed on a tray in an employee cafeteria or a large-scale cafeteria, and an image of a tableware image including an overlapping portion of the tableware and a protrusion is taken. After the image processing, the price of the predetermined tableware is compared with the measurement result of the selected tableware to automatically calculate the food and drink price.

According to this, the pattern recognizing section tracks the outer contour of the shape pattern to be recognized, which is a composite of circular and square tableware, and recognizes various tableware patterns. However, when a plurality of circular patterns or the like form an annular shape and have internal contours, it becomes difficult to recognize the pattern of the recognition target. Therefore, even when there is an outer contour line and an inner contour line due to the contact or overlap of various shape patterns, the inner contour line is extracted from the outer contour line by a geometric calculation method, and the shape to be recognized is extracted. There is a need for an apparatus and method capable of quickly recognizing various shape patterns from patterns.

[0005]

2. Description of the Related Art FIGS. 10 to 12 are explanatory views according to a conventional example. FIG. 10 is a configuration diagram of a pattern recognition device according to a conventional example, and FIG. 11 is an explanatory diagram of a pattern recognition method according to a conventional example. FIG. 12 shows a composite shape pattern diagram for explaining the problem. For example, in FIG. 10, an apparatus to which a pattern recognition method applied by the applicant of the present invention (Japanese Patent Application Laid-Open No. 04-290174) is applied includes an imaging system 1, a signal processing unit 2, a measurement / recognition unit 3, and a control unit. 4
Consists of

[0006] As shown in the processing flowchart of the control section 4, the function of the apparatus is as follows. Here, the acquired image of the recognition target 14 is subjected to binarization signal processing by the signal processing unit 2, and the image data DIN is output to the measurement / recognition unit 3. Next, in step P2, an external contour line is identified, and then, in step P3, it is determined whether the external contour line is a valid measurement line. At this time, if the outer contour is a valid line (YES), the flow shifts to Step P4 or P5. If the external contour is not suitable for measurement (NO), the image processing is terminated and an error display or the like is performed. Therefore, when it is determined in step P3 that the external contour is valid (YES), for example, a circular measurement process is performed in step P4, and a polygon measurement process such as a quadrangle or a triangle is performed in step P5. At P6, registration processing of the measurement result is performed.

[0007] Thus, when the pattern recognition device is used in a cafeteria clearing system, it is possible to automatically pay for food and drink by identifying and measuring tableware. Next, identification of a circular or polygonal tableware according to a conventional example and measurement processing thereof will be described. For example, in the case where the recognized shape pattern H1 is formed only by a circular dish, in FIG. 11A, first, an external contour is extracted from the recognized shape pattern H1 by a contour tracking method. Here, in order to find a bending point, sampling points P1, P2... Pm... Pn... Po.
Then, the direction change values of the inclination vectors V1, V2 between three adjacent sampling points, for example, P1, P2, P3, are sequentially obtained.

Next, it is determined whether or not the direction change values of the vectors V1 and V2 are within a predetermined set value range.
The bending points Pm, Pn, Po, and Pp are determined. After that, it is considered that the contour line between the bending points such as the bending points Pm to Pn is generated from the same circular tableware, and any points Pn− existing on the contour line between the bending points Pm and Pn and the bending point are considered. Consider a circle passing through m. Then, the center coordinates of the circle are calculated, and those whose radius and center coordinates are close to each other are integrated into the same circle based on the calculation result. Thereby, two contour lines P
m to Pn and Po to Pp are integrated as the same circle, and the type of the circular shape of the tableware and the number thereof can be identified.

FIG. 11B shows a circular shape A and a square shape B.
And a case where a recognized shape pattern is formed by the irregular shape D. In FIG. 11B, such a state that the circular shape A and the square B overlap each other,
When the pattern recognition is performed in such a manner that the irregular shape D and the irregular shape D are overlapped with each other, as in the case of the pattern recognition of the circular tableware, the external contour is sampled at a constant interval, and the vector change formed by two adjacent sample points causes the figure to change. A method of recognizing the inflection point indicating the contact point and the inflection point indicating the vertex of the polygon and separating the contour line is adopted.

