JPH05342356A - Picture processor - Google Patents

Abstract

PURPOSE:To surely detect the position of a recognition object with a simple method by setting a window to the position where the existence of the recognition object is estimated and scanning the window picture to detect a one- dimensional density distribution. CONSTITUTION:An arithmetic processor 18 reads out the picture signal stored in a picture memory 13 and sets a window to at least two right and left parts where the existence of the recognition object is estimated. Each window picture is scanned in the horizontal direction and the vertical direction to detect the one-dimensional density distribution. Plural minimum values existing in respective one-dimensional density distributions are obtained, and maximum values before and after arrival at these minimum values and maximum values of inter-picture element differences between these minimum values are defined as likelihood values of respective minimum values. Discrimination values of right and left windows are caluclated in accordance with the width between the minimum value having the maximum value of likelihood values and the minimum value having the second maximum value, and it is judged that the recognition object exists in the window having the discrimination value approximately equal to the diameter of the recognition object.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus used in an automatic assembling apparatus.

[0002]

2. Description of the Related Art An image processing apparatus used in a conventional automatic assembling apparatus or the like picks up an image of an object to be detected (specifically, a groove of a pin having a predetermined diameter into which a hole is inserted or a marking mark). The image signal output from the image capturing unit is subjected to processing such as binarization to detect the position of the detection target,
A position signal is output to the automatic assembling device, which allows the automatic assembling device and the like to operate accurately. Particularly, in the automatic assembly device for a fuel injection pump, when the cam disc is mounted on the connector mounted on the drive shaft, the pin position of the cam disc and the key groove side of the drive shaft must be mounted on the same side. Therefore, it is necessary to reliably detect the pin position of the cam disc.

However, in the conventional image processing, due to the structure of the object to be detected, the detection environment, etc., (1) the contrast of the input image is weak and it is difficult to determine the threshold value when performing the binarization processing, so that good binary Image cannot be obtained, (2) When the metal surface has rough irregularities (surface irregularity), it is difficult to distinguish the marking and the surface irregularity. (3) When there are oil drops or thick oil film on the metal surface However, there are problems such as deterioration of contrast and glare in the pin groove image.

As an image enhancement method for a low-contrast image, there is a method of performing binarization processing after performing spatial filtering processing such as Laplacian calculation. However, noise in the image is emphasized or erroneous recognition occurs. There is a problem.

Further, regarding the determination of an appropriate binarization threshold value, a discriminant analysis binarization method focusing on the density histogram of an image (“Basic of image recognition [I], Ohmsha Co., Ltd.”), and
A processing method in which a threshold value is dynamically changed according to a detection environment like the image processing device disclosed in Japanese Patent No. 154976 is disclosed, but it is difficult to completely discriminate between the disorder of the metal surface and the marking. There is.

[0006]

Therefore, the present invention is to provide an image processing apparatus capable of surely detecting the pin position of a cam disk by a simple method.

[0007]

SUMMARY OF THE INVENTION In the first aspect of the present invention, however, an image signal picked up by the image pickup means is inputted to
In an image processing device for performing image recognition after D / D conversion and storing in a storage device, windows are set at at least two left and right positions where it is estimated that there is a recognition target object on the image captured by the imaging unit. The window setting means and the left and right window images set by the window setting means are scanned in two predetermined directions, and the densities of pixels on the scanning lines are detected to detect the one-dimensional density distribution of each scanning line. A one-dimensional concentration distribution detecting means, a minimum value detecting means for detecting a plurality of minimum values existing in each one-dimensional concentration distribution, and a minimum value from each minimum value to a maximum value before and after this minimum value. The difference between pixels is calculated, and the maximum value of this difference is used as the likelihood value of this minimum value, and the maximum value is calculated from the likelihood value of each minimum value calculated by this likelihood value calculation means. Likelihood And the minimum value having the second likelihood value is detected, the width of these two minimum values is calculated to be the judgment value of this scanning line, and the judgment values of the left and right windows are calculated from the judgment value of this scanning line. The judgment value calculating means compares at least the judgment value of the left window and the judgment value of the right window with the diameter of the recognition object, and confirms that the recognition object is present in the window having the judgment value substantially equal to the diameter of the recognition object. According to the second aspect of the invention, the recognition target position determining means is provided, and after the image signal captured by the image capturing means is input, A / D converted and stored in the storage device, In an image processing apparatus for performing image recognition, window setting means for setting windows at at least two left and right positions where it is estimated that there is a recognition target object on the image captured by the image capturing means,
A one-dimensional density distribution in which the left and right window images set by the window setting means are scanned in predetermined two directions, the densities of pixels on the scanning lines are detected, and the one-dimensional density distribution of each scanning line is detected. Detection means and one of each
A minimum value detecting means for detecting a plurality of local minimum values existing in the three-dimensional density distribution, and a difference between each local pixel from each local minimum value to a local maximum value before and after this local minimum value is obtained, and a maximum value of this difference is calculated. Likelihood value calculating means for setting the likelihood value of this minimum value,
The minimum value having the maximum likelihood value and the second likelihood value is detected from the likelihood values of the respective minimum values calculated by the likelihood value calculation means, and the scanning line is calculated from the density values of these two minimum values. And a judgment value calculation means for calculating the judgment values of the left and right windows from the judgment values of the two scanning lines for scanning each window, and at least the judgment value of the left window and the judgment value of the right window. In comparison, a recognition target position determining means for determining that a recognition target object is present in a window having a low determination value is provided.

