JPH0137789B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention is applicable to an automatic uranium pellet appearance inspection device, etc., in which a cylindrical object such as a uranium pellet is rotated to image the cylindrical surface.
The present invention relates to a cylindrical imaging pattern boundary detection method that detects the boundary (pattern edge) between the object and the background in order to cut out the cylindrical pattern part of the object from the imaging screen.

[Prior art]

FIG. 1 is a perspective view showing an example of the configuration of an apparatus for capturing an image of a cylindrical surface pattern of a cylindrical object such as a uranium pellet, for example in an automatic uranium pellet appearance inspection apparatus. In Figure 1, a cylinder (cylinder)
A sample of a shaped object, such as a uranium pellet 1, is placed on a rotating roller 2, rotated in the circumferential direction, and imaged by a linear image sensor 4 through an imaging lens 3. 5 is a light source for illumination. Here, the linear image sensor 4 is installed facing the rotation axis direction of the pellet 1, and scans the rotation axis direction (X direction) by internal scanning, and also scans the circumferential direction (Y direction) by rotation of the pellet 1. Obtain a two-dimensional image. At this time, by making the field of view of the linear image sensor 4 wider than the length of the pellet, it is possible to obtain an image of the entire cylindrical surface of the pellet while the pellet 1 is rotating. In order to obtain a pellet image by cutting out the target cylindrical surface pattern part of the pellet 1 from the two-dimensional image obtained in this way, it is necessary to
It is necessary to detect the boundary between the image and the background (pattern edge).

Note that FIG. 2 is a top view illustrating a case where pellets 1 are placed in a row on the rotating roller 2 of FIG. The pellets 1 are sequentially imaged while moving in the direction of the rotation axis on the row of pellets shown in FIG. In this case, it is necessary to detect the boundaries between the pellet to be imaged and the pellets before and after it and cut out only the image of the target pellet.

As a method for detecting boundaries (pattern edges) of such a cylindrical surface imaged pattern, if the cylindrical surface pattern part of the target pellet is bright and the background is dark in the captured image, the image signal is binarized using an appropriate threshold value. The most common method is

However, with such a conventional method, if there is a defect such as a chip or a crack on the cylindrical surface of the object, for example, the uranium pellet 1, a correct boundary (edge) may not be detected. FIG. 3 is a perspective view illustrating the external appearance when the pellet 1 has a chip 11, for example, and FIG. 4 is a perspective view illustrating the appearance when the pellet 1 has a crack 12. FIG. 5 shows, for example, an illumination light source 5 in which the normal surface of the pellet 1 shown in FIG. 3 is detected brightly and the chipped portion 11 and the background are detected darkly.
This is a plan view illustrating a binary image of the cylindrical surface imaging pattern of a chipped pellet when the illumination method according to (Fig. 1) is used. EG1 is the left edge position of the pellet (in the figure), EG2 is the right edge position, and W is the pellet length. Note that the X direction is the rotation axis direction and Y
The direction indicates the circumferential direction (Fig. 1). 6 and 7 are waveform diagrams illustrating corresponding image signals on each of the X-direction scanning lines S _A and S _B in FIG. 5. FIG. That is, the image signal from the linear image sensor 4 (FIG. 1) on the X-direction scanning line S _A in FIG. 5 has a waveform like the image signal 6 in FIG. The brightness changes greatly at the positions EG1 and EG2, and the "1" and "0" inversion positions of the binary signal binarized with the threshold value V _Th are the correct left and right edge positions EG ₁ and EG _2. Therefore, correct binarized image data 7 can be obtained. However, the image signal on the X-direction scanning line S _B in FIG. 5 shows a waveform like image signal 8 in FIG. Therefore, the "1" and "0" inversion positions of the binary signal binarized using the threshold value V _Th deviate from the true left edge position EG ₁ and therefore do not represent the correct left edge position EG ₁ and are incorrect. 2
Value image data 9 is obtained.

