Image processing system, object recognition system, image processing method, and program

Abstract

To provide an image processing system for recognizing a recognition target accurately, a target recognition system, an image processing method, and a program.SOLUTION: An image processing system 2 includes an image acquisition unit 21 and an image recognition unit 22. The image acquisition unit 21 acquires data of a captured image obtained by imaging a recognition target. The image recognition unit (22) determines a similarity between a reference patten and a detection patten which is a combination of color and graphic detected in a target region including the recognition target, and recognizes the recognition target included in the target region on the basis of the similarity. The reference pattern has, as reference graphics, a center graphic and at least one frame graphics surrounding the center graphic in layers. Each of colors of the reference graphics is different from a color of a reference image adjacent to a reference image out of the multiple reference graphics.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to an image processing system, an object recognition system, an image processing method, and a program.

Background Art

[0002] The reading device of Patent Document 1 reads a color ring code from a video signal obtained by imaging a concentric color ring code. The color ring code is a code in which information based on an N -ary color is added to a recognition - use concentric ring composed of a plurality of circles arranged concentrically.

[0003] The reading device includes color image processing means. The color image processing means processes a video signal obtained by a color television camera imaging a color ring code and searches for a color code signal from the video signal. Specifically, the color image processing means performs A / D conversion processing on the video signal for one screen among the video signals continuously input from the color television camera to quantize color information, and stores the quantized color information as color - digitized information in an internal memory. Then, the color image processing means scans the color - digitized information stored in the internal memory in the same direction as or perpendicular to the scanning direction of the color television camera, sequentially extracts it, and performs a detection process for the color code.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] It is difficult to accurately recognize a recognition target such as the color ring code of Patent Document 1.

[0006] The purpose of this disclosure is to provide an image processing system, an object recognition system, an image processing method, and a program that can accurately recognize objects to be recognized. [Means for solving the problem]

[0007] An image processing system according to one aspect of this disclosure includes an image acquisition unit that acquires data of an image of a target to be recognized, and an image recognition unit that performs image recognition processing on the image. The object of recognition is located on a hat worn on a person's head. The captured image has a target region in which the object to be recognized is depicted. The image recognition unit determines the similarity between a detection pattern, which is a combination of colors and shapes detected from the target region, and a reference pattern, which is a predetermined combination of colors and shapes, and recognizes the object to be recognized depicted in the target region based on the similarity. The reference pattern has a central shape and at least one frame shape that surrounds the central shape in layers, each of which has a reference shape, and the color of each of the plurality of reference shapes is different from the color of the reference shape adjacent to that reference shape among the plurality of reference shapes. The image recognition unit inputs the data of the captured image into a learning model constructed by machine learning and obtains the recognition result of the object to be recognized from the learning model. The training image of the learning model is created from the captured images taken of each of the multiple people, each of whom is wearing a hat with the same reference pattern formed on it.

[0008] A target recognition system according to one aspect of this disclosure comprises the image processing system described above and an imaging device that captures the target to be recognized and generates data of the captured image.

[0009] The image processing method relating to one aspect of this disclosure is: The image processing method is performed by a computer system.The process includes an image acquisition step of acquiring data from an image of a target object, and an image recognition step of applying image recognition processing to the image. The target object is provided on a hat placed on a person's head. The image has a target region in which the target object is depicted. The image recognition step calculates the similarity between a detection pattern, which is a combination of colors and shapes detected from the target region, and a reference pattern, which is a predetermined combination of colors and shapes, and recognizes the target object depicted in the target region based on the similarity. The reference pattern has a central shape and at least one frame shape that surrounds the central shape in layers, each of which has a different color from the reference shapes adjacent to it among the plurality of reference shapes. In the image recognition step, the image data is input to a learning model constructed by machine learning, and the recognition result of the target object is obtained from the learning model. The training images for the learning model are created from images of multiple people, each of whom is photographed wearing the hat with the same reference pattern.

[0010] A program according to one aspect of the present invention causes a computer system to execute the image processing method described above. [Effects of the Invention]

[0011] As explained above, this disclosure has the effect of enabling accurate recognition of the object to be recognized. [Brief explanation of the drawing]

[0012] [Figure 1] Figure 1 is a block diagram showing the configuration of an object recognition system equipped with an image processing system according to an embodiment. [Figure 2] Figure 2 is a schematic diagram showing the installation of the imaging device included in the aforementioned object recognition system. [Figure 3] Figure 3 is a schematic diagram showing the placement of the objects to be recognized in the aforementioned object recognition system. [Figure 4]FIG. 4 is a plan view showing a recognition target used in the same object recognition system. [Figure 5] FIG. 5 is a diagram showing a reference pattern used in the same object recognition system. [Figure 6] FIG. 6 is a diagram for explaining the color used in the same object recognition system. [Figure 7] FIG. 7 is a diagram showing a captured image in the same object recognition system. [Figure 8] FIG. 8 is a diagram showing an extracted image in the same object recognition system. [Figure 9] FIG. 9 is a diagram showing another extracted image in the same object recognition system. [Figure 10] FIG. 10 is a flowchart showing the same image processing method. [Figure 11] FIG. 11A is a plan view showing a recognition target used in the first modification. FIG. 11B is a diagram showing a reference pattern. [Figure 12] FIG. 12A is a plan view showing a recognition target used in the second modification. FIG. 12B is a diagram showing a reference pattern.

BEST MODE FOR CARRYING OUT THE INVENTION

[0013] The following embodiments relate to an image processing system, an object recognition system, an image processing method, and a program. More specifically, they relate to an image processing system, an object recognition system, an image processing method, and a program for recognizing a recognition target based on a captured image.

[0014] Note that the embodiments described below are merely examples of the embodiments of the present disclosure. The present disclosure is not limited to the following embodiments, and various modifications can be made according to the design and the like as long as the effects of the present disclosure can be achieved.

[0015] (Embodiment) (1) Outline of Object Recognition System The object recognition system of this embodiment uses image processing technology of a computer system to recognize an object to be recognized shown in a captured image captured by an imaging device.

[0016] The object to be recognized is attached to a person or an object, etc., and the recognition result of the object to be recognized is used for identification of a person or an object, detection of a movement trajectory of a person or an object, etc.

[0017] FIG. 1 shows a block configuration of an object recognition system 1 of this embodiment. The object recognition system 1 includes an image processing system 2 and an imaging device 3. The imaging device 3 captures an object to be recognized 9 and generates data of a captured image G1 (see FIG. 7).

[0018] The image processing system 2 includes an image acquisition unit 21 that acquires data of a captured image G1 (see FIG. 7) that has captured the object to be recognized 9, and an image recognition unit 22 that performs image recognition processing on the captured image G1. The captured image G1 has a target area R9 (see FIG. 7) in which the object to be recognized 9 is shown. The image recognition unit 22 obtains the similarity between a detection pattern that is a combination of colors and shapes detected from the target area R9 and a reference pattern Pn (see FIG. 5) that is a combination of colors and shapes determined in advance, and recognizes the object to be recognized 9 shown in the target area R9 based on the similarity. The reference pattern Pn has each of a central figure Pn1 and at least one frame figure Pn2 that surrounds the periphery of the central figure Pn1 in a layered manner as reference figures, and the colors of the plurality of reference figures Pn1 and Pn2 are different from the colors of the reference images adjacent to the reference image among the plurality of reference figures Pn1 and Pn2. Note that n described above is a positive integer.

[0019] The object recognition system 1 and the image processing system 2 having the above-described configuration can accurately recognize the object to be recognized 9.

[0020] (2) Details of the object recognition system As shown in Figure 1, the object recognition system 1 comprises an image processing system 2 and an imaging device 3, and performs recognition processing to recognize an object 9 within space F1 shown in Figure 2. Space F1 is a space in which at least one person 7 exists, such as a factory, store, or commercial facility. Although Figure 2 illustrates one person 7, multiple people 7 may exist in space F1.

[0021] Space F1 has a floor F11 and a ceiling F12. Person 7 is positioned above the floor F11, below the ceiling F12. Person 7 in space F1 can stand, sit, walk, run, and stand still.

[0022] (2.1) Recognition target As shown in Figure 3, person 7 is wearing a hat 8 on their head. On the top surface (top crown) of the hat 8, a recognition object 9A, as shown in Figure 4, is formed. The recognition object 9A is a double circle and is formed on the top surface of the hat 8 by methods such as printing, sewing, or attaching. The shape of the top surface of the hat 8 on which the recognition object 9A is formed may be either a flat or curved surface. For example, in a plan view when the hat 8 worn on person 7's head is viewed from directly above, the recognition object 9A will be circular.

[0023] The recognition object 9A comprises a circular central part 91A and an annular frame part 92A that surrounds the central part 91A along its outer circumference. The central part 91A and the frame part 92A are coaxially located. Furthermore, the colors of the circular central part 91A and the annular frame part 92A are different from each other. Moreover, the combination of the colors of the circular central part 91A and the annular frame part 92A is associated one-to-one with each person 7. That is, when multiple people 7 exist in space F1, the recognition object 9A formed on each of the multiple people 7 has a different combination of the colors of the circular central part 91A and the annular frame part 92A.

[0024] (2.2) Reference Patterns The color and shape of the central part 91A and the frame part 92A of the recognition target 9A will be one of the multiple reference patterns Pn (where n is a positive integer) shown in Figure 5. In other words, the specific combination of the color and shape of the central part 91A and the color and shape of the frame part 92A will be one of the multiple reference patterns Pn.

[0025] Reference pattern Pn is a double-circular figure composed of a central figure Pn1 and a frame figure Pn2. Central figure Pn1 is circular. Frame figure Pn2 is an annular shape that surrounds central figure Pn1 along its outer perimeter. Central figure Pn1 and frame figure Pn2 are coaxial. Furthermore, the colors of the circular central figure Pn1 and the annular frame figure Pn2 are different from each other. Moreover, the combination of the colors of central figure Pn1 and frame figure Pn2 is different for each reference pattern Pn. In other words, reference pattern Pn is a combination of shapes (circular central figure Pn1 and annular frame figure Pn2) and colors (color of central figure Pn1 and color of frame figure Pn2).

[0026] In this embodiment, the reference pattern Pn has one frame shape Pn2. As a result, compared to the case where a reference pattern with multiple frame shapes is used, the processing load on the image recognition unit 22 (described later) to recognize the recognition target 9A is reduced, and the accuracy of the recognition process is improved even with low image resolution.

[0027] Furthermore, it is preferable that the shape of the outline of the central figure Pn1 and the shape of the frame figure Pn2 are similar to each other. In Figure 5, the outline of the central figure Pn1 is annular, the frame figure Pn2 is annular, and the shape of the outline of the central figure Pn1 and the shape of the frame figure Pn2 are similar to each other. As a result, the load on the recognition process in which the image recognition unit 22, described later, recognizes the recognition target 9A is reduced, and the accuracy of the recognition process is improved.

[0028] Furthermore, it is preferable that the shape of the outline of the central figure Pn1 and the shape of the frame figure Pn2 are rotationally symmetric. In Figure 5, the outline of the central figure Pn1 is annular, the frame figure Pn2 is annular, and the shape of the outline of the central figure Pn1 and the shape of the frame figure Pn2 are rotationally symmetric. As a result, regardless of the imaging direction of the recognition target 9A, the image recognition unit 22 described later can more easily recognize the recognition target 9A, reducing the load on the recognition processing performed by the image recognition unit 22 when recognizing the recognition target 9A, and improving the accuracy of the recognition processing.

[0029] Furthermore, it is preferable that the shape of the outline of the central figure Pn1 and the shape of the frame figure Pn2 are annular, and that the central figure Pn1 and the frame figure Pn2 are arranged coaxially. In Figure 5, the central figure Pn1 is circular, and the frame figure Pn2 is an annular shape that surrounds the central figure Pn1 along its outer circumference, and the central figure Pn1 and the frame figure Pn2 are arranged coaxially. As a result, regardless of the imaging direction of the recognition target 9A, the image recognition unit 22 described later can more easily recognize the recognition target 9A, reducing the load on the recognition processing performed by the image recognition unit 22 when recognizing the recognition target 9A, and improving the accuracy of the recognition processing.

[0030] Furthermore, it is preferable that the colors of each of the multiple reference shapes (center shape Pn1 and frame shape Pn2) are colors in which the mixing amounts of each of the multiple colors are individually set to the maximum or minimum value. As a result, the difference between the colors of each of the multiple reference shapes (center shape Pn1 and frame shape Pn2) becomes larger, making it easier for the image recognition unit 22, described later, to recognize the recognition target 9A, and improving the accuracy of the recognition process in which the image recognition unit 22 recognizes the recognition target 9A.

[0031] For example, the color of each of the multiple reference shapes (center shape Pn1 and frame shape Pn2) is preferably a color obtained by mixing the three colors C1, C2, and C3 shown in Figure 6, with each of their mixing amounts individually set to the maximum or minimum value. By individually setting the mixing amounts of the three colors C1, C2, and C3 to the maximum or minimum value, eight colors C(X1, X2, X3) are formed. In C(X1, X2, X3), X1 is the mixing amount of color C1. In C(X1, X2, X3), X2 is the mixing amount of color C2. In C(X1, X2, X3), X3 is the mixing amount of color C3. That is, C(X1, X2, X3) is a color produced by mixing colors C1, C2, and C3 when the ratio of their mixing amounts is X1:X2:X3.

[0032] Then, if we set the maximum mixing amount of color C1 to "C1m", the maximum mixing amount of color C2 to "C2m", the maximum mixing amount of color C3 to "C3m", and the minimum mixing amount of each of colors C1, C2, and C3 to "0", the following eight colors will be generated: C(0,0,0), C(C1m,0,0), C(0,C2m,0), C(0,0,C3m), C(C1m,C2m,0), C(C1m,0,C3m), C(0,C2m,C3m), and C(C1m,C2m,C3m). In this case, for each of the multiple reference patterns Pn shown in Figure 5, the colors of the central figure Pn1 and the frame figure Pn2 will be one of the following: C(0,0,0), C(C1m,0,0), C(0,C2m,0), C(0,0,C3m), C(C1m,C2m,0), C(C1m,0,C3m), C(0,C2m,C3m), or C(C1m,C2m,C3m). Note that the colors of the central figure Pn1 and the frame figure Pn2 are different.

[0033] Specifically, colors C1, C2, and C3 are defined as the three primary colors of light. That is, color C1 is red, color C2 is green, and color C3 is blue. In this case, C(0,0,0) is black, C(C1m,0,0) is blue, C(0,C2m,0) is red, C(0,0,C3m) is green, C(C1m,C2m,0) is magenta, C(C1m,0,C3m) is cyan, C(0,C2m,C3m) is yellow, and C(C1m,C2m,C3m) is white. Therefore, in each of the multiple reference patterns Pn shown in Figure 5, the color of the central figure Pn1 and the color of the frame figure Pn2 will be one of the following: black, blue, red, green, magenta, cyan, yellow, or white.

[0034] Note that colors C1, C2, and C3 do not have to be the three primary colors of light.

[0035] (2.3) Imaging device As shown in Figure 2, the imaging device 3 is installed on the ceiling F12 of space F1. The imaging device 3 forms an imaging area below itself and generates image data by imaging this area. In Figure 2, the imaging device 3 installed on the ceiling F12 is positioned higher than the recognition target 9 of the person's head 7 on the floor F11, and it images the recognition target 9 from above.

[0036] Specifically, the imaging device 3 has a lens and multiple image sensors. The image sensors are CCDs (Charged Coupled Devices) or CMOSs ​​(Complementary Metal Oxide Semiconductors), etc. The imaging device 3 can capture color video by receiving light focused by the lens into the image sensors. The color video consists of many frames (still color images). In other words, the imaging device 3 outputs the data of the color video, which consists of many frames, as captured image data to the image processing system 2.

[0037] The imaging device 3 may always output the captured image data to the image processing system 2, or it may output the captured image data only when a person 7 is present in the imaging area of ​​the imaging device 3.

[0038] (2.4) Image processing system As shown in Figure 1, the image processing system 2 comprises an image acquisition unit 21, an image recognition unit 22, a storage unit 23, and an output unit 24.

[0039] The image processing system 2 preferably includes a computer system. That is, in the image processing system 2, some or all of the functions of the image processing system 2 are realized by a processor, such as a CPU (Central Processing Unit) or MPU (Micro Processing Unit), reading and executing a program stored in memory. The image processing system 2 primarily includes a processor that operates according to the program as its hardware configuration. The type of processor is not limited as long as it can realize its functions by executing a program. The processor consists of one or more electronic circuits, including a semiconductor integrated circuit (IC) or an LSI (Large Scale Integration). Here, we refer to them as ICs and LSIs, but the name changes depending on the degree of integration; they may also be called system LSIs, VLSIs (Very Large Scale Integrations), or ULSIs (Ultra Large Scale Integrations). Field-programmable gate arrays (FPGAs), which are programmed after the LSI is manufactured, or reconfigurable logic devices that allow for the reconfiguration of internal junction relationships or the setup of internal circuit compartments, can also be used for the same purpose. Multiple electronic circuits may be integrated on a single chip or provided on multiple chips. Multiple chips may be arranged in a cluster or distributed.

[0040] (2.4.1) Image acquisition unit The image acquisition unit 21 acquires image data generated by the imaging device 3 by performing wired or wireless communication with the imaging device 3. In this embodiment, the image data is color video data. Wired communication is, for example, wired communication via twisted pair cable, dedicated communication line, or LAN (Local Area Network) cable. Wireless communication is, for example, wireless communication conforming to standards such as Wi-Fi (registered trademark) or unlicensed low-power radio (specified low-power radio).

[0041] Specifically, the image acquisition unit 21 acquires the data of the captured image G1 shown in Figure 7 from the imaging device 3.

[0042] (Image captured) The captured image G1 shown in Figure 7 is a single frame from a color video captured from above of a person 7 (see Figure 3) wearing a hat 8 on which the recognition target 9A is formed. The captured image G1 includes a person region R7 in which person 7 is visible, a hat region R8 in which the hat 8 is visible, and a target region R9 in which the recognition target 9A is visible.

[0043] The target region R9 includes a central region R91 in which the central part 91A of the recognition target 9A is captured, and a frame region R92 in which the frame portion 92A of the recognition target 9A is captured.

[0044] (2.4.2) Image recognition unit, memory unit The image recognition unit 22 determines the similarity between the detection pattern, which is a combination of colors and shapes detected from the target region R9 of the captured image G1, and the reference pattern Pn (see Figure 5), which is a predetermined combination of colors and shapes. Based on the similarity, the image recognition unit 22 recognizes the recognition target 9A that is captured in the target region R9.

[0045] In this embodiment, the image recognition unit 22 inputs the data of the captured image G1 into a learning model constructed by machine learning. The learning model calculates the similarity between the detected pattern and each of the multiple reference patterns Pn, and recognizes the recognition target 9A that is captured in the target region R9 based on each similarity. The image recognition unit 22 obtains the recognition result of the recognition target 9A from the learning model. The storage unit 23 stores the data of the learning model constructed by machine learning.

[0046] The learning model used by the image recognition unit 22 extracts the extracted image G2 shown in Figure 8 from the captured image G1. The extracted image G2 is a region that includes the person region R7, the hat region R8, and the target region R9, and the size of the extracted image G2 is smaller than the size of the captured image G1. The learning model then detects the combination of color and shape of the target region R9 in the extracted image G2 as the detection pattern for the target region R9. The learning model calculates the similarity between the detection pattern of the target region R9 and each of the multiple reference patterns Pn. The learning model then recognizes the reference pattern Pn with the highest similarity among the multiple reference patterns Pn as the detection pattern for the target region R9.

[0047] Specifically, deep learning is employed as the machine learning method, and the learning model is constructed using deep learning with a Convolutional Neural Network (CNN). Deep learning uses numerous frames contained in a video of the recognition target 9A, which is set to one of several reference patterns Pn as shown in Figure 5, as training images. The numerous training images are pre-classified to determine which of the multiple reference patterns Pn they correspond to. The learning model then extracts an image (similar to the extracted image G2 shown in Figure 8) from the training images and performs recognition processing on the extracted image to learn. During the learning process, the learning model calculates the similarity between the detection pattern of the training image and each of the multiple reference patterns Pn, and recognizes the reference pattern Pn with the highest similarity as the detection pattern of the training image.

[0048] In the learning process for constructing the learning model, it is preferable to have multiple people wear hats 8 with the same reference pattern Pn and move around in space F1, in order to use a diverse set of training images. The imaging device 3 captures images of each of the multiple people wearing hats 8 with the same reference pattern Pn and generates data from the captured images, and training images are created from these captured images. By constructing a learning model using such training images, the accuracy of the recognition process in recognizing the recognition target 9A improves even when the person 7 wearing the hat 8 changes.

[0049] Furthermore, in order to shorten the time required to classify a large number of training images during the learning process, it is preferable to use training images that show only one person wearing a hat (8).

[0050] Furthermore, a general-purpose algorithm can be used for extracting images from training images.

[0051] Here, if the imaging environment of the imaging device 3 changes, the appearance of the recognition target 9A in the target area R9 of the captured image G1 also changes. The imaging environment includes, for example, the lighting environment of space F1, the amount and angle of sunlight incident on space F1, the installation and operation status of equipment in space F1, and the position and posture of person 7. When the imaging environment changes, the color of the target area R9 in the captured image G1 also changes, making it difficult to recognize the recognition target 9A that is captured in the target area R9.

[0052] Therefore, the image recognition unit 22 of this embodiment recognizes the recognition target 9A that is captured in the target region R9 of the captured image G1 using a learning model constructed by deep learning using a CNN (Convolutional Neural Network). For example, by using a large amount of training data created under various imaging environments when learning a single reference pattern Pn, the recognition accuracy of the reference pattern Pn can be improved even if the imaging environment changes. As a result, the image recognition unit 22 can accurately recognize the recognition target 9A that is captured in the target region R9 even if the imaging environment changes. Furthermore, even if the recognition target 9A is distorted or the captured image G1 is a low-resolution image, the image recognition unit 22 can accurately recognize the recognition target 9A that is captured in the target region R9.

[0053] The learning model may be constructed using deep learning with FCN (Fully Convolutional Networks). Alternatively, the learning model may be constructed using other algorithms such as Support Vector Machines.

[0054] The learning model used by the image recognition unit 22 may extract the extracted image G3 shown in Figure 9 from the captured image G1. The extracted image G3 is a region that includes the hat region R8 and the target region R9, and the size of the extracted image G3 is even smaller than the size of the extracted image G2. The learning model then detects the combination of color and shape of the target region R9 in the extracted image G3 as the detection pattern for the target region R9.

[0055] (2.4.3) Output section The output unit 24 outputs the recognition result (recognition result) from the image recognition unit 22 to the external system 4. The output unit 24 outputs the recognition result data to the external system 4 by performing wired or wireless communication with the external system 4. Wired communication is, for example, wired communication via twisted pair cable, dedicated communication line, or LAN cable. Wireless communication is, for example, wireless communication conforming to standards such as Wi-Fi (registered trademark) or unlicensed low-power radio (specified low-power radio).

[0056] The external system 4 performs tasks such as identifying person 7 in space F1 or detecting the movement trajectory of person 7 based on the recognition results of the image processing system 2. For example, the external system 4 uses the recognition results of the image processing system 2 to analyze the movement of person 7 as they move together with the recognition target 9A. Such movement analysis is used to improve the work efficiency of workers in a factory.

[0057] (3) Image processing method The image processing method performed by the image processing system 2 described above is summarized in the flowchart of Figure 10. This image processing method includes an image acquisition step S1 and an image recognition step S2. Preferably, the image processing method further includes an output step S3.

[0058] In the image acquisition step S1, the image acquisition unit 21 acquires data from the captured image G1 in which the recognition target 9 is captured. The captured image G1 has a target region R9 in which the recognition target 9 is captured.

[0059] In the image recognition step S2, the image recognition unit 22 performs image recognition processing on the captured image G1. Then, in the image recognition step S2, the image recognition unit 22 calculates the similarity between the detection pattern, which is a combination of color and shape detected from the target region R9, and the reference pattern Pn, which is a predetermined combination of color and shape, and recognizes the recognition target 9 that is captured in the target region R9 based on the similarity. The reference pattern Pn has a central figure Pn1 and at least one frame figure Pn2 that surrounds the central figure Pn1 in layers as reference figures, and the color of each of the multiple reference figures Pn1 and Pn2 is different from the color of the reference image adjacent to the reference image among the multiple reference figures Pn1 and Pn2.

[0060] The image processing method described above can accurately recognize the object to be recognized 9.

[0061] (4) First variation The recognition target 9 may also be the recognition target 9B shown in Figure 11A. The recognition target 9B is a triple circle and comprises a circular central part 91B and two annular frame parts 92B and 93B that surround the central part 91B in a layered manner. Frame part 92B surrounds the central part 91B along its outer circumference. Frame part 93B surrounds frame part 92B along its outer circumference. The color of frame part 92B is different from the color of the adjacent central part 91B and frame part 93B. Furthermore, the combination of the color of the central part 91B, the color of frame part 92B and the color of frame part 93B is associated with each person 7 on a one-to-one basis. That is, when multiple people 7 exist in space F1, the recognition target 9B formed on each of the multiple people 7 has a different combination of the color of the central part 91B, the color of frame part 92B and the color of frame part 93B.

[0062] In this case, the reference pattern Pn is a triple circular figure, as shown in Figure 11B, having the central figure Pn11 and the frame figures Pn12 and Pn13 as reference figures. The central figure Pn11 is circular. The frame figure Pn12 is an annular shape that surrounds the central figure Pn11 along its outer perimeter. The frame figure Pn13 is an annular shape that surrounds the frame figure Pn12 along its outer perimeter. The central figure Pn11, the frame figures Pn12 and Pn13 are coaxial. The shape of the outline of the central figure Pn11, and the shapes of the frame figures Pn12 and Pn13 are all annular.

[0063] Furthermore, the color of the frame shape Pn12 is different from the color of the adjacent center shape Pn11 and the frame shape Pn13. In addition, the combination of the colors of the center shape Pn11, the frame shape Pn12, and the frame shape Pn13 is different for each reference pattern Pn. In other words, the reference pattern Pn in Figure 11B is a combination of shapes (a circular center shape Pn11, and annular frame shapes Pn12 and Pn13) and colors (the color of the center shape Pn11, the color of the frame shape Pn12, and the color of the frame shape Pn13).

[0064] In the reference pattern Pn shown in Figure 11B, the shape of the outline of the central figure Pn11, and the shapes of the frame figures Pn12 and Pn13 are rotationally symmetric. Furthermore, the shape of the outline of the central figure Pn11, and the shapes of the frame figures Pn12 and Pn13 are annular, and the central figure Pn11 and the frame figures Pn12 and Pn13 are arranged coaxially. In addition, the colors of the central figure Pn11 and the frame figures Pn12 and Pn13 are colors in which the mixing amounts of three colors C1, C2, and C3 are individually set to their maximum or minimum values ​​(see Figure 6).

[0065] (5) Second variation The recognition target 9 may also be the recognition target 9C shown in Figure 12A. The recognition target 9C is a triple regular pentagon, comprising a central regular pentagon 91C and two regular pentagonal frame parts 92C and 93C that layerly surround the central regular pentagon 91C. Frame part 92C surrounds the central regular 91C along its outer periphery. Frame part 93C surrounds frame part 92C along its outer periphery. The color of frame part 92C is different from the color of the adjacent central regular 91C and frame part 93C. Furthermore, the combination of the color of the central regular 91C, the color of frame part 92C and the color of frame part 93C is associated one-to-one with person 7. That is, when multiple people 7 exist in space F1, the recognition target 9C formed on each of the multiple people 7's hats 8 will have different combinations of the color of the central regular 91C, the color of frame part 92C and the color of frame part 93C.

[0066] In this case, the reference pattern Pn is a triple regular pentagonal figure, as shown in Figure 12B, with the central figure Pn21 and the frame figures Pn22 and Pn23 each serving as reference figures. The central figure Pn21 is a regular pentagon. The frame figure Pn22 is a regular pentagonal frame that surrounds the central figure Pn21 along its outer perimeter. The frame figure Pn23 is a regular pentagonal frame that surrounds the frame figure Pn22 along its outer perimeter. The central figure Pn21, the frame figures Pn22 and Pn23 are coaxial. That is, the shape of the outline of the central figure Pn21, and the shapes of the frame figures Pn22 and Pn23 are each regular pentagons (polygons).

[0067] Furthermore, the color of the frame shape Pn22 is different from the color of the adjacent central shape Pn21 and the frame shape Pn23. In addition, the combination of the colors of the central shape Pn21, the frame shape Pn22, and the frame shape Pn23 is different for each reference pattern Pn. That is, the reference pattern Pn in Figure 12B is a combination of shapes (a regular pentagonal central shape Pn21, and regular pentagonal frame shapes Pn22 and Pn23) and colors (the color of the central shape Pn21, the color of the frame shape Pn22, and the color of the frame shape Pn23).

[0068] In the reference pattern Pn shown in Figure 12B, the shape of the outline of the central figure Pn21, and the shapes of the frame figures Pn22 and Pn23 are rotationally symmetric. Furthermore, the shape of the outline of the central figure Pn21, and the shapes of the frame figures Pn22 and Pn23 are regular pentagonal frames, and the central figure Pn21 and the frame figures Pn22 and Pn23 are arranged coaxially. In addition, the colors of the central figure Pn21 and the frame figures Pn22 and Pn23 are colors in which the mixing amounts of three colors C1, C2, and C3 are individually set to their maximum or minimum values ​​(see Figure 6).

[0069] (6) Third variation The image recognition unit 22 is not limited to a configuration that uses a learning model. For example, the image recognition unit 22 may determine the similarity between the detected pattern and each of the multiple reference patterns Pn by pattern recognition processing.

[0070] The shape of the recognition target 9 is not limited to the shapes shown in Figures 4, 11A, and 12A (circular, regular pentagon). The shape of the recognition target 9 may be, for example, an ellipse, a polygon other than a regular pentagon, or an irregular shape.

[0071] The recognition object 9 may be provided on an object other than a person. For example, in a warehouse, factory, or store, the recognition object 9 may be provided on a product, inventory, or product for the purpose of managing products, inventory, or goods.

[0072] The functions of the image processing system 2 may be distributed among multiple devices, so that multiple devices constitute the image processing system 2. For example, the image processing system 2 may be implemented using cloud computing technology.

[0073] Furthermore, the image processing system 2 may consist of a single device, such as a personal computer.

[0074] (7) Summary The first embodiment of the image processing system (2) comprises an image acquisition unit (21) that acquires data of an image (G1) in which a recognition target (9) has been captured, and an image recognition unit (22) that performs image recognition processing on the image (G1). The image (G1) has a target region (R9) in which the recognition target (9) is captured. The image recognition unit (22) determines the similarity between a detection pattern, which is a combination of color and shape detected from the target region (R9), and a reference pattern (Pn), which is a predetermined combination of color and shape, and recognizes the recognition target (9) captured in the target region (R9) based on the similarity. A reference pattern (Pn) has as its reference shapes a central shape (Pn1, Pn11, Pn21) and at least one frame shape (Pn2, Pn12, Pn13, Pn22, Pn23) that surrounds the central shape (Pn1, Pn11, Pn21) in layers, and the color of each of the multiple reference shapes (Pn1, Pn11, Pn21, Pn2, Pn12, Pn13, Pn22, Pn23) is different from the color of the reference image of the multiple reference shapes (Pn1, Pn11, Pn21, Pn2, Pn12, Pn13, Pn22, Pn23) that is adjacent to that reference image.

[0075] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0076] In the second embodiment of the image processing system (2) according to the embodiment, in the first embodiment, it is preferable that the shape of the contour of the central figure (Pn1, Pn11, Pn21) and the shape of at least one frame figure (Pn2, Pn12, Pn13, Pn22, Pn23) are similar to each other.

[0077] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0078] In the third embodiment of the image processing system (2) according to the embodiment, it is preferable that the shape of the contour of the central figure (Pn1, Pn11, Pn21) and the shape of at least one frame figure (Pn2, Pn12, Pn13, Pn22, Pn23) are rotationally symmetric, in the first or second embodiment.

[0079] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0080] In the fourth embodiment of the image processing system (2) according to the embodiment, in any one of the first to third embodiments, it is preferable that the shape of the outline of the central figure (Pn1, Pn11) and the shape of at least one frame figure (Pn2, Pn12, Pn13) are annular. Furthermore, the central figure (Pn1, Pn11) and at least one frame figure (Pn2, Pn12, Pn13) are arranged coaxially.

[0081] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0082] In the fifth embodiment of the image processing system (2) according to the embodiment, in any one of the first to third embodiments, it is preferable that the shape of the outline of the central figure (Pn21) and the shape of at least one frame figure (Pn22, Pn23) are polygonal. Furthermore, the central figure (Pn21) and at least one frame figure (Pn22, Pn23) are arranged coaxially.

[0083] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0084] In the sixth embodiment of the image processing system (2), it is preferable that there is only one frame figure (Pn2) in any one of the first to fifth embodiments.

[0085] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0086] In the seventh embodiment of the image processing system (2) according to the embodiment, in any one of the first to sixth embodiments, the color of each of the multiple reference figures (Pn1, Pn11, Pn21, Pn2, Pn12, Pn13, Pn22, Pn23) is preferably a color in which the mixing amount of each of the multiple colors (C1, C2, C3) is individually set to the maximum value (C1m, C2m, C3m) or the minimum value (0).

[0087] The image processing system (2) described above can accurately recognize the object to be recognized (9).

[0088] In the eighth embodiment of the image processing system (2) according to the embodiment, in any one of the first to seventh embodiments, it is preferable that the image recognition unit (22) inputs the data of the captured image (G1) into a learning model constructed by machine learning and obtains the recognition result of the recognition target (9) from the learning model.

[0089] The image processing system (2) described above can accurately recognize the recognition target (9) even when the imaging environment changes.

[0090] The object recognition system (1) according to the ninth embodiment comprises an image processing system (2) according to any one of the first to eighth embodiments, and an imaging device (3) that captures a recognition object (9) and generates data of an captured image (G1).

[0091] The aforementioned object recognition system (1) can accurately recognize the object to be recognized (9).

[0092] In the tenth embodiment of the object recognition system (1) according to the embodiment, in the ninth embodiment, it is preferable that the imaging device (3) is positioned higher than the object to be recognized (9).

[0093] The aforementioned object recognition system (1) makes it easier to capture images of moving objects to be recognized (9).

[0094] In the 11th embodiment of the object recognition system (1), in the 9th or 10th embodiment, the object to be recognized (9) is preferably provided on a hat (8) worn on the head of a person (7).

[0095] The aforementioned object recognition system (1) makes it easier to identify people (7).

[0096] In the twelfth embodiment of the object recognition system (1), in any one of the ninth to eleventh embodiments, it is preferable that the recognition result of the recognition object (9) by the image recognition unit (22) is used for analyzing the movement of a person (7) that moves with the recognition object (9).

[0097] The aforementioned object recognition system (1) can detect the movement of a person (7).

[0098] An image processing method according to a thirteenth embodiment includes an image acquisition step (S1) of acquiring data from an image (G1) in which a recognition target (9) is captured, and an image recognition step (S2) of performing image recognition processing on the image (G1). The image (G1) has a target region (R9) in which the recognition target (9) is captured. The image recognition step (S2) determines the similarity between a detection pattern, which is a combination of color and shape detected from the target region (R9), and a reference pattern (Pn), which is a predetermined combination of color and shape, and recognizes the recognition target (9) captured in the target region (R9) based on the similarity. A reference pattern (Pn) has as its reference shapes a central shape (Pn1, Pn11, Pn21) and at least one frame shape (Pn2, Pn12, Pn13, Pn22, Pn23) that surrounds the central shape (Pn1, Pn11, Pn21) in layers, and the color of each of the multiple reference shapes (Pn1, Pn11, Pn21, Pn2, Pn12, Pn13, Pn22, Pn23) is different from the color of the reference image of the multiple reference shapes (Pn1, Pn11, Pn21, Pn2, Pn12, Pn13, Pn22, Pn23) that is adjacent to that reference image.

[0099] The image processing method described above can accurately recognize the object to be recognized (9).

[0100] The program of the 14th embodiment causes a computer system to execute the image processing method of the 13th embodiment.

[0101] The program described above can accurately recognize the object to be recognized (9). [Explanation of Symbols]

[0102] 1 2 Image Processing System 21 Image acquisition unit 22 Image Recognition Unit 3. Imaging device 7 people 8 hat 9 (9A, 9B, 9C) Recognition target G1 Image R9 Target Area Pn Reference Pattern Pn1, Pn11, Pn21: Central figures (reference figures) Pn2, Pn12, Pn13, Pn22, Pn23 Frame shapes (reference shapes) C1, C2, C3 colors C1m, C2m, C3m maximum value S1 Image acquisition step S2 Image Recognition Step

Claims

1. An image acquisition unit that acquires data from an image of the object to be recognized, The system includes an image recognition unit that performs image recognition processing on the captured image, The object of recognition is located on a hat worn on a person's head. The captured image has a target region in which the recognition target is captured. The image recognition unit determines the similarity between a detection pattern, which is a combination of colors and shapes detected from the target area, and a reference pattern, which is a predetermined combination of colors and shapes, and recognizes the object to be recognized in the target area based on the similarity. The aforementioned reference pattern has a central figure and at least one frame figure that surrounds the central figure in layers, each of which has a different color from the color of the adjacent reference figures among the plurality of reference figures. The image recognition unit inputs the data of the captured image into a learning model constructed by machine learning, and obtains the recognition result of the object to be recognized from the learning model. The training image of the learning model is created from the captured images obtained by placing the same reference pattern on the hats of multiple people and capturing each of the multiple people. Image processing system.

2. The shape of the outline of the central figure and the shape of the at least one frame figure are similar to each other. The image processing system according to claim 1.

3. The shape of the outline of the central figure and the shape of the at least one frame figure are rotationally symmetric. The image processing system according to claim 1 or 2.

4. The shape of the outline of the central figure and the shape of the at least one frame figure are annular. The central figure and the at least one frame figure are arranged coaxially. An image processing system according to any one of claims 1 to 3.

5. The shape of the outline of the central figure and the shape of the at least one frame figure are polygonal. The central figure and the at least one frame figure are arranged coaxially. An image processing system according to any one of claims 1 to 3.

6. There is only one frame shape. An image processing system according to any one of claims 1 to 5.

7. The color of each of the aforementioned reference shapes is a color in which the mixing amount of each of the multiple colors is individually set to the maximum or minimum value. An image processing system according to any one of claims 1 to 6.

8. An image processing system according to any one of claims 1 to 7, The system includes an imaging device that captures the recognition target and generates data of the captured image. A target recognition system.

9. The imaging device is positioned higher than the object to be recognized. The object recognition system of claim 8.

10. The recognition result of the image recognition unit for the recognized object is used for analyzing the movement of a person moving with the recognized object. The object recognition system according to claim 8 or 9.

11. An image processing method performed by a computer system, An image acquisition step to obtain data from an image of the object to be recognized, The process includes an image recognition step of applying image recognition processing to the captured image, The object of recognition is located on a hat worn on a person's head. The captured image has a target region in which the recognition target is captured. The image recognition step involves determining the similarity between a detection pattern, which is a combination of colors and shapes detected from the target area, and a reference pattern, which is a predetermined combination of colors and shapes, and recognizing the object to be recognized in the target area based on the similarity. The aforementioned reference pattern has a central figure and at least one frame figure that surrounds the central figure in layers, each of which has a different color from the color of the adjacent reference figures among the plurality of reference figures. In the image recognition step, the data of the captured image is input to a learning model constructed by machine learning, and the recognition result of the object to be recognized is obtained from the learning model. The training image of the learning model is created from the captured images obtained by placing the same reference pattern on the hats of multiple people and capturing each of the multiple people. Image processing methods.

12. Causes a computer system to execute the image processing method described in Claim 11. program.

Images