[0008]

According to the first aspect of the present invention, therefore, when the recognition target object is a pin inserted into a hole, windows are set at at least two left and right positions where the recognition target object is presumed to exist, and the respective window images are set. The inside is scanned in a predetermined two directions, for example, in the horizontal direction and the vertical direction, and each one-dimensional density distribution is detected. Further, a plurality of local minimum values existing in each one-dimensional density distribution are obtained, and the maximum value of the difference between pixels between the local maximum value before and after reaching the local minimum value and the local minimum value is set to the local minimum value. As the likelihood value of the value, the width of the position of the minimum value having the maximum value and the second value of this likelihood value is obtained, and this is set as the determination value of the scanning line. The judgment value of each window is calculated from the judgment value of this scanning line, and the calculation result is used as the judgment value of each window.
The above problem can be achieved in order to compare the judgment value of each window with the diameter of the recognition target object and to judge that the window having the substantially equal judgment value is the window having the recognition target object.

According to the second aspect of the invention, the determination value for the scanning line is obtained from the density of the minimum value having the maximum and second likelihood values, and the determination value for each window is determined from the determination value for this scanning line. Is calculated, and the calculation result is used as the judgment value of each window. The above problems can be achieved in order to compare the determination values of the respective windows and determine that the window having the lower determination value is the window having the recognition object.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. The automatic assembling apparatus shown in FIG. 1A is a video camera when an assembling target member (for example, a fuel injection pump) 3 fixed to a moving table 2 sent on a rail 1 moves and reaches a predetermined position. 4 is used to photograph an assembly component (for example, a cam disk) 5, the image signal is processed by the image processing device 6 to detect the position of the assembly component 5, and the detection result is output to the robot controller 7. By this, the robot 8 is driven to assemble the assembly parts to the assembly target object 3.

As shown in FIG. 1B, the image processing device 6 includes a clock signal generator 9, an A / D converter 10, a bus buffer 11, a data bus 12, an image memory 13, a column address register 14, and a row. At least the address register 15, the buffers 16 and 17, and the arithmetic processing unit 18, which inputs the video signal output from the video camera 4 based on the clock signal output from the clock signal generation unit 9. is there. This video signal is A
The image memory 13 is converted into a digital signal by the / D converter 10, sent to the data bus 12 via the bus buffer 11, and cascaded via the buffers 16 and 17 by the column address register 14 and the row address register 15. Be stored in.

The image signal stored in the image memory 13 is appropriately read out by the arithmetic processing unit 18 for processing, and the arithmetic processing unit 18 is, for example, a central processing unit (CPU) (not shown), Read-only memory (RO
M), a random access memory (RAM), an input / output port (I / O) and the like, which is known per se, processes the image signal based on a predetermined program, and causes the robot controller 7 to process the image signal. The result of image processing is output.

In the image processing of the embodiment according to the present invention, specifically, in the step of mounting the cam disk 20 shown in FIG. 2 on the connector mounted on the drive shaft (not shown) of the distributed fuel injection pump, the cam disk 20 is mounted. It is used to detect the orientation of 20.