In the conventional circumferential surface imaging pattern boundary detection method that simply determines the boundary from the binary signal obtained by converting the image signal into a binary image signal, defective parts such as chips and cracks in the target cylindrical surface pattern may be mistaken as boundaries. There was a drawback that it was not possible to correctly detect boundaries.

[Purpose of the invention]

An object of the present invention is to solve the above-mentioned drawbacks, and to detect the correct boundary (pattern edge) of the cylindrical surface pattern from within the two-dimensional imaging screen when rotating a cylindrical object such as a uranium pet and imaging the cylindrical surface. ) A method for detecting boundaries of cylindrical imaging patterns is provided.

[Summary of the invention]

The present invention solves the problem of detecting errors in the conventional cylindrical surface imaging pattern boundary detection method when there is a defect on the target cylindrical surface such as a uranium pellet.
Focusing on the fact that it is difficult to distinguish true boundaries (edges) from image data scanned in each direction or in a narrow range, we analyzed at least the circumferential direction to ensure that true boundary (edge) information is included. A wide range of binary image data covering a maximum of one rotation from the allowable defect size is integrated in the circumferential direction, and the multi-valued image signal obtained by this integration is again binarized, and this binarized signal is This is a cylindrical surface imaging pattern boundary detection method in which the inverted position of ``1'' and ``0'' is determined to be a boundary (edge) of a cylindrical surface pattern.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 8 is a circuit diagram showing an embodiment of the cylindrical surface imaging pattern detection method according to the present invention. 8th
To explain the structure and operation of the figure, a binary image obtained by binarizing an analog image signal of a cylindrical object, such as a uranium pellet 1, imaged by the linear image sensor 4 in FIG. 1 using a first threshold value. data 2
1 is a VIDEO signal in which the normal surface of the pellet 1 is "1" and the background and defective portions are "0", as illustrated in FIGS. 6 and 7, for example. The address of the memory 23 is selected by an address signal Ax24 synchronized with internal scanning (X-direction scanning) of the linear image sensor 4. Note that the contents of the memory 23 are cleared prior to starting the integration of the binary image signal 21, but portions related to this operation are omitted in FIG. During the integration of the binary image signal 21, the addresses of the memory 23 are sequentially selected by the address signal 24 according to the internal scanning (X direction scanning) of the linear image sensor 4, and the memory read data is latched for each address. Once latched in the circuit 27, this memory read data and the binary image data 21 are added, and the addition result is written to the same location in the memory 23. As a result, the contents of the memory 23 are incremented by 1 at an address where the binary image data 21 is "1", and integral data in the circumferential direction (Y direction) is obtained. When re-binarizing the multivalued image data obtained by integrating this circumferential direction, the second threshold value Th32 from the CPU is applied.
is latched in the latch circuit 28, and the comparator 30 compares this threshold value 32 with the memory read data of the memory 23. At this time, the multiplexer 29 operates to send the power of the latch circuit 28 to the comparator 30. Comparator 3 obtained by this
0 binary signal output to 1”, “0” inversion detection circuit 31
The inverted position of this binary signal is detected, and this inverted position is determined to be the boundary (edge) of the pellet 1. The X coordinate value of this inverted position is determined by the latch circuit 25.
Latch on and read to CPU. Note that the second threshold 32 when binarizing the integral data again is 2
Prior to the value conversion operation, the peak value of the integral data is
It is read by the CPU and calculated based on this, so
A latch circuit 26 is provided for peak value detection and peak value reading. When detecting the peak value, the latch circuit 26 is
The latch circuit 26 is first cleared, and then the data in the memory 23 is sequentially read out and compared with the contents of the latch circuit 26. If the read data is large, the latch circuit 26 is cleared. By updating the contents of , the peak value is finally made to remain in the latch circuit 26.

FIG. 9 shows that when there is only one pellet 1 within the field of view of the linear image sensor 4 as shown in FIGS. This is a waveform diagram illustrating multivalued image data obtained by integrating in the (Y direction). ) EG ₁ ,
This shows that EG ₂ can be detected correctly.

