JP2021071979A - Template matching method - Google Patents

Abstract

To provide a template matching method having an excellent detection accuracy.SOLUTION: A template matching method includes: a step S2 of generating a similarity map M1 in which similarities of a reference image with a picked-up image from an image pick-up device are arranged according to the coordinates of the reference image in the picked-up image by using an imaging-type encoder; a step S3 of generating a processing similarity map M2 in which processing similarities obtained based on the plurality of similarities adjacent to each other in the similarity map M1 are arranged according to the coordinates of the similarities in the similarity map; and a step S4 of extracting the maximum processing similarity from the processing similarity map M2.SELECTED DRAWING: Figure 4

Description

The present invention relates to a template matching method.

An optical rotary encoder is generally known as a kind of encoder. For example, the encoder described in Patent Document 1 reads a code plate on which a numerical pattern such as a Gray code and a striped pattern are formed by an image sensor, and detects a position from the read numerical pattern and the striped pattern.

Further, as an encoder capable of easily and stably realizing higher detection accuracy than an optical encoder as in Patent Document 1, the reference image and the captured image are used and are most similar to the reference image in the captured image. An image pickup type encoder that detects a position by performing template matching that specifies pixel coordinates can be considered.

Japanese Unexamined Patent Publication No. 63-187118

However, in such an image pickup type encoder, the positional relationship between the code plate to be imaged and the image sensor may shift due to distortion of the cord plate and shaft that occur over time, mechanical rattling, and the like. is there. As a result, the positional relationship between the target mark and the pixel may deviate between the reference image and the captured image, and the degree of similarity between the reference image and the captured image may be significantly reduced. Then, even though the marks are different between the reference image and the captured image, there is a risk of erroneous template matching.

In the template matching method according to the application example of the present invention, the similarity between the image captured by the image sensor and the reference image is obtained by using an imaging encoder.
A similarity map in which the similarity is arranged according to the coordinates of the reference image in the captured image is generated.
A machining similarity map is generated in which the machining similarity obtained based on a plurality of adjacent similarity in the similarity map is arranged according to the coordinates of the similarity in the similarity map.
It is characterized in that the maximum processing similarity is extracted from the processing similarity map.

It is a side view which shows the robot which concerns on embodiment of this invention. It is sectional drawing which shows the encoder which the robot of FIG. 1 has. It is a top view which shows the mark provided on the 1st arm. It is a flowchart which shows the template matching method used in the encoder. It is a figure explaining the method of acquiring a reference image. It is a figure which shows the similarity map. It is an enlarged view of the region Q1 in FIG. It is a figure which shows the state which the pixel shift does not occur. It is a figure which shows the state which the pixel shift occurs. It is a figure which shows the similarity degree map in the state where a pixel shift occurs. It is an enlarged view of the region Q21 in FIG. It is an enlarged view of the region Q22 in FIG. It is a figure which shows the area to be set in the similarity map. It is a figure which shows the processing similarity map. It is a figure which shows the processing similarity map. It is a figure which shows the modification of the area set in the similarity map.

Hereinafter, the template matching method of the present invention will be described in detail based on the preferred embodiments shown in the accompanying drawings.

FIG. 1 is a side view showing a robot according to an embodiment of the present invention. FIG. 2 is a cross-sectional view showing an encoder included in the robot of FIG. FIG. 3 is a plan view showing a mark provided on the first arm. FIG. 4 is a flowchart showing a template matching method used in the encoder. FIG. 5 is a diagram illustrating a method of acquiring a reference image. FIG. 6 is a diagram showing a similarity map. FIG. 7 is an enlarged view of the region Q1 in FIG. FIG. 8 is a diagram showing a state in which pixel deviation does not occur. FIG. 9 is a diagram showing a state in which pixel shift occurs. FIG. 10 is a diagram showing a similarity map in a state where pixel deviation occurs. FIG. 11 is an enlarged view of the region Q21 in FIG. FIG. 12 is an enlarged view of the region Q22 in FIG. FIG. 13 is a diagram showing a region to be set in the similarity map. 14 and 15 are diagrams showing processing similarity maps, respectively. FIG. 16 is a diagram showing a modified example of the region set in the similarity map. In the following, for convenience of explanation, the upper side in FIG. 1 is referred to as “upper” and the lower side is referred to as “lower”. Further, the base side in FIG. 1 is referred to as a "base end", and the opposite side (end effector side) is referred to as a "tip side". Further, the vertical direction in FIG. 1 is defined as the "vertical direction", and the horizontal direction is defined as the "horizontal direction".