The cam disk 20 has a plunger insertion hole 21 into which one end of a plunger formed on the head is inserted,
A pin 22 that is inserted into the communication hole so as to project to a predetermined position of the plunger insertion hole 21 and that is inserted into a meshing groove formed at one end of the plunger and that extends downward and is attached to a drive shaft (not shown) It is composed of meshing projections 23 and 24 that mesh with the connector and a cam surface 25 that rotates on a roller held by a housing (not shown). Since the cam disk 20 of this embodiment is compatible with four cylinders, the cam surface has four crests 26 and a bottom 27, and the interval between the crest 26 and the bottom 27 is the suction of the plunger. In the process, the process from the bottom 27 to the peak 26 is the plunger pressure feeding process.

An image of the cam disk 20 taken by the video camera 4 is shown in FIG. 3, and the peaks 26, the bottom 27, the engaging projections 23 and 24, and the pin 22 of the cam surface 25 are projected with high brightness. but, this image is for example vertical: 2 ^m pixels (y-coordinate downward from above), the horizontal direction: is represented by 2-dimensional image composed of 2 ⁿ pixels (chi coordinates from left to right direction), The gradation value (density value) of each pixel is stored in the image memory 13 as an image signal (image data).

Image processing executed in the image processing apparatus 6 will be described below with reference to the flowchart shown in FIG.

In the image processing program started from step 100, the image data of each pixel stored in the image memory 13 is input in step 110.

At step 120, as shown in FIG. 5, the positions of the windows 30 to 33 are fixed, a group having a certain gradation value or more is detected in each of the windows 30 to 33, and the cam is calculated from the center of gravity of the group. Find the center of the disk 20. Specifically, a group of gradation values of a predetermined value or more (bottom portion 27 of the cam surface 25) having an area of a predetermined value or more is detected, and in this case, four windows 30 to 33 are set. Next, the center of gravity G ₃₀ , G ₃₁ , of each of the set windows _{30 to} 33 is set.
G ₃₂ , G ₃₃ are obtained, and further G ₃₀ and G ₃₂ , G ₃₁ and G ₃₃
The straight lines L ₁ and L ₂ passing through are obtained. The intersection of these straight lines L ₁ and L ₂ is the center C _CT of the cam disk 20.

In step 130, as shown in FIG. 6, the pin position determination windows are set at two places on the left and right. Specifically, from the center C _CT of the cam disk 20,
The circular windows 40 and 41 for pin position determination are set at positions separated by a predetermined position in the left-right direction set at a predetermined angle from the inclination of the straight line L ₁ or L ₂ . The circular window 40 is referred to as the left window 40, and the circular window 41 is referred to as the right window 41.

The pin position detection processing according to the first embodiment of the present invention which is executed from step 200 includes two depressions (valleys of the spatial distribution) spaced apart by a certain distance when the pin 22 exists in the circular window. ) Exists, the pin position is detected. This flowchart is shown in FIGS. 7 to 13 and will be described below with reference to this flowchart.

First, in step 210 of FIG.
4, a scanning line 50 for scanning the left window 40 horizontally (in the χ direction), a scanning line 51 for scanning the left window 40 in the vertical direction (y direction), and a scanning line 53 for scanning the right window 41 in the horizontal direction (χ direction). , And the address of the scan line 54 that scans the right window 41 vertically (y direction) is calculated. Specifically, z pixels are set in the vertical and horizontal directions from the center of each window 40, 41.

In step 220, the scan line 5
0 (z pixels to the left and right from the center of the left window 40)
The gradation level of each pixel above is read to detect the spatial distribution G _LX (N) in the left window χ direction shown in FIG. 15. Similarly, in step 230, from the scanning line 51 (z pixels vertically from the center of the left window 40) to FIG.
The left window y direction spatial distribution G _LY (N) shown in 5 is detected.

In step 240, the scan line 5
3 (z pixels to the left and right from the center of the right window 41)
The gradation degree of each pixel above is read and the right window χ direction spatial distribution G _RX (N) shown in FIG. 16 is detected. Similarly, in step 250, from the scanning line 54 (z pixels above and below the center of the right window 41), as shown in FIG.
The right window y direction spatial distribution G _RY (N) shown in 6 is detected. The spatial distributions G _LX (N), G _LY (N), G
_RX (N) and G _RY (N) indicate so-called one-dimensional density distribution of each scanning line.