For example, contour points P1 to P11 are detected counterclockwise by a contour tracing method, and the contours are followed to extract the contour point coordinates (x, y) and vectorize them. Here, the vector is formed by connecting adjacent contour points counterclockwise, and the angle formed by the vector is the ± vector between the target vector Vm, Vn and the immediately preceding vector Vm-2, Vn-2. Define and calculate. Here, when the angle formed by the vectors becomes negative, the convex inflection point P1, and when the angle suddenly becomes positive, the concave inflection point P4. Here, a to h, j are partial contour lines, which are formed by sides between contour points P1 → P9, P10 → P11. Also, i and k are arc contours,
It is formed by an arc between the contour points P9 → P10 and P11 → P1.

Next, when measuring a circular shape, first, a radius and a center coordinate are obtained by a least-square method based on an arc contour line having a sufficient size to take the number of sampling points with respect to the arc contour line. Thereafter, it is determined whether the circular measurement is possible. For example, the starting point P9 of the arc contour line → the center point → P10
The angle θ between the two vectors is calculated, and the ratio of 360 degrees is calculated. If the angle is larger than the identification relative allowable angle, the angle is measured. It is determined whether or not the arc contour lines i and k originate from the same circle. If it is determined that the arc contour lines i and k originate from the same circle, they are integrated and the circular shape is measured.

In the case of measuring a quadrangle, a quadrangle is extracted from a circular shape or an irregular shape on the basis of a side contour having a size large enough to take the number of sampling points with respect to the side contour. It is determined whether measurement is possible. For example, when the interior angle of the vector V1 on the side a and the vector V2 on the next side is a right angle, when the vector V1 and the vector V3 on the side g are parallel and opposite, the vectors V1 and V3 have an overlap. Is the case. After verifying whether a rectangle satisfies this or not, if it can be measured, numbering is performed on each side to calculate the distance between two facing sides (the size of the rectangle).

Thus, even when a recognized shape pattern is formed by the circular shape A, the square B, and the irregular shape D, the circular shape A and the square B can be accurately extracted. In addition, when the pattern recognition method is adopted in a cafeteria clearing system, it is possible to instantaneously judge and identify tableware such as a circular shape A and a square B and irregular shapes D such as protruding, and measure the same. Can be performed.

[0014]

By the way, according to the conventional pattern recognition method, as shown in the processing flowchart of the control section 4 in FIG. 10, in step P2, a circle, a triangle, a quadrangle and the like are combined. The outer contours of the recognized shape patterns H1 to H3 are tracked, and various shape (tableware) patterns are recognized.

Therefore, when the circular shape A, the square B, and the irregular shape D are superimposed as shown in FIG. 11B, the shape pattern can be easily recognized. However, a plurality of circular patterns as shown in FIG. 12A form an annular shape, and it becomes difficult to recognize a pattern of the recognition target 14 having an internal contour. For example, the recognition target 14 is composed of five circular patterns A to E, and the patterns A and B, B and C, C and D, D and E, and E and A are in contact or overlap. . In such a case, even if the recognized shape pattern of the recognized object 14 is subjected to the data processing by the contour tracking method, the information amount of the identifiable partial contour is reduced, and the roundness as shown in FIG. Only the extracted data related to the pentagonal outer contour line is obtained.

Accordingly, when the outline is separated into the outside and the inside by the graphic patterns touching or overlapping each other, only the tracking and extraction of the outside outline is performed.
Extracted data relating to the internal contour as shown in (C) cannot be obtained, and it becomes difficult to individually identify the five circular patterns A to E. As a result, when the pattern recognition method is employed in a cafeteria clearing system, there is a problem that an error occurrence frequency increases and the reliability of the system decreases.

The present invention has been made in view of the problems of the prior art described above. Even when an outer contour line and an inner contour line exist due to the contact or overlap of various shape patterns, It is an object of the present invention to provide a pattern recognition device and a pattern recognition method capable of extracting an internal contour from an external contour by a geometric calculation method and quickly recognizing various shape patterns from a recognized shape pattern.

[0018]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention relates to a pattern recognition apparatus, which comprises an outer contour defining an outer shape and an inner contour defining an inner cavity. A pattern recognition device that recognizes a combined pattern in which a single pattern of a shape, or a plurality of patterns including the same pattern or a pattern of a different shape overlap or touch each other, acquires an image of the single pattern or the combined pattern, and outputs a signal. Image acquisition means for performing processing, based on a signal from the image acquisition means, a shape recognition means for recognizing the outer contour and the inner contour,
Control means for controlling input and output of the image acquisition means and the shape recognition means, wherein the shape recognition means sequentially samples a predetermined interval of the outer contour lines and recognizes the outer contour lines by a first vector. And determining a bending angle by determining a bending portion where the two lines of the outer contour line intersect based on an angle change of the first vector, and calculating the bending angle at a predetermined angle in the range of the bending angle. The apparatus is characterized in that a predetermined interval between the inner contour lines is sequentially sampled by invading the inner contour side from the contour line side, and the inner contour line is recognized by the second vector.

Further, in order to solve the above-mentioned problem, the present invention relates to a pattern recognition method, and comprises a single pattern of a predetermined shape having an outer contour defining an outer shape and an inner contour defining an inner cavity. Or a pattern recognition method of a composite pattern in which a plurality of patterns including the same pattern or a pattern having a different shape overlap or touch each other, sequentially sampling a predetermined interval of the outer contour line and using a first vector to form the outer contour line. And, based on the change in the angle of the first vector, determine a bending portion where the two lines of the outer contour line intersect to determine a bending angle of the bending portion, and determine the bending angle at a predetermined angle in the range of the bending angle. 2 vectors from the outer contour side to the inner contour side, and sequentially samples a predetermined interval of the inner contour line to obtain the second vector. And recognizes the inner contour by.

In the above-described pattern recognition method, when specifying the bent portion, a bending direction and an angle between two adjacent first vectors are obtained, and the bending direction is a predetermined direction, and When the angle exceeds a predetermined value,
The vicinity of the intersection of the two first vectors is a bent portion.

In the above-described pattern recognition method, the second vector is made to enter the inner contour portion at an angle that divides the bending angle by about half.

[0022]

According to the pattern recognition apparatus of the present invention, as shown in FIG. 1, an image acquisition unit 11, a shape recognition unit 12, and a control unit 13 are provided. The search control for the inner contour is performed based on the search control for the outer contour of the patterns A, B, C.

For example, an image of the object to be recognized 14 is acquired by the image acquiring means 11, and the signal-processed image acquisition data DIN is output to the shape recognizing means 12. The shape recognition means 12 extracts and measures various shape patterns from the recognition target 14 via the control means 13. At this time, the control means 13 performs the search control of the internal contour based on the search control of the external contour of the various shape patterns A, B, C.

For this reason, even when various shape patterns are in contact with each other or overlap each other and the outer contour and the inner contour exist, the inner contour is extracted from the outer contour by a geometric calculation method. Can be. This makes it possible to quickly recognize various shape patterns from the recognized shape pattern based on the search and extraction data D1 of the outer contour line of the recognized shape pattern and the search and extraction data D2 of the inner contour line.

Further, according to the pattern recognition method of the present invention, as shown in the processing flowchart of FIG. 2A, in step P2, the internal contour of the shape to be recognized is determined based on the external contour search and extraction processing. Search and extraction processing is being performed. For example, in step P2A, an internal search process is performed from a contour point on the outer contour line where the shape pattern to be recognized overlaps, to a region surrounded by the outer contour line. At this time, as shown in FIG. 2B, the vectors V1 and V2 in two directions to the contour points p1 and p2 on the external contour are based on the bending point pi on the external contour of the shape to be recognized.
2 is set, and the inverted vector −V [θ] of the vector V [θ] that divides the bending angle θ formed by the two vectors V1 and V2 into two.
The search processing is continued in the direction of. Next, in step P2B, a partial outline is extracted in the area by the outline tracking method, and then, in step P2C, the partial outline is integrated.

For this reason, the various shape patterns come into contact with each other or overlap each other to form a recognized shape pattern in which an outer contour line and an inner contour line exist. Search extraction data D2
Can be obtained. From this, step P3
Thus, for example, by adding the search / extraction data D2 of the internal contour to the search / extraction data D1 of the external contour, the separation / identification processing of the various shape patterns A, B, C... Can be.

This makes it possible to provide a highly reliable cafeteria clearing system or the like to which the pattern recognition device or method is applied.

[0028]

Next, an embodiment of the present invention will be described with reference to the drawings. 3 to 9 are diagrams illustrating a pattern recognition device and a pattern recognition method according to an embodiment of the present invention, and FIG. 3 is a configuration diagram of the pattern recognition device according to the embodiment of the present invention. For example, in FIG. 3, a pattern recognition device applicable to a cafeteria clearing system for automatically clearing food and drinks by identifying and measuring tableware is shown in FIG.
/ D conversion circuit 21B, logic filter circuit 21C, image memory
21D and a shape recognition control system 100.