The robot 100 shown in FIG. 1 is a horizontal articulated robot, and is used, for example, in a manufacturing process for manufacturing precision equipment and the like, and can grip and transport precision equipment and parts. The robot 100 includes a base 110, a first arm 120 that rotates around the first axis J1 with respect to the base 110, and a second arm 130 that rotates around the second axis J2 with respect to the first arm 120. A work head 140 provided at the tip of the second arm 130, which includes a spline shaft 141 that rotates around the third axis J3 with respect to the second arm 130 and moves up and down along the third axis J3. It has an end effector 150 mounted on the tip of the spline shaft 141. Then, such a robot 100 includes a first encoder that detects the rotation angle of the first arm 120 around the first axis J1 with respect to the base 110, and a second axis J2 of the second arm 130 with respect to the first arm 120. A second encoder that detects the rotation angle around it is provided. The template matching method of the present invention can be applied to both the first encoder and the second encoder. Hereinafter, the configuration applied to the first encoder will be described, and the first encoder will be described as the encoder 1.

As shown in FIG. 2, the encoder 1 is provided on the scale unit 2 provided on the first arm 120, the detection unit 3 provided on the base 110 to detect the scale unit 2, and electrically on the detection unit 3. It has a circuit unit 4 and a connected circuit unit 4.

The scale portion 2 is provided at a portion of the first arm 120 facing the base 110. As shown in FIG. 3, the scale portion 2 has a unique pattern provided on the surface of the first arm 120 and arranged along the first axis J1 at a position different from that of the first axis J1. doing. The "unique pattern" means that when the scale portion 2 is rotated around the first axis J1 over a required angle range, the same pattern does not appear twice or more in a predetermined region in the captured image G described later. .. Therefore, each of the plurality of portions of the scale portion 2 having different positions can be used as the mark 21 for identifying the position of the scale portion 2 in the circumferential direction. Note that FIG. 3 illustrates a case where a plurality of marks 21 are arranged along the circumference centered on the first axis J1. Further, the positions, sizes, numbers, and the like of the marks 21 shown in FIG. 3 are examples, and are not limited thereto.

The unique pattern of the scale portion 2 is formed by uniquely arranging a plurality of dots 20 on a black background on a white background. The shape of the dot 20 is circular in the drawing, but is not limited to this, and may be, for example, an ellipse, a quadrangle, or an irregular shape. Further, the pattern of the scale unit 2 is not limited to a dot pattern such as the above-mentioned pattern composed of a plurality of dots 20, and is composed of, for example, a pattern composed of linear lines and a curved line. It may be a pattern composed of a combination of at least two of a pattern, a dot, a linear line, and a curved line, or an inverted pattern thereof. Further, the pattern of the scale unit 2 may be a pattern that can be detected by the detection unit 3, and may be, for example, a pattern formed by ink such as dye or pigment, a pattern having an uneven shape, a pattern formed by a natural object, or the like. There may be.

Further, the mark 21 of the scale unit 2 is not limited to the one using the above-mentioned pattern, and for example, a number may be used, characters such as Roman characters, Arabic characters, and Kanji characters may be used, and characters other than characters may be used. Symbols, symbols, signs, marks, designs, one-dimensional barcodes, QR codes (registered trademarks) and the like may be used.

Further, the scale portion 2 is not limited to the case where it is provided directly on the surface of the first arm 120, and may be provided on, for example, a sheet-like member attached to the surface of the first arm 120. However, it may be provided on a plate-shaped member provided so as to rotate together with the first arm 120. That is, the member provided with the scale portion 2 may be a member that rotates around the first axis J1 with respect to the base 110 together with the first arm 120.

As shown in FIG. 2, the detection unit 3 includes an image sensor 31 provided in the base 110 and an imaging optical system 32 provided in the opening of the base 110. Then, the image sensor 31 images the image pickup region RI which is a part of the scale unit 2 in the circumferential direction via the image pickup optical system 32. The detection unit 3 may be provided with a light source for illuminating the image pickup region RI of the image pickup device 31, if necessary. The image pickup device 31 is not particularly limited, and for example, a CCD (Charge Coupled Devices), a CMOS (Complementary Metal Oxide Semiconductor), or the like can be used.