Next, at step 260 shown in FIG. 8, the flag F _{1 for} designating the spatial distribution for which the determination process is to be performed.
The return to the initial value "0", it is determined whether the flag F ₁ is "0" or in step 270. In this determination, if the flag F ₁ is “0” (Y), the process proceeds to step 280 and the left window χ direction spatial distribution G _LX (N) is set as the calculation spatial distribution G (N), and it is not “0”. (N)
Goes to step 290 to determine whether the flag F ₁ is "1". If it is "1" in this judgment, the spatial distribution for calculation G (N) is set to the left window y direction spatial distribution _GLY (N) in step 300, and if it is not "1", it is moved to step 310. Then, it is determined whether the flag F ₁ is "2".

If the flag F ₁ is "2" in step 310, the flow advances to step 320 to set the right window χ direction spatial distribution G _RX (N) to the calculation spatial distribution G (N), and "2" is set. Otherwise, the calculation spatial distribution G (N)
The right window y direction spatial distribution G _RY (N) is set to.

As a result, after the calculation spatial distribution G (N) is set, all the pixel discrimination flags F _P (N) are cleared (“0”) in step 340, and
In step 350, the initial value "1" is set in the minimum value designation variable n.

After that, the routine proceeds to step 360, where an initial value "1" is set to the variable N indicating the pixel position, and the density comparison value S
Is set to a predetermined value α (specifically 256), and a variable a indicating the minimum value “1” is set to “0”.

In step 370, the flag F _P
It is determined whether (N) is "0". This determination is to determine whether or not the pixel designated by N has already been used in the minimum value search. In the case of "0", the pixel has not been used yet and therefore the process proceeds to step 380. The density comparison value S and the gradation value G (N) of the pixel designated by N are compared. The density comparison value S is a predetermined value α at the first time, and the gradation value of G (N) set in step 390 described below is substituted for the second time and thereafter.

Since N = 1 at the first time, G
(1) is compared with S for which the predetermined value α is set, and G (1)
If is smaller, the process proceeds to step 390 and the comparison value S
Is set to G (1), the variable a is set to "1", the next N (in this case, "2") is set in step 400, and this N reaches a predetermined value β in step 410. Up to step 370. Note that the predetermined value β is 2z + 1 pixels including the center because β pixels are set to the left, right, top, and bottom from the center of the window in this embodiment (β = 2z +).
1).

If the flag F _P other than "0" is already set in this pixel in the determination in step 370, the processing in steps 380 and 390 is avoided and the processing proceeds to step 400. When the gradation value G (N) of the pixel is larger than the density comparison value S in the determination of 380, the processing of step 390 is avoided and the density comparison value S is held.

This process is repeatedly executed up to the predetermined value β, and the pixel value S and the pixel position a having a low gradation value are constantly updated, whereby the minimum value designation variable n is "1" (first time), The minimum value q ₁ (a, S) having the lowest gradation value among the unprocessed pixels is obtained.

Next, proceeding to step 420 shown in FIG. 10, a is set to the pixel position N, and the initial value "0" is set to the left side difference sl. The minimum value q _n set by this
(A, S) is intended for calculating the maximum value sl _n, step 430 or 600 follows sr _n the left and right difference of.

In step 430, it is judged whether or not the value of the flag F _P (N) (F _P (a)) of the pixel to be processed is "0". If it is a value other than "0", steps 440 to 490 are bypassed and step 500 is proceeded to.

In step 440, G (N-1)
And G (N) are compared. This is between each pixel from q _n (N, G (N)) set as the minimum value to the maximum value P _{L1 in} the left direction (before the minimum value) of the spatial distribution G (N) for calculation shown in FIG. To detect the difference between
If (N-1) is larger than G (N), it means that the spatial distribution has a characteristic of rising to the left, and therefore the process proceeds to step 450, where G (N-1) and G ( N) and the flag F for this N
Substitute n for _P (N). Specifically, the minimum value q ₁ (a, G (a)) a detected first in FIG.
Since it is 5, the minimum value q ₁ is represented by (25, G (25)), and for this minimum value q ₁ , G (24) ≧ G (25) in the determination of step 440. In step 450, G (24) −G
(25) is substituted, and the flag F _P (25) is set to "1" (since n = 1 at the first time).