That is, the CCD camera 21A, the A / D conversion circuit 21B, the logic filter circuit 21C and the image memory 21D constitute one example of the image acquisition means 11, and the CCD camera 21A
The camera 21A acquires an image of the recognition target 14 and outputs the image acquisition signal Ain to the A / D conversion circuit 21B. A / D
The conversion circuit 21B performs A / D conversion of the image acquisition signal Ain and outputs the digital image data to the logical filter circuit 21C. The logic filter circuit 21C binarizes the digital image data and outputs the image acquisition data DIN to the image memory 22A. The image memory 21D stores the image acquisition data DIN.

The shape recognition control system 100 comprises the shape recognition means 12 and the control means 13 and has a shape extraction /
Measurement unit 22, I / O interface unit (hereinafter simply referred to as I / O interface unit)
23A, ROM (read only memory) 23B,
RAM for contour points (memory that can be written / read at any time) 23
C, a RAM 23D for a circular outline, a RAM 23E for a side outline, and a CPU (central processing unit) 23F.

The shape extraction / measurement unit 22 is a
This is an example of extracting and measuring various shape patterns from the recognition target 14. The shape extraction / measurement unit 22
The function (1) is described in detail in FIGS. 4 to 6 because the processing flowchart of pattern recognition shown in FIGS. I / O section 23A, ROM 23B, RA for contour points
M23C, RAM23D for circular outline, RAM23E for side outline
And the CPU 23F constitute the control means 13,
The I / O section 23A inputs and outputs image acquisition data DIN, recognition result data DOUT, and other control data.

The ROM 23B is a memory for storing a processing program of the pattern recognition device. For example, a condition for judging a circular shape, a condition for judging a square as shown in FIG. The contour point RAM 23C is a memory for storing contour point data D3 of the shape pattern to be recognized. The circular outline RAM 23D is a memory for storing the search and extraction data D1 of the external outline and the search and extraction data D2 of the internal outline. The side outline RAM 23E is a memory for storing the side outline data D4 of the shape pattern to be recognized.

The CPU 23F controls the input and output of the CCD camera 21A and the shape extraction / measurement unit 22. For example, CPU23
F performs search control of the internal contour based on the search control of the external contour of the various shape patterns of the recognition target 14. Here, the search control of the external contour is defined by the shape extraction / measurement unit 22.
In contrast, control for extracting a pattern along the outer contour of the shape pattern to be recognized is assumed. Similarly, the search control of the internal contour line refers to control for extracting a pattern that is considered to be present in the internal region with reference to a bending point on the external contour line of the shape pattern to be recognized.

As described above, according to the pattern recognition apparatus according to the embodiment of the present invention, as shown in FIG. 3, the CCD camera 21A, the A / D conversion circuit 21B, and the logic filter circuit 21C.
And an image memory 21D and a shape recognition control system 100. The CPU 23F controls the search for the inner contour based on the search control for the outer contour of the various shape patterns of the recognition target 14.

For example, an image of the object to be recognized 14 is acquired by the CCD camera 21A, and the image acquisition signal Ain is A / A.
The signal is processed by the D conversion circuit 21B and the logic filter circuit 21C, and the binarized image acquisition data DIN is output to the shape extraction / measurement unit 22. The shape extraction / measurement unit 22 extracts and measures various shape patterns from the recognition target 14 via the CPU 23F. At this time, the CPU 23F controls the search for the inner contour based on the search control for the outer contour of the various shape patterns of the recognition target 14.

For this reason, even when various shape patterns are in contact with or overlap with each other, and the outer contour and the inner contour exist, the inner contour is extracted from the outer contour by a geometric calculation method. Can be. This makes it possible to quickly recognize various shape patterns from the recognized shape pattern based on the search and extraction data D1 of the outer contour line of the recognized shape pattern and the search and extraction data D2 of the inner contour line.

Next, a pattern recognition method according to an embodiment of the present invention will be described while supplementing the operation of the apparatus. 4 to 6 are flowcharts (parts 1 to 3) of a pattern recognition process according to the embodiment of the present invention, and FIG. 7 is an explanatory diagram of a shape pattern to be recognized. 8 and 9 show supplementary explanatory diagrams (parts 1 and 2) of the pattern recognition method, respectively.

For example, in the case of performing pattern recognition processing of a recognition target 14 in which four shape patterns A, B, C, and D are combined as shown in FIG. 7, first, in FIG. Image acquisition processing. At this time, next, in steps P2 to P4, a process of searching for and extracting the outer contour of the recognized shape pattern H14 from the image acquisition data DIN of the recognized object 14 is performed. For example, in step P2, an external contour point of a binarized image (hereinafter, referred to as a recognized shape pattern) H14 of the recognized object 14 is detected. Here, in FIG. 7, an outer contour line, for example, an outer contour line of the shape patterns A, B, C, and D including the bending points p1 to p6 (see FIG. 7). Arcs a, sides b, c, d and arcs e, f) are detected.

Next, in step P3, the outer contour coordinate values of the shape pattern to be recognized H14 are extracted. At this time, following the outer contour line, the outer contour coordinate values are extracted, and the contour point data D3 is vectorized. The contour point data D3 is stored in the contour point RAM 23C. Further, at step P4, the inflection points p1 to p1 on the outer contour line of the shape pattern H14 to be recognized.
Extract p6. Here, in FIG. 8A, the bending angle θ is defined. The bending angle θ is determined by the vectors V1 to V3.
m-1, Vm... Vn-1, Vn... are connected in a counterclockwise direction, and the vector Vm of interest and the vector V
The angle formed by m-1. Further, the bending angle θ is defined by ± directions. For example, as shown in FIG.
Is sharply changed to positive, the bending point p1 is determined to be a concave bending point, and if the bending angle θ is suddenly changed to negative, the bending point p3 is determined to be a convex bending point.

Thereafter, in step P5, it is determined whether or not a contour exists in the area inside the recognized shape pattern H14. At this time, if there is a contour line in the inner area (YES), the flow shifts to Step P6. Here, the number of bending points p1 to p6 obtained by the contour tracking method becomes three or more, and the probability that an internal contour exists is increased. Further, when the number of the bending points p1 to p6 is two or less, it corresponds to the case where there is no contour line (NO), and the process shifts to Step P9.

Therefore, when the number of the bending points is three or more (YES), the contour points on the outer contour line where the shape pattern to be recognized H14 overlaps extend from the contour points on the outer contour line to the area surrounded by the outer contour line. Execute search processing. That is, in steps P6 to P8, search and extraction processing of the internal contour of the recognized shape pattern H14 is performed. First, in step P6, the internal contour points of the recognized shape pattern H14 are detected. Here, an example different from the shape pattern H14 is shown in FIG.
As shown in (B), an internal contour point search process is executed with reference to a bending point Pj on the external contour line of the recognized shape pattern H15 in which the circular pattern is compounded. Since the rotation direction of the bending angle θ is positive, the bending point P on the outer contour line
With reference to j, vectors Vi and Vk are set in two directions to contour points Pi and Pk on the external contour, and the vector V
Vector V [θ] that divides the bending angle θ formed by i and Vk into two
The unit vector ε is extended in the direction of the inverted vector −V [θ], and the search processing for the internal contour point is continued. Further, the unit vector ε extends to an area without label data (an area without the circular pattern data), and when the label data is detected, the unit vector ε is advanced from the coordinate position in the X direction.

Next, in step P7, the internal contour coordinate values of the shape pattern H14 are extracted. At this time, in FIG. 7, the internal contours, for example, the shape patterns A, B, C, and D including the bending points p7 to p10 are counterclockwise contoured with respect to the internal contour points of the shape pattern H14. An inner contour (arc g, side h, arc i, j) is detected. Further, in step P8, the inflection points p7 to p10 on the internal contour of the shape pattern H14 are extracted. At this time, the inner contour is chased, the inner contour coordinate values are extracted, and the contour point data D3 is vectorized. The contour point data D3 is stored in the contour point RAM 23C. Thereby, the search and extraction data D2 of the inner contour obtained by the integration processing of the partial contours is added to the search and extraction data D1 of the outer contour.

Thereafter, in step P9, a classification process of the partial outline is performed. For example, a partial contour composed of arcs and straight lines as shown in FIG. The q value is determined by the identification relative allowable value of the shape extraction / measurement unit 22.
Here, considering the error with respect to small image noise, the angles α1, α between the vector of interest and the vector two ahead are considered.
2, α3... Αq and β1, β2, β3... Βp are calculated, and their average value (α1 + α2 + α3 +... + Αq) /
q, (β1 + β2 + β3 +... + βp) / q is obtained.
When this value is smaller than the allowable set value (when it is close to 0)
Is a straight line generated from the polygonal shape of the recognized shape pattern H14, that is, a side contour, and if it is larger than an allowable set value, it is classified as an arc contour generated from a circular pattern.

