Image processing method and image processing apparatus

Abstract

To provide a technology to reduce a calculation amount in image processing of detecting a specific object from time-series image data.SOLUTION: A image processing method for detecting a position of a detection target from a captured image periodically acquired, includes the steps of: a) analyzing a region to be analyzed in the captured image to determine a detection error and the position of the detection target (S103); b) changing the region to be analyzed on the basis of magnitude of the detection error (S104-S107); and repeating a) and b) a plurality of times. Thereby, the region to be analyzed can be reduced while keeping the detection error within a predetermined range. Therefore, a calculation amount can be reduced without increasing the detection error.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to a technique for analyzing captured images periodically input from the outside.

Background Art

[0002] In various industrial devices such as printing devices and laser drawing devices, the position of a specific object such as an alignment mark may be detected from captured image data, and the position control of the device may be performed based on the detection result.

[0003] In such a device, if image processing is performed on the entire range of the captured image every time the captured image is input to perform position detection, the amount of calculation increases and it is inefficient. Regarding such a problem, a method for reducing the amount of calculation is described in, for example, Patent Document 1.

[0004] In Patent Document 1, the image is reduced in resolution, and the low-resolution data is analyzed to determine the presence or absence of an object. Then, only when it is determined that there is an object, a method of reducing the amount of calculation by performing the calculation again using the original high-resolution image is used.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] The method of Patent Document 1 is effective when there is a timing when the object to be detected does not exist in the captured image. However, when performing position detection of an alignment mark, the object to be detected is always within the imaging range and its position change is tracked. For such a purpose, even if the method of Patent Document 1 is used, the amount of calculation cannot be reduced.

[0007] This invention has been made in view of these circumstances, and aims to provide a technique for reducing the amount of computation required in image processing to detect a specific object from time-series image data. [Means for solving the problem]

[0008] To solve the above problems, the first invention of this application is an image processing method for detecting the position of a target from periodically acquired captured images, wherein a) the analysis target region of the captured image is analyzed, Inference is performed using convolution on the entire region of the analysis image from which the target region for analysis has been extracted. The position of the object to be detected, The estimation error, which indicates the likelihood of the position being determined as an inference result, is calculated, and the estimation error is calculated Detection error Let's assume The process includes step a) and step b) changing the analysis target area based on the magnitude of the detection error, and steps a) and b) are repeated multiple times.

[0009] The second invention of this application is an image processing method of the first invention, wherein in step b), the area to be analyzed is reduced when the detection error is smaller than a predetermined first threshold.

[0010] The third invention of this application is an image processing method of the second invention, wherein in step b), when the area to be analyzed is reduced in size, it is made to be at least a predetermined minimum size.

[0011] The fourth invention of this application is an image processing method of the second or third invention, wherein in step b), the area to be analyzed is enlarged when the detection error is greater than a predetermined second threshold.

[0012] The fifth invention of this application is an image processing method of any of the first to fourth inventions, further comprising the step of increasing the area to be analyzed when the displacement of the object to be detected is greater than a reference value, after step a) and step b).

[0013] The sixth invention of the present application is an image processing apparatus that detects the position of a detection target from a periodically acquired captured image, and includes an analysis region extraction unit that extracts an analysis target region of the captured image input from the outside and generates an analysis image, and analyzes the analysis image, Inference is performed using convolution operations on the entire region of the aforementioned analysis image. the position of the detection target, The estimation error, which indicates the likelihood of the position being determined as an inference result, is calculated, and the estimation error is calculated detection error Let's assume a mark position analysis unit, and an analysis region determination unit that changes the analysis target region based on the detection error.

Advantages of the Invention

[0014] According to the first to sixth inventions of the present application, the analysis target region can be reduced while keeping the detection error within a predetermined range. Therefore, the amount of calculation can be reduced while suppressing the reduction in detection accuracy.

[0015] In particular, according to the second invention of the present application, the amount of calculation can be reduced while suppressing the reduction in detection accuracy.

[0016] In particular, according to the fourth invention of the present application, when the detection error increases, the analysis target region can be expanded to reduce the detection error, so that the detection error can be within a predetermined range.

Brief Description of the Drawings

[0017] [Figure 1] It is a perspective view of a drawing apparatus provided with an image processing apparatus. [Figure 2] It is a block diagram showing the electrical connection between the control unit and each part in the drawing apparatus. [Figure 3] It is a control block diagram of a position detection unit as an image processing apparatus. [Figure 4] It is a flowchart showing the flow of image processing in the position detection unit. [Figure 5] It is an image diagram showing the relationship between the captured image and the analysis target region. [Figure 6] It is an image diagram showing the relationship between the captured image and the analysis target region.

Embodiments for Carrying Out the Invention

[0018] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0019] <1. Configuration of the drawing device> Hereinafter, a drawing device 1 including a position detection unit 90, which is an image processing device according to an embodiment of the present invention, will be described with reference to FIGS. 1 and 2. FIG. 1 is a perspective view of the drawing device 1. FIG. 2 is a block diagram showing the electrical connection between a control unit 10 including the position detection unit 90 and each unit in the drawing device.

[0020] This drawing device 1 is a device that irradiates spatially modulated light onto the upper surface of a substrate W such as a semiconductor substrate or a glass substrate coated with a photosensitive material to draw an exposure pattern on the upper surface of the substrate W. As shown in FIGS. 1 and 2, the drawing device 1 includes a transport mechanism 20, a frame 30, a drawing processing unit 40, a camera 50, and a control unit 10.

[0021] The transport mechanism 20 is a device that horizontally transports a flat stage 22 in a substantially constant posture on the upper surface of a base 21. The transport mechanism 20 includes a main scanning mechanism 23, a sub-scanning mechanism 24, and a rotation mechanism 25 (see FIG. 2). The main scanning mechanism 23 is a mechanism for transporting the stage 22 in the main scanning direction, which is one direction in the horizontal direction. The sub-scanning mechanism 24 is a mechanism for transporting the stage 22 in the sub-scanning direction, which is orthogonal to the main scanning direction, in the horizontal direction. The substrate W is held in a horizontal posture on the upper surface of the stage 22 and moves in the main scanning direction and the sub-scanning direction together with the stage 22. The rotation mechanism 25 can adjust the angle of the stage 22 around the vertical axis with respect to the base 21.

[0022] The frame 30 is a structure for holding the drawing processing unit 40 above the base 21. The frame 30 has a pair of support columns 31 and a bridging section 32. The pair of support columns 31 are erected at intervals in the sub-scanning direction. Each support column 31 extends upward from the upper surface of the base 21. The bridging section 32 extends in the sub-scanning direction between the upper ends of the two support columns 31. The stage 22 holding the substrate W passes between the pair of support columns 31 and below the bridging section 32.

[0023] The drawing processing unit 40 has two optical heads 41. The two optical heads 41 are fixed to the bridging section 32 with a gap between them in the sub-scanning direction. The drawing processing unit 40 also has an illumination optical system and a laser oscillator (not shown), and a laser drive unit 42 (see Figure 2). The illumination optical system, laser oscillator, and laser drive unit 42 are housed, for example, in the internal space of the bridging section 32. The laser drive unit 42 is electrically connected to the laser oscillator. When the laser drive unit 42 is operated, pulsed light is emitted from the laser oscillator. The pulsed light emitted from the laser oscillator is then introduced to the optical heads 41 via the illumination optical system.

[0024] An optical system including a spatial modulator is provided inside the optical head 41. The pulsed light introduced into the optical head 41 is modulated into a predetermined pattern by the spatial modulator and irradiated onto the upper surface of the substrate W. As a result, a photosensitive material such as resist coated on the upper surface of the substrate W is exposed.

[0025] Camera 50 (see Figure 2), although not shown in Figure 1, is attached, for example, to the lower surface of the bridging portion 32 of the frame 30. Camera 50 photographs the upper surface of the substrate W placed on the stage 22 and inputs the image to the control unit 10. The control unit 10 can detect the position and orientation of the substrate W by detecting the position of the alignment marks M provided on the upper surface of the substrate W from the image Di captured by camera 50.

[0026] The control unit 10 is a means for controlling the operation of each part of the drawing device 1. As conceptually shown in Figure 1, the control unit 10 is composed of a computer having a processor 101 such as a CPU, memory 102 such as RAM, and a storage unit 103 such as a hard disk drive. The storage unit 103 stores a computer program P for controlling the operation of the drawing device 1.

[0027] Furthermore, as shown in Figure 2, the control unit 10 is electrically connected to the drawing processing unit 40 (including the optical head 41 and the laser drive unit 42), the main scanning mechanism 23, the sub-scanning mechanism 24, the rotation mechanism 25, and the camera 50. The control unit 10 reads the computer program P and data D stored in the storage unit 103 into the memory 102, and the processor 101 performs calculation processing based on the computer program P and data D, thereby controlling the operation of each of the above parts in the drawing device 1. This allows the drawing process in the drawing device 1 to proceed.

[0028] The control unit 10 includes a position detection unit 90, a transport control unit 91, and a drawing control unit 92. The functions of the position detection unit 90, the transport control unit 91, and the drawing control unit 92 are realized by the operation of the computer processor 101 of the computer constituting the control unit 10 according to the computer program P.

[0029] The position detection unit 90 receives captured image Di from the camera 50 at regular intervals. The position detection unit 90 analyzes the captured image Di and detects the position, size, angle, etc., of the alignment mark M, thereby detecting the position and orientation of the substrate W. The position detection unit 90 then passes the detection result Do, which is the information obtained through the analysis, to the transport control unit 91.

[0030] The transport control unit 91 controls each part of the transport device 20. The transport control unit 91 transports the substrate W in a predetermined manner, correcting the position of the substrate W based on the detection result Do input from the position detection unit 90.

[0031] The drawing control unit 92 controls each part of the drawing processing unit 40. When the drawing device 1 is in operation, the operation of the transport mechanism 20 by the transport control unit 91 and the operation of the drawing processing unit 40 by the drawing control unit 92 are performed in conjunction. Specifically, exposure by the optical head 41 and transport of the substrate W by the transport device 20 are repeatedly performed. More specifically, the sub-scanning mechanism 24 transports the stage 22 in the sub-scanning direction while irradiating it with pulsed light from the optical head 41, thereby exposing a strip-shaped region (swath) extending in the sub-scanning direction, and then the main scanning mechanism 23 transports the stage 22 by one swath in the main scanning direction. The drawing device 1 draws a pattern over the entire upper surface of the substrate W by repeatedly performing this exposure in the sub-scanning direction and transporting the stage 22 in the main scanning direction.

[0032] While this drawing process is being performed, the position detection unit 90 analyzes the position, size, angle, etc. of the alignment marks M based on the captured image Di input from the camera 50 and inputs this information to the transport control unit 91. As a result, the transport control unit 91 checks whether the position of the substrate W has shifted from the desired position and corrects the position of the substrate W if a shift has occurred. This enables accurate pattern drawing on the upper surface of the substrate W.

[0033] <2. Configuration of the image processing device> Next, a position detection unit 90 as an image processing device according to one embodiment of the present invention will be described with reference to Figure 3. As described above, the position detection unit 90 is an image processing device implemented by the control unit 10, which detects the position of the alignment mark M to be detected from the captured image Di acquired periodically by the camera 50. As shown in Figure 3, the position detection unit 90 has an analysis region extraction unit 81, a mark position analysis unit 82, and an analysis region determination unit 83.

[0034] The analysis region extraction unit 81 extracts the region to be analyzed from the captured image Di input from outside the position detection unit 90 and generates an analysis image D1. Specifically, the analysis region extraction unit 81 generates an analysis image D1 by cutting out a portion of the image from the captured image Di taken by the camera 50 based on the region information D3 input from the analysis region determination unit 83. Then, it passes the analysis image D1 and the region information D3, which is the position information of the analysis image D1, to the mark position analysis unit 82.

[0035] The mark position analysis unit 82 analyzes the analysis image D1 and determines the detection result Do, which includes the position, size, and angle of the alignment mark M, and the detection error D2. The mark position analysis unit 82 then passes the detection result Do to the transport control unit 91, and passes the detection result Do and the detection error D2 to the analysis area determination unit 83.

[0036] The mark position analysis unit 82 performs inference using convolution operations over the entire area of ​​the analysis image D1 to detect the position of the alignment mark M. At the same time as position detection, an estimation error indicating the certainty of the result is obtained. The mark position analysis unit 82 uses this estimation error as the detection error D2. Since the amount of computation increases or decreases depending on the size of the analysis image D1, the smaller the size of the analysis image D1, the less computation is required, and the faster the computation can be performed.

[0037] The analysis region determination unit 83 changes the analysis target region based on the detection result Do and detection error D2, and passes the region information D3, which is the coordinate information of the analysis target region, to the analysis region extraction unit 81. Specifically, the analysis region determination unit 83 determines the size and position of the analysis target region based on the detection result Do and detection error D2. Then, the analysis region determination unit 83 passes the region information D3, which includes the size and position of the analysis target, to the analysis region extraction unit 81.

[0038] Specifically, the analysis area determination unit 83 of this embodiment reduces the analysis target area when the detection error D2 is smaller than a predetermined first threshold. However, even when the detection error D2 is smaller than the first threshold, i.e., when the analysis target area is reduced, the analysis area determination unit 83 ensures that the analysis target area is at least a predetermined minimum size.

[0039] Furthermore, the analysis area determination unit 83 of this embodiment enlarges the analysis target area when the detection error D2 is greater than or equal to a predetermined second threshold. The analysis area determination unit 83 does not change the size of the analysis target area when the detection error D2 is greater than or equal to the first threshold but less than the second threshold.

[0040] Furthermore, the analysis area determination unit 83 may enlarge the analysis area based on the displacement of the alignment mark M after changing the size of the analysis area due to the detection error D2. In this implementation, the analysis area determination unit 83 enlarges the analysis area when the displacement of the alignment mark M is greater than a predetermined reference value.

[0041] Furthermore, the analysis region determination unit 83 changes the position of the analysis target region in accordance with the displacement of the alignment mark M in the detection result Do. For example, the analysis region determination unit 83 changes the position of the analysis target region so that the center position of the analysis target region is as close as possible to the center position of the nearest alignment mark M.

[0042] <3. Image Processing Flow> Next, the image processing flow in the position detection unit 90 will be explained with reference to Figures 4 to 6. Figure 4 is a flowchart showing the image processing flow in the position detection unit 90. Figures 5 and 6 are illustrative diagrams showing the relationship between the captured image Di and the region to be analyzed. The image processing flow will be explained below with reference to the examples in Figures 5 and 6.

[0043] In the examples of Figures 5 and 6, the captured image Di is square in shape, with the same number of pixels in the vertical and horizontal directions. However, the present invention is not limited to this. The captured image Di may be rectangular in shape, with different numbers of pixels in the vertical and horizontal directions. Also, in the examples of Figures 5 and 6, the alignment mark M is cross-shaped, and its intersection is referred to as the center position. The alignment mark M may be any other shape. Furthermore, the center position of the alignment mark M may be, for example, the center of the two-dimensional coordinate range of the rectangle in which the alignment mark M exists, or it may be the centroid position.

[0044] The image processing shown in Figure 4 is started in the drawing device 1 simultaneously with the start of substrate transport W. In the image processing, first, the position detection unit 90 acquires the first captured image Di. Specifically, periodic imaging by the camera 50 is started, and the first captured image Di is input to the position detection unit 90 (step S101).

[0045] The analysis region extraction unit 81 extracts the region to be analyzed from the input captured image Di and generates the analysis image D1 (step S102). In the first step S102, since the region to be analyzed has not been set, the region information D3 at the start of image processing is set to the entire range of the captured image Di. Therefore, the analysis region extraction unit 81 uses the captured image Di as is for analysis image D1.

[0046] Next, the mark position analysis unit 82 analyzes the analysis image D1 and detects the position, angle, size, etc. of the alignment mark M (step S103). At the same time, the mark position analysis unit 82 also calculates the detection error D2, which is an estimated error. The mark position analysis unit 82 passes the detection result Do, which includes the position, angle, and size of the alignment mark M, and the detection error D2 to the analysis region determination unit 83.

[0047] Next, the analysis region determination unit 83 compares the value of the detection error D2 with predetermined first and second thresholds (step S104). Then, the analysis region determination unit 83 changes the size of the analysis target region according to the magnitude of the value of the detection error D2 (steps S105 to S111). The second threshold has a value greater than the first threshold.

[0048] If the detection error D2 is determined to be less than the first threshold (step S104: less than the first threshold), the analysis region determination unit 83 reduces the size of the analysis region (step S105). Specifically, for example, the size of the analysis region is reduced by p pixels in both the vertical and horizontal directions. That is, if the size of the analysis region at that time is m pixels × n pixels, it becomes (mp) pixels × (n-p) pixels.

[0049] Furthermore, the analysis region determination unit 83 changes the position of the analysis region so that the center position of the analysis target region coincides as closely as possible with the center position of the alignment mark M.

[0050] In this way, the analysis area determination unit 83 reduces the analysis area when the detection error D2 is relatively small. This makes it possible to reduce the amount of computation in the next image analysis step (step S103) while suppressing a decrease in detection accuracy.

[0051] Following step S105, the analysis region determination unit 83 determines whether the size after reduction is greater than or equal to a predetermined minimum region size (step S106). The minimum region size may be a predetermined fixed value, or it may vary based on the detection result Do. For example, the minimum region size may vary based on the size of the alignment mark M in the detection result Do. In that case, for example, if the size of the alignment mark M is approximately x pixels square, the size of the analysis target region may be (x+q) pixels square, or a constant multiple of x pixels square.

[0052] In step S104, if it is determined that the detection error D2 is greater than or equal to the first threshold and less than the second threshold (step S104: greater than or equal to the first threshold and less than the second threshold), the analysis region determination unit 83 does not change the size of the analysis target region (step S108). At this time, the analysis region determination unit 83 changes the position of the analysis target region so that the center position of the analysis target region coincides as closely as possible with the center position of the alignment mark M.

[0053] In step S104, if it is determined that the detection error D2 is greater than or equal to the second threshold (step S104: greater than or equal to the second threshold), the analysis region determination unit 83 expands the size of the analysis target region (step S109). Specifically, for example, the size of the analysis target region is increased by r pixels in both the vertical and horizontal directions. That is, if the size of the analysis target region at that time is m pixels × n pixels, it is made (m+r) pixels × (n+r) pixels.

[0054] Thus, the analysis area determination unit 83 expands the analysis area when the detection error D2 is relatively large, that is, when the detection error D2 increases. This reduces the detection error D2 in the next image analysis step (step S103) and suppresses a decrease in detection accuracy.

[0055] After the determination in step S104, and after steps S105-S107, S108, or S109 are completed, the analysis domain determination unit 83 determines whether the displacement of the alignment mark M is greater than the reference value (step S110). The displacement of the alignment mark M is calculated, for example, the difference between the center position of the most recent alignment mark M and the center position of the alignment mark M one step prior. Alternatively, the displacement of the alignment mark M may be calculated, for example, the difference between the center position of the most recent alignment mark M and the center position of the alignment mark M a predetermined number of steps prior, or it may take into account not only the difference in center position but also the angle of the alignment mark M.

[0056] In step S110, if the displacement of the alignment mark M is greater than a predetermined reference value (step S110: Yes), the size of the analysis target area is enlarged (step S111). Specifically, for example, the size of the analysis target area is increased by s pixels in both the vertical and horizontal directions. That is, if the size of the analysis target area at that time is m pixels × n pixels, it is made (m+s) pixels × (n+s) pixels. After that, the analysis area determination unit 83 passes the area information D3, which includes the size and position of the analysis target area, to the analysis area extraction unit 81 and returns to step S101.

[0057] On the other hand, in step S110, if the displacement of the alignment mark M is less than or equal to a predetermined reference value (step S110: No), the size and position of the analysis target region at that time are passed to the analysis region extraction unit 81 as region information D3, and the process returns to step S101.

[0058] In this way, the size of the analysis area is adjusted each time the captured image Di is acquired and analyzed, according to the value of the detection error. This makes it possible to reduce the analysis area within a range where the detection error does not exceed a predetermined second threshold. In other words, the amount of computation required for analysis can be reduced while maintaining a certain level of detection accuracy for the alignment marks M.

[0059] Here, Figures 5 and 6 show examples of changes in the analysis target area in the image processing of this embodiment. Figures 5 and 6 show the case where the position and angle of the cross-shaped alignment mark M do not move (no displacement), the detection error is less than the first threshold in the 1st to 7th steps S104, and the displacement of the alignment mark M is less than or equal to the reference value in the 1st to 7th steps S110.

[0060] In Figures 5 and 6, the range of the captured image Di is shown by a solid line. The analysis areas A1 (1st), A2 (2nd), A3 (3rd), A4 (4th), A5 (5th), A6 (6th), A7 (7th), and A8 (8th) are shown by dashed lines. Note that the range of the first analysis area A1 is shown slightly inside the solid line representing the captured image Di. The actual A1 is the same range as Di, but it is intentionally offset to avoid ambiguity when the solid and dashed lines overlap.

[0061] In the examples of Figures 5 and 6, the captured image Di and the first analysis target area A1 are 200 pixels square. In the first step S104, since the detection error is less than the first threshold, in step S105, the size of the analysis target area is reduced by 20 pixels at a time until it is 180 pixels square. Here, as shown in the upper part of Figure 5, if the analysis target area is set so that its center coincides with the center of the alignment mark M, the area shown as A2' in the upper part of Figure 5 becomes the analysis target area. Since area A2' has an area that does not overlap with the captured image Di, in the examples of Figures 5 and 6, the position of the analysis target area is corrected, and within the captured image Di, area A2, which is 180 pixels square and whose center is closest to the center of the alignment mark M, is set as the analysis target area.

[0062] Next, in the second step S104, since the detection error is less than the first threshold, in step S105, the size of the analysis target area is reduced by 20 pixels at a time until it is 160 pixels square. Here, as shown in the lower part of Figure 5, if the analysis target area is set so that its center coincides with the center of the alignment mark M, the area shown as A3' in the lower part of Figure 5 becomes the analysis target area. Since area A3' has an area that does not overlap with the captured image Di, the position of the analysis target area is corrected, and within the captured image Di, area A4, which is 160 pixels square and whose center is closest to the center of the alignment mark M, is set as the analysis target area.

[0063] In the third step S104, the size of the analysis area is similarly reduced by 20 pixels each time, until it is 140 pixels square. Here, as shown in the upper part of Figure 6, if the analysis area is set so that its center coincides with the center of the alignment mark M, the area shown as A4' in the upper part of Figure 6 becomes the analysis area. Since area A4' has an area that does not overlap with the captured image Di, the position of the analysis area is corrected, and within the captured image Di, area A4, which is 140 pixels square and whose center is closest to the center of the alignment mark M, is set as the analysis area.

[0064] Similarly, in steps S104 from the 4th to the 7th, the size of the analysis area is reduced by 20 pixels each time, to 120 pixels square, 100 pixels square, and 80 pixels square, respectively. As shown in the lower part of Figure 6, in these cases, if the analysis area is set so that its center coincides with the center of the alignment mark M, then analysis areas A5 to A8 shown in the lower part of Figure 6 are all within the captured image Di, so there is no need to correct their position.

[0065] As shown in the examples in Figures 5 and 6, the amount of computation in image analysis can be appropriately reduced by shrinking the analysis area to a portion of the alignment mark M so that the detection error falls within a predetermined range. In actual image processing, the position of the alignment mark M fluctuates. The analysis area also follows the fluctuations of the alignment mark M, including the alignment mark M and its surrounding area. At that time, the analysis area is appropriately reduced so that the detection error of the alignment mark M falls within a predetermined range.

[0066] <4. Variation> Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above.

[0067] In the above embodiment, the control object of the control unit 10, which has a position detection unit 90 that is an image processing device, was a drawing device 1. Therefore, the position detection unit 90 appropriately adjusted the analysis target area in order to detect the position and angle of the alignment marks M provided on the substrate W. However, the objects detected by the image processing device of the present invention are not limited to this. For example, it can be applied to various applications such as detecting alignment marks provided on a survey object in a surveying device, or detecting alignment marks provided on a substrate in a substrate inspection device.

[0068] Furthermore, in the above embodiment, the detection error was divided into three ranges depending on the value of the detection error: less than the first threshold, between the first threshold and the second threshold, and greater than or equal to the second threshold. Three steps were then selected: reducing the area to be analyzed, leaving it unchanged, and expanding it. However, the present invention is not limited thereto.

[0069] For example, the first threshold and the second threshold may be the same single threshold. Furthermore, the analysis area may be reduced if the detection error is less than the threshold, and expanded if the detection error is greater than or equal to the threshold.

[0070] Alternatively, for example, three thresholds—a first threshold, a second threshold, and a third threshold—can be used to define four ranges for the detection error. If the detection error is less than the first threshold, the analysis area can be reduced by two steps. If the detection error is between the first and second thresholds, the analysis area can be reduced by one step. If the detection error is between the second and third thresholds, the analysis area can be expanded by one step. If the detection error is greater than or equal to the third threshold, the analysis area can be expanded by two steps.

[0071] Furthermore, the elements that appear in the above embodiments and modifications may be combined as appropriate, to the extent that no contradictions arise. [Explanation of Symbols]

[0072] 1 Drawing device 10 Control Unit 50 Cameras 81 Analysis area extraction part 82 Mark Position Analysis Unit 83 Analysis area determination section 90 Position detection unit D1 Image for analysis D2 Detection Error D3 Area Information Di Photographed Images Do detection results

Claims

1. An image processing method for detecting the position of a target from periodically acquired captured images, a) The process of analyzing the region to be analyzed in the captured image, performing inference by convolution on the entire region of the analysis image from which the region to be analyzed has been extracted, determining the position of the object to be detected and the estimation error indicating the certainty of the position which is the inference result, and using the estimation error as the detection error, b) A step of changing the analysis target area based on the magnitude of the detection error, Includes, An image processing method comprising repeating steps a) and b) multiple times.

2. The image processing method according to claim 1, An image processing method in step b) above, wherein if the detection error is smaller than a predetermined first threshold, the area to be analyzed is reduced in size.

3. The image processing method according to claim 2, An image processing method in which, when reducing the size of the area to be analyzed in step b), the size is made to be at least a predetermined minimum range.

4. The image processing method according to claim 2, An image processing method in which, in step b), if the detection error is greater than a predetermined second threshold, the area to be analyzed is enlarged.

5. An image processing method according to any one of claims 1 to 4, c) After steps a) and b), if the displacement of the object to be detected is greater than the reference value, a step to enlarge the area to be analyzed. An image processing method further comprising the following.

6. An image processing device for detecting the position of a target from images acquired periodically, An analysis region extraction unit extracts the region to be analyzed from the captured image input from an external source and generates an image for analysis, A mark position analysis unit analyzes the aforementioned analysis image, performs inference by convolution operation on the entire region of the analysis image, determines the position of the detection target and the estimation error indicating the likelihood of the position being determined as an inference result, and uses the estimation error as the detection error. An analysis region determination unit that changes the analysis target region based on the detection error, An image processing apparatus having

Images