In step 460, sl _n is compared with k set in step 450,
If k is large, this k is set to s in step 470.
Substitute for l _n . If k is small, step 4
The difference sl _n between pixels is retained by avoiding 70. After that, in step 480, N-1 is substituted for N to set a pixel shifted by one to the left in the drawing as a processing reference, and in step 490, the pixel position N is "1".
In the case of the above, the process returns to step 430 and the same process is performed. Specifically, at the initial stage, sl ₁ is determined to be 420 in step 420.
Since sl ₁ is smaller than k because it is “0” in the setting of, the value of k (G (2
4) -G (25)) is set to sl ₁ and step 480
Then, N is set to 24 (25-1).

The above processing is executed when it is determined in step 430 and step 440 that the flag F _P is a value other than "0" (when the pixel is already processed).
Alternatively, when G (N-1) becomes smaller than G (N) (when it passes the maximum value), the process of steps 450 to 490 is avoided and the process proceeds to step 500, where the flag F of the last pixel position is set. N is set in _P (N).

As a result, as shown in FIG. 17, the difference between each pixel from the minimum value q ₁ to the maximum value P _L1 in the left direction is detected, and the maximum difference is left as sl ₁ . In the case of n = 2, sl ₂ is obtained for the second minimum value q ₂ , and the flag F _P (N) of each pixel is set to “2”. Hereinafter, similar processing is performed until n = ε.

Further, in order to obtain the difference sr ₁ of each pixel to the right of the minimum value q ₁ (after the minimum value), step 5
In step 10, the flag F _P (a) having the minimum value is set to “0” again, and in step 520, the pixel position N is set to a again and the rightward difference sr _n is set to the initial value “0”. To do.

In step 530, it is judged whether or not the value of the flag F _P (N) (F _P (a)) is “0” for the pixel to be processed. If the value is other than "0", steps 540 to 590 are bypassed and step 600 is proceeded to.

In step 540, G (N + 1)
And G (N) are compared. This is between each pixel from q _n (N, G (N)) set as the minimum value to the maximum value P _{R1 in} the right direction (after the minimum value) of the spatial distribution G (N) for calculation shown in FIG. To detect the difference between
If (N + 1) is larger than G (N), it means that the spatial distribution has a characteristic of rising to the right. Therefore, the process proceeds to step 550, where k (G + 1) and G (N) are combined. Substitute the difference and flag F for this N
Substitute n for _P (N). Specifically, the minimum value q ₁ (a, G (a)) a detected first in FIG.
Since it is 5, the minimum value q ₁ is represented by (25, G (25)), and for this minimum value q ₁ , G (26) ≧ G (25) in the determination of step 540. In step 550, G (26) −G
(25) is substituted, and the flag F _P (25) is set to "1" (since n = 1 at the first time).

In step 560, k set in step 550 is compared with sr _n ,
If k is large, this k is set to s in step 570.
It is assigned to r _n. If k is small, step 5
The difference sr _n between pixels is retained by avoiding 70. After that, in step 580, N + 1 is substituted for N to set a pixel shifted by one to the right in the drawing as a processing reference. In step 590, when the pixel position N is γ or less, the process returns to step 530. Then, the same processing is performed. Specifically, in the initial stage, the determination in step 560, in order sr ₁ is "0" in the setting of step 520, the value of k in step 570 in order sr ₁ is smaller than k (G ( 26) -G
(25)) is set to sr ₁ and N is set to 26 in step 580 (25 + 1).

In the above processing, if it is determined in step 530 and step 540 that the flag F _P has a value other than "0" (if the pixel has already been processed),
Alternatively, when G (N + 1) becomes smaller than G (N) (when it passes the maximum value), the process of steps 550 to 590 is avoided and the process proceeds to step 600, where the flag F _P ( N is set in N).

As a result, as shown in FIG. 17, the difference between each pixel from the minimum value q ₁ to the maximum value P _R1 in the left direction is detected, and the maximum difference is left as sr ₁ . In the case of n = 2, sr ₂ is obtained for the second minimum value q ₂ and “2” is set to the flag F _P (N) of each pixel. Hereinafter, similar processing is performed until n = ε.

At step 610, the value a indicating the position of the minimum value q _n is substituted into the pixel position array P (n). Specifically, the position a of the minimum value q ₁ detected first time
(25) is substituted into P (1).

Next, in steps 620 to 670, the likelihood value of the minimum value q _n is set. Step 620
And in step 630, if one of the left side difference sl _n and the right side difference sr _n is “0”, step 6
In step 40, "0" is set in the likelihood value array C (n). If the left-side difference sl _n and the right-side difference sr _n are not both “0” in steps 620 and 630, the magnitudes of the left-side difference sl _n and the right-side difference sr _n are compared in step 650 to determine that the left-side difference sl _n is If it is large, the process proceeds to step 660, substituting sl _n for C (n), and if the right difference sr _n is large, C (n)
Sr _n is substituted into

As a result, the likelihood value array C (n) and C (1) are set at the first time, and in step 680, n + 1 is set to the minimum value designating variable n to search for the next minimum value.
Is set, and it is determined in step 690 whether or not this minimum value designating variable n has reached a predetermined value ε. If it has not reached the predetermined value ε, step 36
By returning to 0, the processes of steps 360 to 680 described above are performed to search for the second minimum value q ₂ and set the likelihood value. The predetermined value ε is normally set to 5, and five minimum values q _{1 to} q ₅ are detected.

As a result, the likelihood value array C (n) (C
(1) to C (ε)) and pixel position array P (n) (P
(1) to P (ε)) are set.

Next, the judgment value of each spatial distribution is set according to the flow chart shown in FIG. Step 71
In step 0, the likelihood value array C (n) is rearranged in the descending order of values, and in step 720, the pixel position array P (n) corresponding to the largest (maximum) likelihood value C (n). ) Is substituted for P ₁ and the pixel position array P corresponding to the second largest likelihood value C (n) is obtained in step 730.
Substitute (n) into P ₂ . This results in step 74
At 0, two local minimum values q _J1 (P ₁ , G (P ₁ )) and local minimum value q _J2 (P ₂ , G (P ₂ )) are set.

Thereafter, in step 750, it is judged whether the flag F ₁ is "0". If it is "0",
In step 760, the left wind χ direction spatial distribution judgment value D _LX is set by the difference (| P ₁ −P ₂ |) between the pixel positions P ₁ and P ₂ of the two local minimum values q ₁ and q ₂ , and step 770. At ₁ , the flag F ₁ is set to "1". As a result, the determination in step 860 returns to step 270, and the determination in step 290 changes the calculation spatial distribution G (N) into the left window y-direction spatial distribution G.
_LY (N) is set and the same processing as above is performed.
Accordingly, next, in the determination of step 750, since F ₁ is set to “1”, the process proceeds to step 780, and the process proceeds to step 790 in this determination.

At this step 790, the difference between the pixel positions P _{1 and} P ₂ is determined by the left window y direction spatial distribution judgment value D.
Is set to _LY, sets "2" to F ₁ in step 800. As a result, the process returns from step 860 to step 270 again, and then passes through steps 750 and 780 to reach the determination of step 810. F in this judgment
₁ = a if 2, sets the difference between the pixel positions P _1, P ₂ in the right window χ direction spatial distribution determining value D _RX proceeds to step 820, and sets "3" to F ₁ in step 830 .. If F ₁ = 3, the process proceeds to step 840 in the determination of step 810, and this step 84
At 0, the difference between the pixel positions P _{1 and} P ₂ is set to the right window y direction spatial distribution determination value D _RY , and at step 850, F
Set “4” to ₁ . This results in step 860
In the determination of, the process proceeds to the next step shown in FIG.

In step 870 shown in FIG. 13, it is determined whether or not the left window χ direction spatial distribution determination value D _LX is within a predetermined range (λ ₁ <D _LX <λ ₂ ).
In this determination, if D _LX is not within the predetermined range, the process proceeds to step 880, and the determination result L _X is set to “0”. If it is within the predetermined range, the process proceeds to step 890 and the determination result L _X is set. "1" is set. This λ ₁
And λ ₂ are set with an allowable range with respect to the diameter λ of the pin 22.

Next, in step 900, the left window y-direction spatial distribution judgment value D _LY is judged, and if it is outside the predetermined range, the judgment result L is judged in step 910.
"0" is set to _Y, and if it is within the predetermined range, the determination result L _Y is set to "1" in step 920.

Similarly, in steps 930 to 980, the determination results R _X and R _Y are set by determining the right window χ direction spatial distribution determination value D _RX and the right window y direction spatial distribution determination value D _RY. Is.

As a result, in step 990, the determination results L _X and L _Y of the scanning lines in the left window are "1" and at least one or both of the determination results R _X and R _Y of the scanning lines in the right window are "1". If it is "0", it can be determined that the position of the pin 22 is in the left window, so "1" is set to the pin position output t (0).