As a result, arcs a, e, f, g, i, j and sides b,
c, d, and h are classified, and arc contour lines a, e, f, g,
The coordinate values of i and j are stored as search and extracted data D1 and D2 in the RAM 23D for circular contour lines. Also, the side contour lines b,
The coordinate values of c, d, and h are stored in the edge contour RAM 23E as edge contour data D4.

Next, in step P10, prior to the measurement processing of the shape pattern to be recognized H14, it is determined whether or not the contour is a contour from which a sampling point can be taken. At this time, if the contour line can take a sampling point (YES), step P11 or P18
Move to In the case of a contour line that cannot be obtained (N
In O), the control is terminated as in the conventional example. Therefore, when it is determined in step P10 that the contour line can take a sampling point (YES), the circular measurement processing is executed in steps P11 to P17, and the rectangle measurement processing is executed in steps P18 to P28. As a result, various shape patterns A, B, C, and D of the recognition target 14 are separated and identified based on the outer contour search and extraction data D1 and the inner contour search and extraction data D2 of the recognized shape pattern H14. be able to.

For example, if the circular measurement processing is to be executed first, the center coordinates and the radius of the circular arc are obtained in step P11. Here, the center coordinate O is calculated by the least squares method with respect to the arc contour line having a sufficient length for obtaining the number of sampling points. Here, the condition for determining the circular shape is read from the ROM 23B. Next, in step P12, it is determined whether or not the contour is an arc to be integrated. At this time, the calculated radii and center coordinates O are compared with respect to each of the arc contour lines a, e, f, g, i, and j. Considered to have occurred. As a result, it is determined that the outline is an arc to be integrated (YES), and the routine goes to Step P13 to integrate the partial arc outline. Thereafter, the center coordinates and the radius are obtained from the integrated arc in step P14. Specifically, the bending angle θ1 formed by the two vectors of the starting point p1 → the center point → the ending point p2 of the plurality of arc contours by using the center coordinates O again by the least square method using the center coordinates O. And calculate the ratio of 360 [°].

Thereafter, returning to step P12, if the arc is not integrated with the contour (NO), step P15
Then, it is determined whether or not the ratio of the arc is within a measurable allowable value. At this time, if the ratio of the arc is at the allowable value (YES), the process shifts to Step P16 to determine whether or not the contour is a measurable circle. Here, if the arc contour lines a and e shown in FIG. 7 are generated from the same circle and are determined to be measurable circles (YES), the circular pattern of the recognized shape pattern H14 is determined in step P17. Execute the measurement processing of.

If the bending angle θ1 is smaller than the identification relative allowable angle in step P15, there is a risk of erroneous recognition, so it is determined that the ratio of the arcs does not reach the measurable allowable value (NO). , And the control ends. Also, if the circle is not measurable in step P16 (NO), it is not measured. If it is determined in step P10 that the contour is a contour line at which sampling points can be taken (YES), and if the process of measuring the rectangular pattern of the shape pattern to be recognized H14 is to be executed, the process proceeds to step P18. More specifically, the side contour data D4 stored in the side contour RAM 23E is read out, and the shape extraction / measurement unit 22 extracts a side contour line having a sufficient length for obtaining the number of sampling points in the side contour line. Extract and measure polygons from irregular shapes.

That is, in step P18, it is determined whether or not the outline of the recognized shape pattern H14 is a side to be integrated. At this time, if the outline is a side to be merged (YES), the flow shifts to step P19 to integrate the partial straight outline. If it is not the edge to be integrated (NO),
The program shifts to Step P20. Specifically, in FIG.
If the polygon C meets the following conditions, the polygon C is measured as a “square”. If the polygon C is not matched, there is a risk of erroneous recognition, and the polygon C is not measured. Here, the condition for judging the square is read from the ROM 23B. The contents are, as shown in the condition of FIG.
The inner angle between 1 and the vector V2 must be a right angle. When this is applied to the polygon C in FIG. 7, the interior angle between the vector V1 by the side b and the vector V2 by the next side c is a right angle.

