JP2000276598A - Method for recognizing image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make accurately performable the recognition processing and to make recognizable a recognition target even though the scale factors of an image are changed and the size of an input image is mutually different. SOLUTION: A recognition target composed of line segments passing through one point and has the angular information of an angle made by adjacent line segments with each other as characteristic quantity. A line segment detecting part 11 detects line segments in received image date, and an angle detecting part detects the angular information (angle plus array information) of an angle made by adjacent line segments with each other among detected line segments. A target matching part 13 judges whether or not calculated angular information coincides with a correct recognition target, and a non-target condition checking part 14 judges whether or not angles and line segments that do not exist in the correct recognition target, exist. A judging part 15 judges that a recognition target exists when the angular information is correct and non-target information does not exist.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method.

[0002]

2. Description of the Related Art For example, when checking the suitability and type of a work flowing on a transport line in an FA system, a method is employed in which a mark (pattern) printed on the work surface is determined. Then, an image recognition method may be used to make such a determination.

[0003] As is well known, one of the techniques for performing recognition determination (pattern matching) is that a template between a reference recognition pattern (template) and a recognition target image (object) is used. Evaluation is performed using distance and cross-correlation. Therefore, F
When applied to A, the same mark (pattern) as the mark to be printed on the work surface is stored and held as a reference recognition pattern. Then, the workpiece flowing on the transport line is imaged by a camera, and the obtained image data is matched with a recognition pattern, and it is determined whether or not they match. As a result, it is possible to specify the type of the work and determine whether or not the mark is missing.

[0004]

The above-described image recognition method has the following problems. In other words, when the transfer line has a certain width and the work is transferred randomly on the transfer line, the distance from the camera may vary depending on the position of the work on the line. Then, naturally, if the distances are different, dimensions on image data obtained when the same object is imaged are different (the farther the distance, the smaller the size).

By the way, the above-mentioned evaluation (recognition processing)
Size identity is also an important factor. That is,
Even if the shapes are the same (similar), if the dimensions are different, it is recognized as not a recognition pattern. Therefore, in the case of a system in which the distance between the camera and the work fluctuates as described above, even if the work is the same, it may be determined that the recognition pattern is the same or not the same depending on the distance to the camera. .

In order to avoid the influence of such a difference in dimensions, there is a method of preparing a plurality of recognition patterns of different dimensions and sequentially matching them. In the case of this method, not only does the number of matchings increase, alienating high-speed processing, but also holding a plurality of reference recognition patterns requires a large memory capacity of the dictionary.

If the magnification is known, for example, by measuring the distance with another device, the data obtained by thinning out the captured image or performing interpolation processing is used. There is also a method of normalizing the image by enlarging / reducing itself, and matching it with a reference recognition pattern. In this case, it is only necessary to create one recognition pattern based on the difference in size, and the matching process can be performed only once. However, this method is based on the premise that it is necessary to know an accurate magnification ratio, so that an FA or the like requires a separate distance measuring device, which is not preferable because the system becomes large.

On the other hand, it is considered that human vision recognizes the shape of an object in a complicated process, but the evaluation of the absolute size is not good, and the slight difference in size is recognized (distinguished). Is difficult to do. Therefore, even if images are distinguished without being determined to be the same in image recognition, if the difference is a slight difference, the same may be visually recognized as the same.

In particular, when the enlargement / reduction rate (hereinafter referred to as "magnification ratio") of the recognition pattern is about several percent to several tens percent, the two (the recognition pattern and the object to be recognized) are arranged side by side and directly compared. When compared, it is easy to understand because the size is different, but it is difficult to understand when only the recognition target is viewed alone.

[0010] By utilizing this property, for example, when the inspector moves on the transfer line and looks at the mark of the transferred work as described above, even if there is a displacement of the work on the transfer line as described above, The difference in the size of the mark due to the displacement is absorbed, and it can be determined whether or not the mark is the same as the recognition pattern.

Therefore, even in the recognition algorithm of the image recognition apparatus, in the case of a small magnification ratio based on the difference in distance as in the case of the above-mentioned human, there is no need for a pre-processing for absorbing the dimensional difference, and it is simple and small. It is desired to develop a device / method that can be recognized by the number of times.

The present invention has been made in view of the above background, and an object of the present invention is to perform accurate recognition processing even if the size of image data being processed is different due to scaling. An object of the present invention is to provide an image recognition method capable of determining a predetermined image.

[0013]

In order to achieve the above object, an image recognition method according to the present invention is a recognition method for a recognition device for recognizing a recognition target included in an image, wherein the recognition target is a line. And the angle information of the angle between the line segments is used as the feature amount. Then, perform a line segment detection step of detecting a line segment in the received image data, and then perform an angle detection step of obtaining angle information of an angle between predetermined line segments among the detected line segments,
A determination step is provided for determining whether or not the recognition target exists in the image data based on the obtained angle information (claim 1).

Even if the magnification of the image to be processed such as enlargement / reduction changes with respect to the original image (the image whose reference size is 100%), the angle information is not affected. Therefore, by performing image recognition based on the angle information, the magnification becomes strong, and an image including the recognition target can be reliably recognized.

[0015] Preferably, the angle information is an angle and an arrangement order of the angle. By adding the arrangement order to the judgment criteria, erroneous recognition is suppressed, and even if the arrangement order is included in the judgment condition, the processing load associated therewith does not increase significantly.

A recognition method in a recognition apparatus for recognizing a recognition target included in an image, wherein the recognition target is constituted by a line segment, and the inclination information of the line segment is used as a feature amount, and Performing a line segment detection step of detecting a line segment of the detected line segment, performing a tilt detection step of obtaining inclination information of a predetermined line segment among the detected line segments, based on the obtained inclination information, A determination step for determining whether or not a recognition target exists may be provided (claim 3).

Similar to the first aspect, even if the magnification changes with respect to the original image, the inclination information is not affected. Therefore, by performing image recognition based on such tilt information, the magnification becomes strong, and an image including the recognition target can be reliably recognized.

[0018] More preferably, non-target information (corresponding to an angle that does not exist or a line segment that does not pass through a point in the embodiment) that does not exist in the recognition target is stored and held. It is more preferable to perform the process by taking into account information on whether or not the non-target information exists in the data (claim 4).

When non-target information is included in image data, there is a high possibility that there is no recognition target in the image data. Therefore, by taking such information into account, the recognition rate is further improved. As a method of taking into account, as in the embodiment, a kind of AND (determining that there is a recognition target for the first time when there is a feature amount and there is no non-target information) is appropriately weighted and comprehensively performed. Of course, it may be determined.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the image recognition method according to the present invention will be described below in detail with reference to the accompanying drawings.
FIG. 1 shows an example of a recognition target used in the present invention. As shown in the figure, the recognition target 1 of the present embodiment is constituted by a plurality of line segments 1a to 1f intersecting at one point P. The conditions for this line segment are line segments 1d, 1f
, One end is located at one point P, and the other line segment 1
Although it is separated from the turn P as in a to 1c and 1e, any case where a point P exists on the extension line can be adopted. However, a line segment penetrating the point P, that is, a line segment having the point P on the line is not used.

By doing so, a radial pattern, a set of rectangular parallelepipeds, and the like, which are common in general printed matter and objects, etc.
The possibility of erroneously detecting a plurality of overlapping shapes or the like can be suppressed as much as possible. Also, from this point, it is preferable that the angle between adjacent line segments does not become a commonly used angle such as 30 degrees, 45 degrees, 60 degrees, and 90 degrees.

The reference data for recognition is angle information of the angle between the line segments constituting the recognition target 1. In the present embodiment, the angle between adjacent line segments is used, and the order is also taken into account. Therefore, as shown in FIG.
c, d, e, f] and their order.

In the present embodiment, whether or not the recognition target 1 having the above-described shape exists in the input image data is recognized based on the angle information. As a specific device configuration, as shown in FIG. 2, the input image data is sent to the line segment detection unit 11, and a line segment existing in the image data is extracted.

As the line segment extraction method in the line segment detection section 11, various methods such as a template matching method and a Hough transform method can be used. And
In the present embodiment, extraction is performed using template matching, and more specifically, it has a function of executing the flowchart shown in FIG.

That is, a template is applied to the input image data to extract a line segment (ST1, ST1).
2). Next, with respect to each line segment existing within a certain distance range with respect to the pixel to be processed (the above-mentioned one point P at the center of each line segment), the line segment is located on the pixel to be processed (one end is located at one end). In the case where the pixel touches the line segment, both ends of the line segment are located on both sides of the pixel to be processed. It is determined whether or not the line segment passes through the processing target pixel (one point) as existing (ST3).

If the branch determination in step 3 is Yes, the specifications of the line segment (for example, the position coordinates of both ends) are output to the angle detector 12 (ST4), and the branch determination in step 3 is No. In this case, the specifications of the line segment (for example, position coordinates of both ends) are output to the non-target matching unit (ST
4).

On the other hand, the angle detecting section 12 calculates the angle between adjacent line segments, and specifically implements the function shown in the flowchart of FIG. That is, an arbitrary line segment among the given line segments is determined as the first line segment, and the specifications of the line segments are sequentially extracted clockwise (ST11, ST1).
2). As a result, the position of two adjacent line segments is known, and the angle is obtained (ST13), and the angle is output, that is, given to the target matching unit 13 and the non-target condition check unit 14 (ST14).

When the last line segment is processed, that is, when the next line segment is the first extracted line segment, the angle is obtained and the process is terminated with the output state (ST).
14).

As a result, for example, when a line segment as shown in FIG. 5 is detected, a line segment L1 extending diagonally up and down is selected first (ST11), and then from step 12 to 1
By repeatedly executing step 4, [f, a, b, c, d, e] are obtained in order as the angle of the adjacent line segment,
This angle group is output. Actually, the angle may be output each time the angle is calculated, or may be output all at once after the angle is calculated.

The target matching unit 13 refers to the data stored in the first dictionary 13a in which the correct target angles are associated with the order of the targets. Of these, it is determined whether or not the angles between adjacent line segments are arranged in the correct order, and more specifically, it has a function of realizing a flowchart as shown in FIG.

That is, first, it is determined whether or not the angle group given from the angle detection unit 12 matches the correct target in both the order and the angle method (ST21). If they match, the object is determined to be a detection target, and information (true) of target matching (with target) is output (ST2).
4). On the other hand, if they do not match, the process jumps to step 22, turning the head angle of the angle group to the tail (ST22),
In this turning process, it is determined whether the angle group has made one round, that is, whether the angle group has returned to the initial state (ST23). If it has not completed one round, the process returns to step 21 to determine again whether or not it matches the correct target.

If there is no match even after one round of the angle rearrangement, the process proceeds from step 23 to step 25, where it is determined that the object to be processed is not a recognition target, and no target information (false) is output. .

To explain with a specific example, as shown in FIG. 1B, the angle groups [a, b, c, d,
e, f] are stored in the first dictionary 13a. Then, the target matching unit 13 sets the first dictionary 13a
To obtain the information of the correct angle group.

In this case, a line segment as shown in FIG. 5 is detected, and as described above, [f, a,
b, c, d, e], the first step 2
Since the judgment at 1 does not match, the process proceeds to step 22, and by turning the head angle f to the tail, [a, b, c,
d, e, f] are obtained. Then, since it matches the reference data, the result is Yes in step 21 and the information of the target matching is output.

On the other hand, for example, a line segment as shown in FIG. 7A (the number of line segments indicated by arrows is larger than the line segment of a correct target).
Is detected, the angle information is [a, b,
c, d, e, g, h], and any angle at the head of the angle group does not match the case of the correct target. Therefore, information without a target is finally output.

The non-target condition checking unit 14
Is based on the angle group given from the angle detection unit 12 and the information of the “line segment that does not pass through one point” from the line segment detection unit 11,
It is determined whether or not it is a non-target. When checking the angle group, the second dictionary 1
The determination is made with reference to the information on the angle that does not exist in the correct target stored in 4a. Here, information about angles that do not exist in the correct target can be significantly increased if all the non-existing angles are included.
Angles that make up the target (does not include order information)
Is stored, and when an angle other than the angle is detected, it is determined that “there is an angle not included in the target”.

Further, as other information, for example, 9
If an angle that is not used as an angle of a line segment constituting the target, such as 0, 60, or 45 degrees, is determined in advance, the angle may be stored.
In that case, if it matches the stored angle,
It is determined that there is an angle that is not included in the target.

The specific processing is as shown in the flowchart of FIG. That is, first, it is determined whether or not there is an angle that is not included in a correct target, based on the angle information stored in the second dictionary 14a, for each piece of angle information constituting the given angle group (ST31). If there is an angle that is not included, it can be determined that the angle is not a correct target, and the process proceeds to step 34, where target matching invalidity (true: matches the non-target condition) is output.

If there is no angle that is not included in the correct target, the process proceeds to step 32, where the line segment detector 11
It is determined whether there is an invalid line segment that does not pass through one point from the target. In this process, if there is actually a detection signal from the line segment detection unit 11, the determination is Yes, and the process proceeds to step 34, and the target matching invalid (true) is output. Also,
If there is no invalid line segment, the process proceeds to step 33 and outputs target matching validity (false: does not satisfy the non-target condition).

A specific example will be described.
It is assumed that (a, b, c, d, e, f: no information on the order) is stored in a. Then, the non-target condition is an angle other than the above angle.

In this case, when a line segment as shown in FIG. 7A (the number of line segments indicated by arrows is larger than the line segment of the correct target) is detected, the angles g, Since h is detected, the condition is met (Yes) in step 31 and target matching invalid (true) is output.

When a line segment as shown in FIG. 3B (a line segment that does not pass through one point indicated by an arrow) is detected, there is no angle not included in the correct target, but a line indicated by an arrow is detected. Since the minute does not pass through the pixel to be processed, the condition is met (Yes) in the judgment of step 32, and target matching invalid (false) is output.

When the output of the target matching unit 13 is “true” and the output of the non-target condition checking unit 14 is “false”, the determination unit 15 determines that the output is “true” and that the recognition target exists. Judgment is made and the recognition result is output by the output unit 16.

The above-described processing is performed each time the pixel to be processed (one point P at the center of each line segment) is sequentially moved with respect to the given image data by, for example, raster scanning.

If the dictionary data stored in the second dictionary 14a is only a correct target angle, if the step 21 of the target matching unit 13 is Yes, the judgment of the step 31 of the non-target condition check unit 14 must be performed. No, and if Step 21 is No, Step 31 becomes Yes.

Nevertheless, the reason why this branch decision is made is that by performing a double check on both,
The algorithm that minimizes the probability of misjudgment as much as possible, if there is a different angle, immediately recognizes that it is abnormal and decides that it is not the target as soon as possible. This is for the purpose of commonality with the above.

Therefore, in order to simplify the determination process, step 31 need not be provided. In that case,
Since the determination in step 32 uniquely corresponds to the presence or absence of the output from the line segment detection unit 11, the output of the line segment detection unit 11 may be provided to the determination unit 15 directly or via a delay unit or the like. Good.

In the above embodiment, the angle between adjacent line segments is used as angle information.
The present invention is not limited to this. For example, it is possible to take various measures such as, for example, setting every other line, setting the angle between each line segment with respect to a certain line segment, and the like. However, taking the angle between the adjacent line segments is preferable in that the processing is simple and the processing can be performed at high speed.

Furthermore, in the above-described embodiment, the condition of angle + order (arrangement order) is used as a condition. However, information of order may be omitted from the search condition, and the search may be performed based on only angle information. In that case, for example, not only angles formed by adjacent line segments but also all angles of 180 degrees or less formed by any combination of line segments are extracted, and array information is extracted as a combination of angles. You may. As a specific example, in the case of FIG. 5, (a, b, c, d, e, f, a +
b, b + c, c + d, d + e, e + f, f + a, a + b
+ C, b + c + d, f + a + b).

FIGS. 9 to 11 show a second embodiment of the present invention. As shown in FIG.
Are directly applied to the target matching unit 13 'and the non-target matching unit 14', respectively.
The outputs of 3 'and 14' are supplied to a determination unit 15 to determine whether or not the pattern is correct. That is, when the matching pattern is detected (true) by the target matching unit 13 'and the matching pattern is not detected (false) by the non-target matching unit 14', the output of the determination unit 15 is true and the detection is determined. .

More specifically, the target matching unit 13 'detects the first black pixel as shown in FIG.
A mask M having a mask region M1 and a second mask region M2 for detecting white pixels located on both sides of the mask region M1 is prepared, and the mask M is arranged at a predetermined position of each line of a correct target.

With this configuration, even if the magnification changes slightly, the angle does not change, so that "true" is output for a correct target. That is, if there is no line segment where it should be, "false" is output. However, if all the line segments of the correct target are present and there are many line segments, “true” is output. In this case, however, the non-target matching unit detects such a large number of line segments. There is no.

On the other hand, the non-target matching section 14 '
As shown in FIG. 11, in a certain target area centered on a certain point P (a point through which each line segment of the target passes), a region excluding a portion where a line segment exists and an extension thereof is defined as a mask region M3. Is used. Then, it is determined whether or not a black pixel exists in the mask area M3, and if so, “true” is output.

In this embodiment, instead of directly forming a line segment, a black pixel (a pixel forming a line) exists at a position where a line segment should originally exist, and a white pixel exists at a position where there is no line segment. By recognizing that there is, it is possible to indirectly recognize the presence of a line segment and determine that the angle of the line segment is in a predetermined condition by using a template, so that it can be easily realized with a simple circuit, and Since it is template matching,
You can make a quick decision.

However, since there is a possibility of erroneous recognition in a situation where noise or the like is extremely large, the threshold value of the non-target condition is increased if necessary, and noise removal processing or the like is performed as preprocessing.

Further, since the target is affected by the attitude (rotation angle) of the target, the mask (template) used in both the target matching unit 13 'and the non-target matching 14' has a plurality of masks corresponding to different rotation angles. I have prepared a template,
Alternatively, it is necessary to rotate the template at the time of determination.

In each of the above-described embodiments and modified examples, the determination is made based on the angle information of the angle formed by an arbitrary line segment. However, the present invention is not limited to this. The determination may be made based on the combination of the inclination of the object itself.

That is, as shown in FIG. 12, the inclination (a ', b',
c ', d', e ', f'), and use them as reference data. The hardware configuration of the first embodiment shown in FIG. 1 can be used as it is. The difference is that the angle detection unit does not calculate the angle of the line segment in the process of step 12, but uses a reference line (for example, a horizontal line).
The inclination is calculated by obtaining an angle from the angle.

The target matching unit determines whether or not the detected inclination (group of angles) of each line segment matches the reference data registered in advance. If they match, "true" is output, and if they do not finally match, "false" is output. At this time, the posture (rotation angle) of the recognition target may be different from that of the reference data. Therefore, it is necessary to correct the rotation so that the rotation angle is not affected by the posture.

As for the function of the non-target condition checking unit, considering the correction of the rotation angle, the process is simplified if the step 31 shown in FIG. 31 may be provided).

[0061]

As described above, in the image recognition method according to the present invention, even if the size of the input image is different from that of the original image (100% of the same image as the recognition pattern), the line segment The angle and tilt do not change. Therefore, if the determination is made based on the angle (inclination) information, it can be accurately determined whether or not the target is correct regardless of the presence or absence of deformation.

According to the second aspect of the present invention, misrecognition is reduced because array information of angles is added. Further, in the invention according to claim 4, since the judgment is made in consideration of the non-target information, the probability of erroneous recognition is further reduced.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating an example of a recognition target.

FIG. 2 is a block diagram showing an image recognition device suitable for implementing the first embodiment of the image recognition method according to the present invention.

FIG. 3 is a flowchart illustrating a function of a line segment detection unit.

FIG. 4 is a flowchart illustrating a function of an angle detection unit.

FIG. 5 is a diagram illustrating the operation of an angle detection unit.

FIG. 6 is a flowchart illustrating functions of a target matching unit.

FIG. 7 is a diagram illustrating an operation.

FIG. 8 is a flowchart illustrating a function of a non-target condition checking unit.

FIG. 9 is a block diagram showing an image recognition device suitable for carrying out the second embodiment of the image recognition method according to the present invention.

FIG. 10 is a diagram illustrating an example of a template used in a target matching unit.

FIG. 11 is a diagram illustrating an example of a template used in a non-target matching unit.

FIG. 12 is a diagram illustrating a third embodiment of the present invention.

[Description of Signs] 1 Recognition target 1a to 1f Line segment 11 Line segment detection unit 12 Angle detection unit 13, 13 'Target matching unit 14 Non-target condition checking unit 14' Non-target matching unit 15 Judgment unit

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Keitaro Taniguchi F-term 5L096 BA03 CA02 DA02 FA03 FA67 HA07 JA11 JA16 KA01 KA04 in Omron Co., Ltd.

Claims (4)

[Claims] 1. A recognition method in a recognition device for recognizing a recognition target included in an image, wherein the recognition target is formed of line segments, and angle information of an angle between the line segments is used as a feature amount. Performing a line segment detection step of detecting a line segment in the detected image data, performing an angle detection step of obtaining angle information of an angle formed by predetermined line segments among the detected line segments, based on the obtained angle information. And determining whether the recognition target exists in the image data. 2. The image recognition method according to claim 1, wherein the angle information is an angle and an arrangement order of the angle. 3. A recognition method in a recognition device for recognizing a recognition target included in an image, wherein the recognition target is constituted by a line segment, and the inclination information of the line segment is used as a feature amount. Performing a line segment detection step of detecting a line segment of the line segment; performing a tilt detection step of obtaining inclination information of a predetermined line segment among the detected line segments; based on the obtained inclination information, An image recognition method comprising: a determination step of determining whether a recognition target exists. 4. The non-target information that does not exist in the recognition target is stored and held, and the determination step is performed in consideration of information on whether or not the non-target information exists in the image data. The image recognition method according to claim 1.