In step 1000, at least one or both of the left window scanning line determination results L _X and L _Y are "0", and the right window scanning line determination results R _X and R _Y are both "0". If it is "1", it can be determined that the position of the pin 22 is in the right window, so "2" is set to the pin position output t (0).

Further, in step 1010, if the determination result of each scanning line is other than the above, the pin position output t (0) is set to "0" as unrecognizable, and step 1
The routine returns from 020 to the main control routine.

Then, in step 1060 shown in FIG. 4, the pin position output t (0) is set to the robot controller 7
To the end of the process in step 1070.

Therefore, in the above-described pin position detecting method, the minimum values q ₁ and q ₂ are detected as shown in FIG. 17, the width is compared with the pin width, and the width substantially equal to the pin width is determined. Since the window in which the spatial distribution has exists is determined as the side having the recognition target (in this case, the pin), the pixels of the entire image are not targeted, so that the processing data can be reduced and the processing speed can be increased. Is what you can do. In addition, according to this method, it is possible to detect even those which are determined to be unrecognizable by the conventional method (method by binarization processing or the like).

Further, in the pin position detecting method according to the second aspect of the invention, since the circular window in which the pin 22 exists has two depressions (valleys in the spatial distribution), the gradation of the circular window having the depressions becomes low. The pin position is detected based on
This will be described below according to this flowchart. The same steps as those in the first aspect of the invention are omitted, and the same numbers are assigned to omit the description.

In the case of the above-mentioned first invention, the step 7
At 40, two local minimum values q _J1 (P ₁ , G (P ₁ ))
, And q _J2 (P ₂ , G (P ₂ )) are set, steps 760, 790, 820, depending on the value of the flag F ₁ .
At 840, the judgment values D _LX , D _LY , D _RX , and D _RY are set. However, in the second invention, step 76
At 0 ′, the minimum value of the χ-direction spatial distribution q _J1 (P ₁ , G
(P ₁ )) and q _J2 (P ₂ , G (P ₂ )) from the gradation value G
The sum of (P ₁ ) and G (P ₂ ) is calculated and set as the left window χ direction spatial distribution judgment value J _LX (J _LX ← G (P ₁ ).
+ G (P ₂ )), similarly steps 790 ′, 820 ′,
840 ′, the left window y-direction spatial distribution determination value J
_LY , the right wind χ direction spatial distribution judgment value J _RX , and the right window y direction spatial distribution judgment value J _RY are set.

Next, at step 2010, the left window judgment value L _J is set by calculating the sum of J _LX and J _LY (L _J ←
J _LX + J _LY ), and in step 2020, the right window determination value R _J is similarly obtained (R _J ← J _RX + J _RY ).

In step 2030, it is determined whether or not the values of L _J and R _J are equal, and they are equal (L _J =
In the case of R _J ), the process proceeds to step 2040 and it is determined that the pin position output t (1) cannot be recognized and “0” is set. Also,
In the determination of step 2030, they are not equal (L _J
≠ R _J ), go to step 2050 and proceed to L _J and R
_When the size of _J is compared and L _J is large (L _J > R _J ).
To step 2060, the pin position output t (1) is set to "1" assuming that the pin position is on the left window side, and if R _J is large (L _J <R _J ), step 2
At 070, "2" is set to the pin position output t (1).

The method of the second invention can be executed by a program simpler than that of the first invention, and as a result of the experiment, there is almost no false detection or undecidable ratio. The rate can be achieved.

Further, the reliability can be further improved by executing the conventional method at the same time as the method according to the first and second inventions.

[0065]

As described above, according to the first invention and the second invention, windows are set at at least two image positions where the recognition object is estimated to exist, and the windows are set in the horizontal direction. And scan the scan line in the vertical direction to 1
Since the three-dimensional density distribution (spatial distribution) is detected and this spatial distribution is the processing target, the data processing amount can be greatly reduced, so that the processing speed can be increased and the recognition target object The feature, in other words, the width of the pin in the first aspect of the invention and the low gradation value in the second aspect of the invention are used as the determination criteria, so that the recognition target can be accurately detected.

[Brief description of drawings]

FIG. 1A is a schematic explanatory view of an automatic assembly apparatus in which an image processing apparatus is incorporated, and FIG. 1B is a schematic explanatory view showing a configuration of the image processing apparatus.