The image sensor 31 converts the captured image into an electric signal for each pixel and outputs the image. The image sensor 31 can be applied to either a two-dimensional image sensor (area image sensor) or a one-dimensional image sensor (line image sensor). When a two-dimensional image sensor is used, a two-dimensional image having a large amount of information can be acquired, and it is easy to improve the detection accuracy of the mark 21 by template matching described later. As a result, the rotational state of the first arm 120 can be detected with high accuracy. On the other hand, when the one-dimensional image sensor is used, the image acquisition cycle, that is, the frame rate, is high, so that the detection frequency can be increased, which is advantageous at high speed operation.

Here, as shown in FIG. 3, the image pickup region RI of the image pickup device 31 is set on the lower surface of the first arm 120 so as to overlap a part of the scale portion 2 in the circumferential direction. As a result, the image sensor 31 can image the mark 21 in the image pickup region RI. Therefore, the rotational state of the first arm 120 can be known by reading the mark 21 located in the imaging region RI.

As shown in FIG. 2, the circuit unit 4 includes a processing unit 5 such as a CPU (Central Processing Unit) and a template matching dedicated arithmetic circuit, and a storage unit 6 such as a ROM (Read only memory) and a RAM (Random Access Memory). Has. Here, the storage unit 6 stores instructions that can be read by the processing unit 5, and the processing unit 5 realizes various functions by appropriately reading and executing the instructions from the storage unit 6. Such a circuit unit 4 can be configured by using, for example, an ASIC (application specific integrated circuit) or an FPGA (field-programmable gate array). By hardwareizing the circuit unit 4 using the ASIC or FPGA in this way, it is possible to increase the processing speed, reduce the size, and reduce the cost of the circuit unit 4.

The processing unit 5 estimates the relative rotational state of the base 110 and the first arm 120 based on the detection result of the detection unit 3. Examples of the rotation state include a rotation angle, a rotation speed, a rotation direction, and the like. In particular, the processing unit 5 recognizes the mark 21 by template matching the image captured by the image sensor 31 using the reference image, and uses the recognition result to rotate the first arm 120 with respect to the base 110. Estimate the moving angle. The acquisition of the reference image will be described in detail below. Further, the template matching in the processing unit 5 and the estimation of the rotation state using the template matching will be described in detail based on the flowchart shown in FIG.

The template matching in the processing unit 5 and the estimation of the rotation state using the template matching are performed by the captured image acquisition step of acquiring the captured image G using the imaging element 31 and the template matching with respect to the captured image G using the reference image TA. A similarity map generation step for generating a similarity map M1 by performing the above, a machining similarity map generation step for generating a machining similarity map M2 based on the similarity map M1, a machining similarity map M2, and a machining similarity map M1. Has an estimation step of estimating the position of the first arm 120 based on. Hereinafter, a detailed description will be given based on the flowchart shown in FIG.

1. 1. Acquisition of Reference Image The encoder 1 acquires the reference image TA used for template matching in advance prior to estimating the rotational state of the first arm 120 with respect to the base 110 using template matching. The acquisition of the reference image TA may be performed only once before the first template matching, but may be performed after that as needed. In that case, the reference image TA used for template matching can be updated with the newly acquired reference image TA.

When acquiring the reference image, the first arm 120 is appropriately rotated around the first axis J1 with respect to the base 110, and a plurality of marks 21 are imaged for each of the marks 21 by the image sensor 31. Then, by trimming each of the obtained captured images G and deleting unnecessary portions, a reference image TA for each mark 21 is generated. That is, the reference image TA is generated for the number of marks 21. Each of the generated reference image TAs is associated with the pixel coordinate information and the angle information and stored in the storage unit 6.

More specifically, as shown in FIG. 5, the captured image G has a shape corresponding to the imaging region RI, and has two sides extending along the X direction and the Y direction orthogonal to the X direction. It forms a rectangle with two sides extending along. Further, the two sides extending along the X-axis direction of the captured image G are arranged so as to be along the moving direction of the mark 21 indicated by the arrow as much as possible. Further, the captured image G has a plurality of pixels arranged in a matrix in the X direction and the Y direction. Here, the pixel positions are represented by a pixel coordinate system (X, Y) represented by "X" indicating the position of the pixel in the X direction and "Y" indicating the position of the pixel in the Y direction. To. Further, the pixel at the upper left end of the image of the captured image G is set as the origin pixel (0,0) of the image coordinate system (X, Y).

For example, when generating the reference image TA corresponding to the mark 21A in FIG. 5, the first arm 120 is appropriately rotated with respect to the base 110, and the mark 21A is moved to a predetermined position in the captured image G, or the X axis in the drawing. It is positioned on the center line LY set at the center in the direction. The rotation angle θA0 of the first arm 120 with respect to the base 110 when the mark 21 is located on the center line LY has been acquired in advance by measurement or the like.

Next, the state in which the mark 21 is positioned on the center line LY is imaged by the image sensor 31, and the obtained image G is captured in a rectangular pixel range that is the minimum necessary range including the mark 21A. Trim. As a result, the reference image TA corresponding to the mark 21A is obtained, and the obtained reference image TA is stored in the storage unit 6. At this time, the reference image TA includes the above-mentioned angle information regarding the rotation angle θA0 and the pixel information regarding the reference pixel coordinates (Xp0, Yp0) which are the pixel coordinates of the reference pixel P0 included in the pixel range of the reference image TA. It is associated and stored. That is, the reference image TA, the angle information, and the pixel coordinate information form one template set used for template matching.

When the encoder 1 is an absolute type, the angle information in this template set is an absolute angle. When the encoder 1 is an increment type, the angle information in this template set is a relative angle.

2. Template matching and estimation of the rotating state using the template matching The template matching in the processing unit 5 and the estimation of the rotating state using the template matching will be described in detail below.

[Captured image acquisition step]
First, the processing unit 5 uses the image sensor 31 to image the image pickup region RI that is a part of the scale unit 2 to obtain the image pickup image G (step S1).

[Similarity map generation step]
First, the processing unit 5 performs template matching on the image G of the image sensor 31 using the reference image TA prepared in advance to generate the similarity map M1 (step S2). Specifically, the processing unit 5 sets the entire area of the captured image G as the search area, superimposes the reference image TA on the search area, and calculates the similarity of the overlapping portion between the search area and the reference image TA. The processing unit 5 moves the reference image TA along the pixel arrangement of the captured image G, and calculates the similarity of the overlapping portion between the search region and the reference image TA for the pixels in the entire search region. Specifically, the processing unit 5 moves the pixel coordinates (Xp, Yp) of the reference pixel P0 set in the reference image TA from the start coordinate PSp to the end pixel PEp one pixel at a time, and pixels in the entire search area. The degree of similarity between the search area and the overlapping portion of the reference image TA is calculated. That is, the processing unit 5 calculates the similarity for each pixel coordinate (Xp, Yp) of the reference pixel P0 of the reference image TA. Then, the processing unit 5 arranges the similarity calculated for each pixel coordinate (Xp, Yp) corresponding to the position of the pixel coordinate (Xp, Yp), that is, along the order of the pixel coordinates (Xp, Yp). A similarity map M1 as shown in FIGS. 6 and 7 arranged side by side is generated. At this time, the similarity map M1 is a map including a point at which the similarity is maximized, that is, a plurality of peaks. The region Q1 shown in FIG. 7 is a region including pixel coordinates (Xp, Yp) = (Xn, Yn) having the maximum similarity. The generated similarity map M1 is stored in the storage unit 6. As a method for obtaining the degree of similarity, there are SAD, SSD, ZNCC and the like. In SAD and SSD, the value is the minimum, and in ZNCC, the similarity is the maximum when the value is the maximum. In the figure, since the degree of difference obtained by using SAD is described as the degree of similarity, the degree of similarity is maximized when the value is the minimum.

In the case of conventional template matching, the processing unit 5 next selects the maximum similarity degree from the similarity degree map M1 stored in the storage unit 6. In the example of FIG. 7, the similarity of pixel coordinates (Xp, Yp) = (Xn, Yn) is selected. Then, the processing unit 5 determines the pixel coordinates (Xp, Yp) of the reference pixel P0 of the reference image TA corresponding to the selected similarity as the pixel coordinates of the mark 21, and determines the position of the mark 21 in the captured image G. To detect.

Here, for example, as shown in FIG. 8, if the positional relationship between the target mark 21A and the pixel Px in the reference image TA and the captured image G is the same, the similarity map M1 shown in FIG. 7 is obtained. , The similarity measured at pixel coordinates (Xp, Yp) = (Xn, Yn) is sufficiently larger than the similarity measured at other pixel coordinates (Xp, Yp). Therefore, the position of the mark 21A in the captured image G can be detected with high accuracy.

On the other hand, as shown in FIG. 9, the positional relationship between the target mark 21A and the pixel Px in the reference image TA and the captured image G may deviate by about 0.5 pixel. In this case, the peak having the maximum similarity is also shifted by about 0.5 pixel. FIG. 10 shows a similarity map M1 generated in a state where pixel shift occurs. The region Q21 shown in FIG. 11 is a region including pixel coordinates (Xp, Yp) = (Xn, Yn) that maximizes the similarity when there is no pixel deviation. The region Q22 shown in FIG. 12 is a region including pixel coordinates (Xp, Yp) = (Xz, Yz) in which a mark 21 similar to the target mark 21A but different from the mark 21 exists. Since the similarity is calculated discretely in units of one pixel, it is different from the reference image TA of adjacent pixel coordinates (Xp, Yp) = (Xn, Yn) as shown in the similarity map M1 shown in FIG. The similarity of the peak with the maximum similarity is shared with the reference image TA of pixel coordinates (Xp, Yp) = (Xn-1, Yn). Therefore, the similarity measured at the pixel coordinates (Xp, Yp) = (Xn, Yn) is significantly lower than the similarity when there is no pixel deviation. Therefore, when a mark 21 similar to the target mark 21A exists in the captured image G, the pixel coordinates (Xp, Yp) = (as shown in FIGS. 11 and 12) = ( There is a possibility that a “reversal phenomenon” may occur in which pixel coordinates (Xp, Yp) = (Xz, Yz) having a similarity larger than the actually measured similarity in Xn, Yn) occur. In other words, when searching for the peak with the maximum similarity from the similarity map M1, the similarity is discretely sampled in units of one pixel. Therefore, for example, the peak having the maximum sampling range is removed by pixel deviation. And when it coincides with the next largest peak and the difference in similarity between the maximum peak and the next largest peak is small, a "reversal phenomenon" may occur. When such a reversal phenomenon occurs, the processing unit 5 determines the pixel coordinates (Xp, Yp) = (Xz, Yz) as the pixel coordinates of the mark 21A. As a result, the position of the mark 21A is erroneously determined.

Therefore, in order to suppress such erroneous determination of the position of the mark 21, the processing unit 5 of the present embodiment executes the following processing after generating the similarity map M1.

[Processing similarity map generation step]
First, the processing unit 5 processes the similarity map M1 stored in the storage unit 6 to generate the processing similarity map M2 (step S3). Specifically, first, as shown in FIG. 13, the processing unit 5 has a square-shaped region including a total of four pixel coordinates (Xp, Yp) of 2 rows and 2 columns arranged adjacently in the X direction and the Y direction. Set R. Next, the processing unit 5 calculates the average value of the similarity of the four pixel coordinates (Xp, Yp) included in the region R as the processing similarity. The processing unit 5 moves the area R along the coordinates of the similarity map M1 and calculates the processing similarity for the coordinates of the entire similarity map M1. Specifically, the processing unit 5 moves the coordinates (Xs, Ys) of the reference pixel coordinates S0 set in the area R from the start coordinate PSs to the end pixel PEs one by one, and covers the entire similarity map M1. For the coordinates, the processing similarity is calculated. That is, the processing unit 5 calculates the processing similarity for each coordinate (Xs, Ys) of the reference pixel coordinate S0.

Then, the processing unit 5 arranges the processing similarity calculated for each coordinate (Xs, Ys) corresponding to the position of the coordinate (Xs, Ys), that is, along the order of the coordinates (Xs, Ys). A machining similarity map M2 as shown in FIGS. 14 and 15 arranged side by side is generated. The generated processing similarity map M2 is stored in the storage unit 6. The machining similarity map M2 shown in FIG. 14 corresponds to the similarity map M1 shown in FIG. 11, and the machining similarity map M2 shown in FIG. 15 corresponds to the similarity map M1 shown in FIG. The processing similarity is not limited to the average value of the four similarity included in the region R, and may be, for example, the total value of the four similarity. By setting the processing similarity as the average value or the total value of the four similarities included in the region R, it becomes easy to obtain the processing similarity. In addition to this, the processing similarity may be a value obtained by substituting four similarity into a predetermined formula, for example.

[Estimation step]
Next, the processing unit 5 extracts the maximum processing similarity from the plurality of processing similarity included in the processing similarity map M2 stored in the storage unit 6 (step S4). Next, the processing unit 5 extracts the maximum similarity from the four similarity included in the region R corresponding to the extracted processing similarity (step S5). Then, the processing unit 5 sets the pixel coordinates (Xp, Yp) = (Xn, Yn) of the reference pixel P0 of the reference image TA corresponding to the extracted similarity as the matching position, that is, the mark, based on the similarity map M1. It is determined as the pixel coordinates of 21 (step S6). Further, the processing unit 5 outputs the determined matching position, that is, the rotation angle θ based on the pixel coordinates (step S7).

At this time, the position is deviated from the reference image TA at the reference pixel coordinates (Xp0, Yp0) by the number of pixels (Xn−Xp0) in the X-axis direction. Therefore, the distance between the center of the image pickup region RI and the first axis J1 is r, and the width of the region on the image pickup region RI corresponding to one pixel of the image sensor 31 in the X-axis direction (one pixel of the image sensor 31). The rotation angle θ of the first arm 120 with respect to the base 110 can be obtained by using the following equation (1), where W is the visual field size per unit.

In this formula (1), (Xn-Xp0) × W is the actual position corresponding to the reference pixel coordinates (Xp0, Yp0) of the reference image TA and the pixel coordinates of the reference image TA in which the above-mentioned similarity is the maximum value. It corresponds to the distance from the actual position corresponding to (Xn, Yn). Further, 2rπ corresponds to the length (circumferential length) of the locus of the mark 21 when the first arm 120 is rotated 360 ° with respect to the base 110. As described above, θA0 is the rotation angle of the first arm 120 with respect to the base 110 when the mark 21A is positioned at a predetermined position. The rotation angle θ is the angle at which the first arm 120 rotates with respect to the base 110 from the reference state (0 °).

From the above, the rotation angle θ of the first arm 120 with respect to the base 110 is estimated. Such template matching and calculation of the rotation angle θ using the same are performed for the other marks 21 in the same manner. Here, at an arbitrary rotation angle θ, at least one mark 21 is displayed in the captured image G without chipping, and a reference image corresponding to each mark 21 is registered so that template matching is possible. This makes it possible to prevent an angle range in which template matching is not possible.

According to the processing similarity map M2, since the average value of the similarity between the adjacent coordinates is used as the processing similarity, the similarity divided by the adjacent coordinates can be added up, and the similarity in the similarity map M1 can be added up. It produces the same effect as averaging the peaks by pixel width. Therefore, when the similarity is discretely sampled in units of one pixel using the processing similarity map M2, the same sampling value can be obtained regardless of whether the peak that maximizes the sampling range is removed or matches, and the maximum is obtained. Even if the difference in similarity between the peak and the next largest peak is small, there is no mistake. Therefore, the influence of the pixel shift as shown in FIG. 9 can be canceled, and the pixel coordinates of the mark 21 are not the pixel coordinates (Xp, Yp) = (Xn, Yn) but the pixel coordinates (Xp, Yp). It is possible to effectively suppress the erroneous selection of = (Xz, Yz). As described above, according to the above-mentioned method, excellent detection accuracy can be exhibited.

In particular, in the present embodiment, the region R has a square shape including a total of four pixel coordinates (Xp, Yp) of 2 rows and 2 columns arranged adjacently in the X direction and the Y direction. Therefore, the region R includes pixel coordinates (Xp, Yp) adjacent to each other in the X direction, pixel coordinates (Xp, Yp) adjacent to each other in the Y direction, and 45 ° with respect to both directions of the X and Y directions. Pixel coordinates (Xp, Yp) adjacent to each other in the oblique direction of inclination are included. Therefore, even if the pixel shift as shown in FIG. 9 occurs in any of the X direction, the Y direction, and the oblique direction, the influence of the pixel shift can be canceled by the processing similarity map M2. Further, since the pixel shift shown in FIG. 9 is less than one pixel at the maximum, it is sufficient if two adjacent pixel coordinates (Xp, Yp) are included in each of the X direction, the Y direction, and the oblique direction. .. Therefore, it is possible to suppress the inclusion of useless pixel coordinates (Xp, Yp) in the region R, and the noise generated by using the similarity of the useless pixel coordinates (Xp, Yp) for averaging is substantially present. Does not occur. Therefore, better detection accuracy can be exhibited by setting the region R in 2 rows and 2 columns.

As the region R, the above-mentioned two-row, two-column shape is most preferable, but the region R is not limited to this, and two pixel coordinates (Xp) arranged adjacent to each other in at least one direction of the X direction, the Y direction, and the diagonal direction. , Yp) may be included. However, among these, it is preferable to select from a total of nine pixel coordinates (Xp, Yp) of 3 rows and 3 columns arranged adjacently in the X direction and the Y direction. As a result, it is possible to prevent the region R from including unnecessary pixel coordinates (Xp, Yp), and it is possible to obtain a processing similarity with less noise. The region R selected from a total of nine pixel coordinates (Xp, Yp) in 3 rows and 3 columns is not particularly limited, but for example, as shown in FIG. 16, a cross-shaped region R1 and a grid type are used. An eggplant region R2 and the like can be mentioned.

Further, depending on the configuration of the encoder 1, it is conceivable that only the pixel shift in the X direction occurs. In this case, for example, as shown in FIG. 16, the region R3 including two pixel coordinates (Xp, Yp) arranged adjacent to each other in the X direction can be set. Similarly, depending on the configuration of the encoder 1, it is possible that only pixel shift in the Y direction occurs. In this case, for example, as shown in FIG. 16, the region R4 including two pixel coordinates (Xp, Yp) arranged adjacent to each other in the Y direction can be set. Similarly, depending on the configuration of the encoder 1, it is conceivable that only pixel deviation in the oblique direction occurs. In this case, for example, as shown in FIG. 16, the region R5 including two pixel coordinates (Xp, Yp) arranged adjacent to each other in the oblique direction can be set. As a result, it is possible to obtain a processing similarity with less noise.

The template matching method of the present embodiment has been described above. In such a template matching method, as described above, the encoder 1 which is an imaging type encoder is used to determine the similarity of the imaging element 31 with the reference image TA with respect to the captured image G, and the reference image TA and the captured image G are combined. A similarity map M1 is generated in which the similarity of the above is arranged according to the coordinates (Xp, Yp) of the reference image TA in the captured image G, and the processing is obtained based on a plurality of adjacent similarities in the similarity map M1. A processing similarity map M2 in which the similarity is arranged according to the coordinates (Xs, Ys) of the similarity in the similarity map M1 is generated, and the maximum processing similarity is extracted from the processing similarity map M2. According to such a method, the influence of the pixel shift as shown in FIG. 9 can be canceled, and the pixel coordinates of the mark 21 are not the pixel coordinates (Xp, Yp) = (Xn, Yn) but the pixels. It is possible to effectively suppress the erroneous selection of coordinates (Xp, Yp) = (Xz, Yz). Therefore, according to such a method, excellent detection accuracy can be exhibited.

Further, as described above, the template matching method extracts the maximum similarity from a plurality of similarity used to obtain the extracted processing similarity. As a result, it is effective to mistakenly select the pixel coordinates (Xp, Yp) = (Xz, Yz) instead of the pixel coordinates (Xp, Yp) = (Xn, Yn) as the pixel coordinates of the mark 21. Can be suppressed. Therefore, according to such a method, excellent detection accuracy can be exhibited.

Further, as described above, the position of the extracted reference image TA having the similarity, that is, the pixel coordinates (Xp, Yp) is determined as the detection position. As a result, accurate position detection can be performed.

Further, as described above, the processing similarity is an average value of a plurality of similarity. Thereby, the processing similarity can be easily obtained. Further, it is possible to more reliably cancel the influence of the pixel shift as shown in FIG.

Further, as described above, the processing similarity is the total value of a plurality of similarity. Thereby, the processing similarity can be easily obtained. Further, it is possible to more reliably cancel the influence of the pixel shift as shown in FIG.

Further, as described above, in the processing similarity map M2, a plurality of similarities are arranged in a matrix in the X direction and the Y direction which are orthogonal to each other. Then, the processing similarity is obtained by using a plurality of similarities selected from nine similarities adjacent to each other in the X direction and the Y direction arranged so as to form a matrix of 3 rows and 3 columns. As a result, it is possible to obtain a processing similarity with less noise.

Further, as described above, the processing similarity is obtained by using four similarities adjacent to each other in the X direction and the Y direction, which are arranged so as to form a matrix of 2 rows and 2 columns. As a result, it is possible to obtain a processing similarity with less noise.

Although the template matching method of the present invention has been described above based on the preferred embodiment shown in the illustration, the present invention is not limited to this, and the configuration of each part and each step have any configuration having the same function. Can be replaced with one. Further, any other components or steps may be added. Further, the configurations of the two or more embodiments described above may be combined.

Further, the above-mentioned encoder 1 can be applied to both absolute type and incremental type. Further, the encoder 1 can be applied not only to a rotary encoder but also to a linear encoder.

Further, the above-mentioned encoder 1 may be applied to the above-mentioned second encoder that detects the rotation angle around the second axis J2. Further, the encoder 1 described above can be installed in a robot of a type other than the horizontal articulated robot described above, for example, a 6-axis articulated robot, a dual-arm robot, or the like. Further, the encoder 1 may be mounted on an electronic device other than the robot 100, for example, a printer, a projector, or the like. Further, it may be mounted on a moving body such as an encoder 1, a vehicle, an aircraft, or a ship.

1 ... Encoder, 2 ... Scale unit, 20 ... Dot, 21, 21A ... Mark, 3 ... Detection unit, 31 ... Imaging element, 32 ... Imaging optical system, 4 ... Circuit unit, 5 ... Processing unit, 6 ... Storage unit , 100 ... robot, 110 ... base, 120 ... first arm, 130 ... second arm, 140 ... work head, 141 ... spline shaft, 150 ... end effector, G ... captured image, J1 ... first axis, J2 ... 2nd axis, J3 ... 3rd axis, LY ... center line, M1 ... similarity map, M2 ... processing similarity map, P0 ... reference pixel, PEp ... end pixel, PEs ... end pixel, PSp ... start coordinate, PSs ... Start coordinates, Px ... pixels, R, R1, R2, R3, R4, R5 ... regions, RI ... imaging regions, S0 ... reference pixel coordinates, S1 to S7 ... steps, TA ... reference images

Claims (7)

Using an imaging encoder, determine the similarity of the image captured by the image sensor with the reference image.
A similarity map in which the similarity is arranged according to the coordinates of the reference image in the captured image is generated.
A machining similarity map is generated in which the machining similarity obtained based on a plurality of adjacent similarity in the similarity map is arranged according to the coordinates of the similarity in the similarity map.
A template matching method characterized by extracting the maximum machining similarity from the machining similarity map. The template matching method according to claim 1, wherein the maximum similarity is extracted from a plurality of the extracted similarity used to obtain the processing similarity. The template matching method according to claim 2, wherein the position of the reference image having the extracted similarity is determined as a detection position. The template matching method according to any one of claims 1 to 3, wherein the processing similarity is an average value of a plurality of the similarity. The template matching method according to any one of claims 1 to 3, wherein the processing similarity is a total value of a plurality of the similarity. In the processing similarity map, a plurality of the similarities are arranged in a matrix in the X and Y directions orthogonal to each other.
In any one of claims 1 to 5, the processing similarity is obtained by using the four similarities adjacent to each other in the X direction and the Y direction, which are arranged in a matrix of 2 rows and 2 columns. Described template matching method. In the processing similarity map, a plurality of the similarities are arranged in a matrix in the X and Y directions orthogonal to each other.
The processing similarity is obtained using a plurality of the similarity selected from the nine adjacent similarity in the X direction and the Y direction arranged in a matrix of 3 rows and 3 columns. The template matching method according to any one of claims 1 to 5.

Images