In step P20, it is determined whether or not the vectors of the two opposing sides are parallel and opposite. At this time, if the vectors on the two sides are parallel and opposite (YES), the program shifts to Step P21. Here, in order for the polygon C to be a quadrangle, as shown in the condition of FIG.
The vector V3 opposite to the vector V1 must be parallel and opposite. Applying this to polygon C in FIG.
Vector V1 by side b and vector V3 by opposing side d
Are parallel and reversed.

Thus, the process shifts to step P21 to determine whether or not the vectors of the two sides facing each other have an overlapping portion. At this time, if the vector has a portion where the vector overlaps (YES), the program shifts to Step P22. Here, in order for the polygon C to be a quadrangle, the vector V1 and the vector V3 must have an overlapping portion as shown in the condition of FIG. 9B. When this is applied to the polygon C in FIG. 7, there is a difference between the vector V1 by the side b and the vector V3 by the opposite side d.

Accordingly, the process shifts to step P22 to extract a side opposite to the basic side. Here, when there are a plurality of opposing sides, the closest side is extracted. In step P23, the positional relationship around the opposing sides is verified, and then, in step P24, the width of the two sides of the recognized shape pattern H14 is measured. Thereby, the polygon C can be determined as a quadrangle.

If it is determined in step P20 that the vectors of the two sides are parallel and not in opposite directions (NO) or if it is determined in step P21 that there is no overlapping part of the vectors of the two sides (NO), the control is performed. To end. Then, a registration process of the measurement result is performed in step P25. Accordingly, the recognition target 1 in which the circular patterns A and B, the circular pattern B and the square pattern C, the pattern C and the circular pattern D, and the circular patterns D and A are in contact or overlap with each other.
4 can be individually recognized as three circular patterns A to C and one square D.

As described above, according to the pattern recognition method according to the embodiment of the present invention, as shown in the processing flowcharts (Nos. 1 to 3) of FIGS. Steps P6 to P based on the extraction process
In step S8, a search and extraction process of the internal contour of the recognized shape pattern H14 is performed. For example, from the contour point p1 on the external contour line where the recognized shape pattern H14 overlaps in step P6,
The internal search processing is performed toward a region surrounded by the external contour. At this time, as shown in FIG.
With reference to the inflection point Pj on the outer contour line of the shape pattern H14 to be recognized, two points reaching the contour points Pi and Pk on the outer contour line are obtained.
Vectors Vi and Vk are set in the directions, and both vectors V
Vector V [θ] that divides the bending angle θ formed by i and Vk into two
The unit vector ε is extended in the direction of the inverse vector −V [θ], and then, in steps P7 and P8, a partial contour is extracted in the area by the contour tracing method. The partial contour lines are integrated.

For this reason, the various shape patterns H14 come into contact with each other or overlap each other to form a recognized shape pattern H14 in which an external outline and an internal outline exist, and the internal outline is difficult with the conventional processing method. Line search and extraction data D2 can be obtained. From this, in step P8, the search and extraction data D2 of the inner contour is added to the search and extraction data D1 of the outer contour to separate and identify the various shape patterns A, B, C, and D of the recognition target 14. Processing can be performed accurately. That is, in the conventional example, since the conventional method relies only on the process of searching for the outer contour, the side contour h of the polygon C cannot be extracted, and the determination condition of the quadrangle cannot be cleared, so that the measurement cannot be performed immediately. I was On the other hand, according to the present invention, since the side contour h can be accurately extracted, it is possible to easily identify the polygon C as a quadrangle and to measure the same.

Thus, even when the external contour and the internal contour exist due to the contact or overlap of the various shape patterns, the internal contour is extracted from the external contour by the geometric calculation method. By this, various shape patterns A, B,
C and D can be quickly recognized. This makes it possible to provide a highly reliable cafeteria clearing system or the like to which the pattern recognition device or method is applied.

[0057]

As described above, according to the pattern recognition apparatus of the present invention, there are provided an image acquisition means, a shape recognition means and a control means, and the control means comprises an external contour line of various shape patterns to be recognized. The search control of the internal contour is performed based on the search control of. For this reason, even when various shape patterns are in contact with or overlap with each other, and the outer contour and the inner contour exist, the inner contour can be extracted from the outer contour by the geometric calculation method. Accordingly, it is possible to quickly recognize various shape patterns from the recognized shape pattern based on the search and extraction data of the outer contour line of the recognized shape pattern and the search and extraction data of the inner contour line.

Further, according to the pattern recognition method of the present invention, the process of searching and extracting the internal contour of the shape pattern to be recognized is performed based on the process of searching and extracting the external contour. Therefore, by adding the search and extraction data obtained by the search and extraction processing of the internal contour to the search and extraction data obtained by the search and extraction processing of the external contour, various shape patterns can be accurately separated from the recognition target. It becomes possible to identify.

This greatly contributes to providing a high-reliability cafeteria clearing system and the like to which the pattern recognition apparatus and method are applied.

[Brief description of the drawings]

FIG. 1 is a principle diagram of a pattern recognition device according to the present invention.

FIG. 2 is a principle diagram of a pattern recognition method according to the present invention.

FIG. 3 is a configuration diagram of a pattern recognition device according to an embodiment of the present invention.

FIG. 4 is a flowchart (part 1) of a pattern recognition process according to the embodiment of the present invention.

FIG. 5 is a flowchart (part 2) of a pattern recognition process according to the embodiment of the present invention.

FIG. 6 is a flowchart (part 3) of a pattern recognition process according to the embodiment of the present invention.

FIG. 7 is an explanatory diagram of a recognized shape pattern according to the embodiment of the present invention.

FIG. 8 is a supplementary explanatory diagram (part 1) of the pattern recognition method according to the embodiment of the present invention.

FIG. 9 is a supplementary explanatory diagram (part 2) of the pattern recognition method according to the embodiment of the present invention.

FIG. 10 is a configuration diagram of a pattern recognition device according to a conventional example.

FIG. 11 is an explanatory diagram of a pattern recognition method according to a conventional example.

FIG. 12 is a composite shape pattern diagram for explaining a problem according to the conventional example.

[Explanation of symbols]

 11: image acquisition means, 12: shape recognition means, 13: control means, L1: external contour line, L2: internal contour line, A to D: various shape patterns, DIN: image acquisition data, D1, D2: search and extraction data , Θ: bending angle, V1, V2: vector, V [θ], −V [θ]: vector, inverted vector, p1, p2: bending point.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06T 7/60 G01B 11/24 G06T 1/00 280

Claims (4)

(57) [Claims] 1. A single pattern of a predetermined shape or a plurality of patterns including a pattern of the same pattern or a different shape, comprising an outer contour defining an outer shape and an inner contour defining an inner cavity. Or a pattern recognition device for recognizing a contacting synthetic pattern, comprising: an image obtaining unit that obtains an image of the single pattern or the synthetic pattern to perform signal processing; and the outside based on a signal from the image obtaining unit. A shape recognizing unit that recognizes a contour line and the internal contour line, and a control unit that controls input and output of the image acquiring unit and the shape recognizing unit, wherein the shape recognizing unit has a predetermined interval between the outer contour lines. Are sequentially sampled to recognize the outer contour line by a first vector, and the outer contour line is determined based on a change in the angle of the first vector. Determining a bend angle where two lines intersect to determine a bend angle, and causing a second vector to penetrate from the outer contour side to the inner contour side at a predetermined angle within the range of the bend angle, Wherein the predetermined interval is sequentially sampled and the inner contour is recognized by the second vector. 2. A single pattern of a predetermined shape or a plurality of patterns including a pattern of the same pattern or a different shape having an outer contour defining an outer shape and an inner contour defining an inner cavity. Or a method of recognizing a contact pattern of a composite pattern, wherein a predetermined interval of the outer contour line is sampled sequentially and the first
The outer contour line is recognized by a vector, and based on the angle change of the first vector, the outer contour line 2 is recognized.
The bending angle at which the two lines intersect is determined to determine the bending angle, and a second vector is made to penetrate from the outer contour side to the inner contour side at a predetermined angle in the range of the bending angle, and A pattern recognition method characterized by sequentially sampling predetermined intervals and recognizing the inner contour line by the second vector. 3. When specifying the bent portion, a bending direction and an angle between two adjacent first vectors are determined, and the bending direction is a predetermined direction, and the angle is a predetermined value. 3. The pattern recognition method according to claim 2, wherein, when the number of the first vectors exceeds the number, the vicinity of the intersection of the two first vectors is set as a bent portion. 4. The pattern recognition method according to claim 2, wherein the second vector is made to enter the inner contour side at an angle that divides the bending angle by about half.