FIG. 2A is a front view showing an example of a cam disk as an object to be recognized, and FIG. 2B is a plan view.

FIG. 3 is an explanatory diagram showing an image of a cam disc taken by a video camera.

FIG. 4 is a flowchart of a main program for image processing according to an embodiment of the present invention.

FIG. 5 is an explanatory diagram showing a setting situation of a window for detecting the center of the cam disc.

FIG. 6 is an explanatory diagram showing a setting situation of left and right circle windows.

FIG. 7 is a flowchart diagram of a spatial distribution detecting portion of the flowchart of pin position detection according to the embodiment of the first invention.

FIG. 8 is a flowchart of a part for replacing each spatial distribution with a calculation spatial distribution.

FIG. 9 is a flowchart of a local minimum value detecting portion.

FIG. 10 is a flowchart of a portion for obtaining a maximum value of a difference between pixels before and after each minimum value.

FIG. 11 is a flowchart of a portion for obtaining an array for obtaining the likelihood value of each minimum value.

FIG. 12 is a flowchart of a portion for obtaining a determination value for each scanning line.

FIG. 13 is a flowchart of a portion that determines a pin position.

FIG. 14 is an explanatory diagram when the scanning lines of the left and right circular windows are set.

FIG. 15: Left wind χ direction spatial distribution and left window y
It is a characteristic diagram of directional space distribution.

FIG. 16: Right wind χ direction spatial distribution and right window y
It is a characteristic diagram of directional space distribution.

FIG. 17 is an explanatory diagram showing a spatial distribution for calculation and each minimum value.

FIG. 18 is a partial flowchart showing the features of pin position detection of the embodiment according to the second invention.

[Explanation of symbols]

 4 Video camera 8 Robot 20 Cam disk 22 pin

Claims (2)

[Claims] 1. An image processing apparatus for performing image recognition after inputting an image signal captured by an image capturing unit, performing A / D conversion, storing in a storage device, and an image on an image captured by the image capturing unit. A window setting means for setting windows at at least two left and right positions where it is presumed that there is a recognition object, and left and right window images set by the window setting means are scanned in predetermined two directions and hit the scanning line. A one-dimensional density distribution detecting means for detecting a density of a pixel to detect a one-dimensional density distribution of each scanning line; and a minimum value detecting means for detecting a plurality of minimum values existing in each one-dimensional density distribution, Likelihood value calculation means for obtaining the difference between each pixel from each of these local minimum values to the local maximum values before and after this local minimum value, and using the maximum value of this difference as the likelihood value of this local minimum value; The minimum value having the maximum likelihood value and the second likelihood value is detected from the likelihood value of each minimum value calculated by the calculating means, and the width of these two minimum values is calculated to calculate the width of this scanning line. The judgment value is used as a judgment value, and the judgment value calculating means for calculating the judgment values of the left and right windows from the judgment value of this scanning line is compared with the judgment value of at least the judgment value of the left window and the judgment value of the right window and the diameter of the recognition target to recognize An image processing apparatus, comprising: a recognition target position determination unit that determines that a recognition target object is present in a window having a determination value substantially equal to the diameter of the target object. 2. An image processing apparatus for performing image recognition after inputting an image signal photographed by the image pickup means, A / D converting it, and storing it in a storage device. A window setting means for setting windows at at least two left and right positions where it is presumed that there is a recognition object, and left and right window images set by the window setting means are scanned in predetermined two directions and hit the scanning line. A one-dimensional density distribution detecting means for detecting a density of a pixel to detect a one-dimensional density distribution of each scanning line; and a minimum value detecting means for detecting a plurality of minimum values existing in each one-dimensional density distribution, Likelihood value calculation means for obtaining the difference between each pixel from each of these local minimum values to the local maximum values before and after this local minimum value, and using the maximum value of this difference as the likelihood value of this local minimum value; The minimum value having the maximum likelihood value and the second likelihood value is detected from the likelihood value of each minimum value calculated by the calculating means, and the determination value of the scanning line is obtained from the density value of these two minimum values. Comparing at least the judgment value of the left window and the judgment value of the right window with the judgment value calculating means for calculating the judgment values of the left and right windows from the judgment values of the two scanning lines for scanning each window. An image processing apparatus, comprising: a recognition target position determination unit that determines that a recognition target object is present in a window having a low determination value.