In addition, FIG. 10 shows the integrated multivalued image data obtained in FIG. 8 because when the pellets 1 are arranged in a row as shown in FIG. FIG. In this case, in the binary image data 21, the entire circumference of the edge portion of the target pellet is not necessarily at a dark level due to the influence of reflected light from adjacent pellets. Therefore, as shown in FIG. 10, the level of the edge portion of the target pellet in the integrated multivalued image data may exhibit levels such as V _b1 and V _b2 . For this reason, if the threshold value is fixed, it becomes impossible to binarize the edge portion of the pellet, making it impossible to detect the boundary. In order to prevent such inconvenience, the normal plane level of the integrated multilevel image data
Find V _S and the darkest levels V _b1 and V _b2 of the left and right edges,
The threshold values T _h1 and T _h2 are adjusted for each pellet to an appropriate level between the darkest levels V _b1 and V _b2 of the left and right edges, respectively, and the normal surface level V _S.
Calculate and set using CPU. In this way, the correct left and right edge portions EG ₁ and EG ₂ of the target pellet can be detected as shown in FIG.

Note that the above binary image data 21 to be integrated may be appropriately thinned out for X-direction scanning lines and integrated in the circumferential direction (Y-direction), which is effective in reducing the capacity of the memory 23. It is.

As described above, according to this embodiment, binary image data of a cylindrical object such as a uranium pellet is integrated in the circumferential direction, and this wide range of image information is used to determine the boundary using an appropriate second threshold. (edges), it is possible to accurately detect the boundaries of the target cylindrical surface pattern even when the cylindrical surface has defects such as local chips or cracks or when pellets are arranged in rows.

〔Effect of the invention〕

As described above, according to the present invention, when rotating a cylindrical object in the circumferential direction and imaging the cylindrical surface, it is possible to prevent misjudgment due to the influence of locally existing defective parts, and The boundary of the correct cylindrical surface pattern can be stably detected from the imaged screen, and can be widely used, for example, in detecting pellet patterns and edges in automatic uranium pellet appearance inspection equipment.

[Brief explanation of drawings]

Fig. 1 is a perspective view showing an example of the configuration of a cylindrical surface pattern imaging device for a cylindrical object/pellet, and Fig.
Figure 3 is a top view showing the row of cylindrical objects/pellets on the roller; Figure 3 is a perspective view of the chipped pellet;
The figure is a perspective view of a pellet with cracks, and Figure 5 is a perspective view of a pellet with cracks.
Figure 3 is a cylindrical surface binary image pattern diagram, Figure 6 is an image signal waveform diagram on scanning line S _A in Figure 5,
FIG. 7 is a waveform diagram of an image signal on scanning line S _B , FIG. 8 is a circuit configuration diagram showing an embodiment of the present invention, and FIG.
Figure 8 is a waveform diagram of multilevel image data obtained by integrating the binary image signal in Figure 5, and Figure 10 is a waveform diagram of multilevel image data obtained by integrating the binary image signal of the pellet row in Figure 8. FIG. 3 is a waveform diagram of multivalued image data. 1... Cylindrical object/pellet, 2... Roller,
4... Linear image sensor, 21... Binary image data, 22... Adder, 23... Memory, 2
5, 26, 27, 28...Latch circuit, 30...
Comparator, 31... Reversal detection circuit.

Claims (1)

[Claims] 1. A cylindrical object is rotated in the circumferential direction and its outer cylindrical surface is imaged by an imager, and the boundary of the cylindrical surface pattern of the cylindrical object is detected from the imaging screen within the field of view of the imager. When detecting, the analog image signal output from the imager is binarized using a first threshold, and this two
The digitized signal is integrated in the circumferential direction, the multilevel image signal after this integration is binarized again using a second threshold, and the "1" and "0" inverted parts of this binarized signal are set as the boundary position. A cylindrical surface imaging pattern boundary detection method characterized by determining. 2. Detecting the cylindrical surface signal level and the darkest level within the cylindrical surface boundary region in the multivalued image signal after the integration, and setting the second threshold within the range sandwiched between these two levels. A cylindrical surface imaging pattern boundary detection method according to claim 1, characterized in that: