Teacher image data creation device, machine learning program creation device and management device

Abstract

To provide a teacher image data creation device capable of increasing accuracy of object detection and reducing a load of data creation.SOLUTION: A teacher image data creation device includes: a storage unit that stores an object image including an object, object image data including a label of the object, and background image data including a plurality of background images; an object image acquisition unit that acquires one object image from the object image data; a background image acquisition unit that acquires one background image from the background image data; an image composition unit that creates a composite image by composing the object image with the acquired background image; and a label assignment unit that assigns a label included in the object image to the composite image.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a teacher image data creation device, a machine learning program creation device, and a management device. [Background technology]

[0002] Various tasks are performed by construction machines. One method for monitoring the operation of construction machines is to install a camera at the site and check the operation using the image captured by the camera. Patent Document 1 describes an example of a technology for analyzing the image captured by the camera and extracting an object. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2017-173298 A Summary of the Invention [Problem to be solved by the invention]

[0004] Meanwhile, a method for extracting objects using machine learning has been proposed. By extracting objects and automatically evaluating the objects, the burden on the worker can be reduced. One method of machine learning for extracting objects is to create training data and use the training data to learn. In this case, the training data needs to identify information about the object and create a large number of images in the environment in which it is expected to be used. If there is a large difference between the images in the training data and the images acquired during the object extraction process, the accuracy of detection decreases. In addition, the process of preparing a large number of images of training data places a large burden on the worker.

[0005] The present disclosure has been made in consideration of the above-mentioned problems, and aims to provide a teacher image data creation device, a machine learning program creation device, and a management device that create teacher image data that can improve the accuracy of object detection and reduce the burden of data creation. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, a teacher image data creation device of one embodiment of the present disclosure includes a memory unit that stores object image data including an object image and a label of the object, and background image data including a plurality of background images, an object image acquisition unit that acquires one of the object images from the object image data, a background image acquisition unit that acquires one of the background images from the background image data, an image synthesis unit that creates a synthetic image by synthesizing the object image with the acquired background image, and a label assignment unit that assigns a label contained in the object image to the synthetic image.

[0007] In order to achieve the above-mentioned objective, a machine learning program creation device of one embodiment of the present disclosure includes a memory unit that stores teacher image data including a plurality of teacher images created by the teacher image data creation device described above, a machine learning unit that learns images including an object and creates a trained learning model that processes the object from the image, and a learning processing unit that trains the machine learning unit from the teacher image data.

[0008] In order to achieve the above-mentioned objective, a management device of one embodiment of the present disclosure includes a memory unit that stores a trained learning model created by the machine learning program creation device described above, and an image analysis unit that processes images stored by a camera using the trained learning model and analyzes objects contained in the images. Effect of the Invention

[0009] According to the present disclosure, teacher image data is created that can improve the accuracy of object detection and reduce the burden of data creation. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing a schematic configuration of a management system according to the present embodiment. [Diagram 2] FIG. 2 is a flowchart showing an example of processing of the teacher image data creating device of this embodiment. [Diagram 3]FIG. 3 is an explanatory diagram for explaining an example of a method for creating teacher image data. [Figure 4] FIG. 4 is a flowchart showing an example of the learning process of this embodiment. [Diagram 5] FIG. 5 is a flowchart showing an example of the object analysis process of this embodiment. [Figure 6] FIG. 6 is a flowchart showing an example of image processing according to this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] The present invention will be described in detail below with reference to the drawings. Note that the present invention is not limited to the form for carrying out the present invention (hereinafter, referred to as the embodiment). Furthermore, the components in the following embodiment include those that a person skilled in the art can easily imagine, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the components disclosed in the following embodiment can be appropriately combined.

[0012] FIG. 1 is a block diagram showing a schematic configuration of a management system according to the present embodiment. The management system 10 creates a trained learning model that detects an object from a captured image. The management system 10 also creates teacher image data including a plurality of teacher images used to create the trained learning model. The management system 10 also uses the created trained learning model to analyze a captured image, detect an object, and evaluate the object. The object in this embodiment is a construction machine. Hereinafter, the construction machine of the object is also referred to as equipment. The object is not particularly limited. The construction machine is, for example, a backhoe. The backhoe operates a bucket to dig and transport earth and sand. The backhoe has tracks that contact the ground, and can move on uneven ground by driving the tracks with an engine. In this embodiment, the trained learning model is described as a program. However, the trained learning model is not limited in configuration as long as it satisfies the functions. It may be processed using hardware or software.

[0013] The management system 10 of the present embodiment also analyzes a video, i.e., a series of still images captured at a predetermined frame rate, as the image to evaluate the motion of the object. Note that the management system 10 may also analyze a still image as the image.

[0014] The management system 10 of this embodiment includes a teacher image data creation device 12, a machine learning program creation device 14, a management device 16, and a camera 18.

[0015] The teacher image data creation device 12 creates a teacher image and creates teacher image data (teacher data) including a plurality of teacher images. The teacher image data creation device 12 includes an input unit 20, an output unit 22, a communication unit 24, a calculation unit 26, and a storage unit 28. The input unit 20 is a user interface through which a user gives instructions to the calculation unit 26 and inputs information, and has a terminal for connecting image data to a recording medium provided by the storage unit 28. The input unit 20 is composed of, for example, a mouse and keyboard, a USB terminal, etc. The output unit 22 is a device that outputs information from the teacher image data creation device 12, and is, for example, a display, a printer, etc.

[0016] The communication unit 24 communicates with an external communication device and the machine learning program creation device 14 to exchange various data. The communication path of the communication unit 24 may be wired, wireless, or a combination of wired and wireless. In addition, a part or all of the communication path may use a public communication line such as the Internet, or a part or all of the communication path may be provided exclusively for the embodiment. The communication unit 24 acquires images that are to be used as material for the teacher image from an external communication device in accordance with an instruction from the calculation unit 26. In addition, the communication unit 24 transmits the created teacher image data to the machine learning program creation device 14 in accordance with an instruction from the calculation unit 26.

[0017] The calculation unit 26 includes a calculation device such as a central processing unit (CPU) or a graphics processing unit (GPU). The calculation unit 26 reads out and executes software programs and data stored in the storage unit 28. Hereinafter, the term "program" refers to software programs read out by the calculation unit 26 and data referenced when executing the software programs. The calculation unit 26 includes an equipment image acquisition unit 30, a background image acquisition unit 32, an image synthesis unit 34, an image processing unit 36, and a labeling unit 38. The functions of each unit will be described after the storage unit 28.

[0018] The storage unit 28 includes a secondary storage device that stores a program. Specifically, as shown in FIG. 1, the storage unit 28 stores device image data 40, background image data 42, teacher image data 44, an image creation program 46, and a processing condition table 48. Specific configuration examples of such secondary storage devices include hard disk drives, solid state drives, flash memories, and other non-volatile storage devices. The storage unit 28 of the embodiment includes one or more of these. The storage unit 28 also includes a primary storage device that functions as a RAM (Random access memory) used as a storage area when the calculation unit 26 executes the program. Specific configuration examples of such temporary storage devices include a semiconductor memory that functions as a DRAM (Dynamic Random Access Memory).

[0019] The equipment image data 40 stores a plurality of equipment images including equipment as the target object. The equipment images are associated with information on the position of the construction equipment on the image and information on the area to be displayed as labels. The equipment images are also associated with information on the processing performed by the construction equipment in a plurality of consecutive images, such as information on excavation, movement, rotation, and the presence or absence of release of the contents of the bucket, as labels. The equipment images may be images consisting of only elements of the construction equipment. In other words, the equipment images may be composed of pixels of the same color or transparent pixels for parts other than the construction equipment. The equipment images may also include information that can separate the construction equipment from others by associating information on the display area of ​​the construction equipment. The equipment images are preferably created by simulation, but may also be created by processing actual images. The teacher image data creation device 12 acquires the equipment images via the input unit 20 and the communication unit 24 and stores them in the storage unit. The teacher image data creation device 12 may also acquire information on the elements of the images and process them in an image processing unit to create the equipment images.

[0020] The background image data 42 includes a plurality of background images which are images of the locations where the construction machine performs work. For example, the background image is an image of a quarry when the construction machine is used in a quarry. Here, the background image is preferably an image acquired at the target location to be analyzed by the management system 10, but an image created by simulation or an image of various locations with a similar environment may be used. The teacher image data creation device 12 acquires the background images via the input unit 20 and the communication unit 24, and stores them in the storage unit. The teacher image data creation device 12 may also acquire information on the elements of the image, and process it in an image processing unit to create a background image.

[0021] The teacher image data 44 is a plurality of teacher images created by the teacher image data creation device 12. The teacher image is an image in which an equipment image is composited with a background image. The teacher image is associated with a label of the object (construction machinery) included in the composite image. The label is information indicating the action being performed by the object, the type of object, etc. In other words, the label is classification information used in machine learning and information on the position of a bounding box.

[0022] The image creation program 46 is a program that, when executed, realizes the functions of each of the equipment image acquisition unit 30, the background image acquisition unit 32, the image synthesis unit 34, the image processing unit 36, and the label assignment unit 38. The processing condition table 48 is the conditions for execution of the equipment image acquisition unit 30, the background image acquisition unit 32, the image synthesis unit 34, the image processing unit 36, and the label assignment unit 38, in other words, the conditions for creating teacher image data.

[0023] The detailed functions of the calculation unit 26 will be described. The equipment image acquisition unit 30 acquires equipment images from external equipment, etc., and stores them in the equipment image data 40. The equipment image acquisition unit 30 also selects an equipment image from the equipment image data 40 when creating teacher image data. The equipment image acquisition unit 30 sends the selected equipment image to the image synthesis unit 34.

[0024] The background image acquisition unit 32 acquires a device image from an external device or the like and stores it in the background image data 42. In addition, the background image acquisition unit 32 selects a background image from the background image data 42 when creating teacher image data. The background image acquisition unit 32 sends the selected background image to the image synthesis unit 34.

[0025] The image synthesis unit 34 synthesizes the device image with the background image to create a synthetic image in which the object is included in the background image. The synthetic image becomes the teacher image. The method of synthesizing the device image with the background image is not particularly limited, and may involve extracting an object element from the device image and overlaying it on the background image, or making the device image an image in which the parts other than the object are transparent and overlaying the device image on the background image. The image synthesis unit 34 saves the synthetic image in the teacher image data 44.

[0026] The image processing unit 36 ​​processes the device image, background image, or composite image. Based on the settings, the image processing unit 36 ​​executes various processes on the image when extracting an object from the image. Examples of image processes include black and white (grayscale) conversion, binarization, edge extraction, shading, etc.

[0027] The labeling unit 38 assigns a label to the composite image. Here, the labeling unit 38 creates label information to be assigned based on information included in the device image and the position where the device image is composited into the background image. The label is information indicating the action being performed by the object, the type of the object, etc. Note that the label may include position information of the object in the device image.

[0028] Next, the machine learning program creation device 14 executes learning with the teacher image data, extracts objects contained in the images, and creates a trained learning model that evaluates the objects. The machine learning program creation device 14 includes an input unit 50, an output unit 52, a communication unit 54, a calculation unit 56, and a storage unit 58. The input unit 50 is a user interface through which a user gives instructions to the calculation unit 56 and inputs information, and has a terminal that connects the image data to a recording medium provided by the storage unit 58. The input unit 50 is composed of, for example, a mouse and a keyboard, a USB terminal, etc. The output unit 52 is a device that outputs information from the machine learning program creation device 14, such as a display, a printer, etc.

[0029] The communication unit 54 communicates with the teacher image data creation device 12 and the management device 16 to exchange various data. The communication path of the communication unit 54 may be wired, wireless, or a mixture of wired and wireless. In addition, a part or all of the communication path may use a public communication line such as the Internet, or a part or all of the communication path may be provided exclusively for the embodiment. The communication unit 54 acquires teacher image data from the teacher image data creation device 12 according to an instruction from the calculation unit 56. In addition, the communication unit 54 transmits the created trained learning model 74 to the management device 16 according to an instruction from the calculation unit 56. Note that the teacher image data and the trained learning model 74 may exchange data using a recording medium.

[0030] The calculation unit 56 includes a calculation device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The calculation unit 56 reads out and executes software programs and data stored in the storage unit 58. Hereinafter, the term "program" refers to software programs read out by the calculation unit 56 and data referenced when executing the software programs. The calculation unit 56 includes a learning processing unit 60 and a machine learning unit 62.

[0031] The learning processing unit 60 performs a learning process for the machine learning unit 62 using the teacher image data 70. In other words, it creates a learned learning model. The machine learning unit 62 is a processing unit that executes deep learning. The machine learning unit 62 can use various learning methods to extract objects from images. The machine learning unit 62 can use a model that combines a convolutional neural network (CNN) and a recurrent neural network (RNN) as the learning model. It is preferable to use a long short-term memory (LSTM) as the recurrent neural network (RNN). Note that the learning model can use various models that extract objects from images.

[0032] The storage unit 58 includes a secondary storage device that stores a program. Specifically, as shown in FIG. 1, the storage unit 58 stores teacher image data 70, a learning control program 72, and a trained learning model 74. Specific configuration examples of such secondary storage devices include hard disk drives, solid state drives, flash memories, and other non-volatile storage devices. The storage unit 58 of the embodiment includes one or more of these. The storage unit 58 also includes a primary storage device that functions as a RAM (Random access memory) used as a storage area when the program is executed by the calculation unit 56. Specific configuration examples of such temporary storage devices include a semiconductor memory that functions as a DRAM (Dynamic Random Access Memory).

[0033] The teacher image data 70 is the same data as the teacher image data 44. It is a plurality of teacher images created by the machine learning program creation device 14. The learning control program 72 is a program that, when executed, realizes the functions of each of the learning processing unit 60 and the machine learning unit 62. The trained learning model 74 is a program created by executing learning into a learning model using teacher image data in the machine learning unit 62. The trained learning model 74 is a program that inputs an image and outputs an evaluation of an object. The evaluation of an object in this embodiment is the presence or absence of an object and the action being performed by the object.

[0034] The management device 16 evaluates the object in the image captured by the camera using the trained learning model. The management device 16 includes a calculation unit 80 and a storage unit 82. Although not described here, the monitoring device 16 includes an input unit, an output unit, a communication unit, and the like, like other devices.

[0035] The calculation unit 80 includes a calculation device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The calculation unit 80 reads out and executes software programs and data stored in the storage unit 82. Hereinafter, the term "program" refers to software programs read out by the calculation unit 80 and data referenced when executing the software programs. The calculation unit 80 includes an image acquisition unit 90 and an image analysis unit 92.

[0036] The image acquisition unit 90 acquires images from the camera 18. The image analysis unit 92 inputs the images acquired by the image acquisition unit 90 into a trained learning model and acquires the presence or absence of an object and the motion of the object as output. The image analysis unit 92 outputs an evaluation result based on the acquired presence or absence of an object and the motion of the object. As the evaluation result, the type of motion performed by the construction machine may be output, or information on the motions performed by the construction machine may be accumulated and the operating status of the construction machine may be output.

[0037] The storage unit 82 includes a secondary storage device that stores a program. Specifically, as shown in FIG. 1, the storage unit 82 stores an action extraction program 102 and a trained learning model 104. Specific configuration examples of such secondary storage devices include hard disk drives, solid state drives, flash memories, and other non-volatile storage devices. The storage unit 82 of the embodiment includes one or more of these. The storage unit 82 also includes a primary storage device that functions as a RAM (Random access memory) used as a storage area when the calculation unit 80 executes a program. Specific configuration examples of such temporary storage devices include a semiconductor memory that functions as a DRAM (Dynamic Random Access Memory).

[0038] The action extraction program 102 is a program that, when executed, realizes the functions of the image acquisition unit 90 and the image analysis unit 92. The trained learning model 104 is a program created by the machine learning program creation device 14. The trained learning model 104 is a program that inputs an image and outputs an evaluation of an object. In this embodiment, the evaluation of an object is the presence or absence of an object and the action being performed by the object.

[0039] The camera 20 captures the target area where the construction machine is working. The camera 20 can capture images continuously. In other words, the camera 20 can capture video. The camera 20 can also rotate left and right (pan), rotate up and down (tilt), enlarge (zoom in) and reduce (zoom out) the captured image. The camera 20 may be placed on a dolly that is placed at the construction site and made movable. The camera 20 outputs the captured images to the management device 16. The management device 16 may also acquire images from multiple cameras.

[0040] Next, the processing of the management system 10 will be described with reference to Fig. 2 to Fig. 6. First, the processing operation of the teacher image data creation device 12 will be described with reference to Fig. 2 and Fig. 3. Fig. 2 is a flowchart showing an example of the processing of the teacher image data creation device of this embodiment.

[0041] The teacher image data creation device 12 (hereinafter also referred to as the data creation device 12) selects an equipment image (step S12). Specifically, the equipment image acquisition unit 30 selects one equipment image from the equipment image data 40. Next, the data creation device 12 selects a background image (step S14). Specifically, the background image acquisition unit 32 selects one background image from the background image data 42.

[0042] Next, the data creating device 12 creates a composite image by combining the selected device (object) with the background (step S16). Specifically, the composite image is created by superimposing the object of the device image on the background image.

[0043] Next, the data creating device 12 assigns a label associated with the equipment image to the composite image (composite image) and stores it (step S18). The data creating device 12 assigns a label associated with the equipment image to the composite image. Note that a label may be assigned to an image that has been subjected to image processing on the composite image. The data creating device 12 also stores the composite image to which the created label has been assigned in the teacher image data 44 as one of the teacher images.

[0044] Next, the data creation device 12 judges whether to end the creation of the composite image (step S20). Here, the criteria for judging the end can be set in various ways. For example, there is a criterion for judging the end of the process when a set number of composite images have been created, or when composite images of all combinations of equipment and background have been created. If the data creation device 12 judges that the creation is not complete (No in step S20), it returns to step S12 and creates another composite image. Here, the data creation device 12 is not limited to creating another composite image by changing the combination of the equipment image and the background image, and may create another composite image by changing the position and size of the equipment image to be combined with the background image. If the data creation device 12 judges that the creation is complete (Yes in step S20), it ends this process.

[0045] 3 is an explanatory diagram for explaining an example of a method for creating teacher image data. By performing the above processing, the data creation device 12 composites the device image 202 and the background image 204 to create a composite image 206. By overlaying only the image of the object in the device image 202 on the background image 204, a composite image 206 in which only the object is layered on the background can be created.

[0046] Next, FIG. 4 is a flowchart showing an example of the learning process of this embodiment. FIG. 2 is a flowchart showing an example of the process of the machine learning program creation device 14 (hereinafter also referred to as the program creation device 14) of this embodiment. The program creation device 14 reads teacher image data (step S32). The program creation device 14 specifies images from the teacher image data (step S34). In other words, the program creation device 14 specifies teacher images to be used for learning. Here, the program creation device 14, for example, sets a predetermined number of teacher images from the teacher image data as learning data, and sets a predetermined number of teacher images from the teacher image data as verification data. For example, when the learning data is 7, the verification data is set to 3.

[0047] The program creation device 14 uses the identified teacher image to learn the learning model (step S36). The program creation device 14 creates a trained learning model by learning the learning model using the teacher image of the learning data, and inputs an image from which the label of the teacher image of the verification data has been deleted into the trained learning model, and compares the calculated result with the label to evaluate the accuracy.

[0048] The program creation device 14 determines whether the termination condition is satisfied (step S38). If the program creation device 14 determines that the termination condition is not satisfied (No in step S38), for example, if the accuracy does not satisfy the condition, the program creation device 14 returns to step S34 and performs the processes of steps S34 and S36 again. In this case, re-learning may be performed to further learn the current trained learning model, or the combination of teacher images identified from the teacher image data may be changed. If the program creation device 14 determines that the termination condition is satisfied (Yes in step S38), the program creation device 14 saves or registers the trained learning model (step S40) and ends this process.

[0049] Fig. 5 is a flowchart showing an example of an object analysis process according to the present embodiment. Fig. 5 shows a process in which the management device 16 analyzes an image from the camera 18. The management device 16 reads a stored or registered trained learning model (step S52). Next, the management device 16 acquires an image from the camera 18 (step S54).

[0050] Next, the management device 16 performs object extraction using the machine learning unit (step S56). That is, the image analysis unit 92 inputs the image into the trained learning model 104 and obtains the object extraction result as an output. Next, the management device 16 outputs the extraction result (step S58). Here, the extraction result is the identification result of the object and information on the operation being performed by the object.

[0051] The management device 16 determines whether the processing has ended (step S60). If the management device 16 determines that the processing has not ended (No in step S60), it returns to step S54 and executes processing to evaluate the object from the image. If the management device 16 determines that the processing has ended (Yes in step S60), it ends this processing. As a result of the above, the management device 16 can identify the object from the image, identify the behavior of the object, and perform an evaluation. Note that when returning to step S54, image processing may be performed on the image captured by the camera 18.

[0052] Here, the management system 10 may execute image processing before executing the process of extracting an object from an image. FIG. 6 is a flowchart showing an example of image processing in this embodiment. The process in FIG. 6 will be described below as the process of the image processing unit 36. Note that, in the case where image processing is executed before executing the process of extracting an object from an image, the process is executed when creating a teacher image, and when evaluating an image in the management device 16, the image acquisition unit 90 executes the same process as the image processing unit 36. This allows the input images to be under similar conditions.

[0053] The image processing unit 36 ​​acquires an image (step S72). Here, the image in this embodiment is a composite image. The image processing unit 36 ​​converts the image to grayscale (step S74), extracts edges (step S76), and performs binarization (step S78). The image processing unit 36 ​​can convert the image into a black and white line drawing by executing the process of FIG. 6.

[0054] As described above, the teacher image data creation device 12 of this embodiment includes a storage unit that stores an object image including an object and object label, object image data including a plurality of background images, an object image acquisition unit that acquires one of the object images from the object image data, a background image acquisition unit that acquires one of the background images from the background image data, an image synthesis unit that creates a composite image by synthesizing the object image with the acquired background image, and a label assignment unit that assigns a label included in the object image to the composite image. The teacher image data creation device 12 can efficiently create a teacher image by separately preparing an equipment image, which is an image of an object, and a background image, creating a composite image, and creating a teacher image. In addition, by performing learning with the teacher image data of this embodiment, a trained learning model with high detection accuracy can be easily created. In addition, by using this trained learning model, detection can be easily performed with high accuracy.

[0055] By performing tasks on the object in various positions at the actual work site, teacher images that can be obtained, background images are acquired, and multiple object images are combined against multiple background images to create multiple combinations of teacher images. This makes it possible to efficiently create a larger number of comprehensive combinations of teacher images.

[0056] In addition, by using a simulation to create an image of an object, it is possible to create images of the object in various poses more efficiently. Furthermore, the information on the action being performed by the object can be identified with high accuracy through simulation. This allows labels to be assigned with high accuracy and detailed classification.

[0057] By converting the object image into a video, it is possible to create teacher image data and a trained learning model that can more accurately identify the object's movement.

[0058] By setting the target object as construction machinery and the labels as information about the operation of the construction machinery and information about the position of the construction machinery, it is possible to automatically extract the work being performed by the construction machinery, and to evaluate the work with higher accuracy.

[0059] Moreover, by assuming that the background image is a photographed image, the analysis accuracy can be improved. Moreover, the management system 10 may perform image processing on the image so as to use a black and white line drawing. [Explanation of symbols]

[0060] 10 Management Systems 12 Teacher image data creation device 14 Machine learning program creation device 16 Management device 18 Camera 20, 50 Input section 22, 52 Output section 24, 54 Communications Department 26, 56, 80 calculation section 28, 58, 82 storage section 30 Equipment image acquisition section 32 Background image acquisition section 34 Image synthesis unit 36 Image Processing Section 38 Labeling section 40 Equipment image data 42 Background image data 44 Teacher image data 46 Image Creation Programs 48 Processing Condition Table 60 Learning processing unit 62 Machine Learning Department 70 Teacher image data 72 Learning Control Program 74 Pre-trained learning models 90 Image acquisition unit 92 Image Analysis Section 102 Motion Extraction Program 104 Pre-trained learning models

Claims

1. A storage unit that stores object image data including an object image and a label of the object, and background image data including a plurality of background images; an object image acquisition unit that acquires one of the object images from the object image data; a background image acquisition unit that acquires one of the background images from the background image data; an image synthesis unit for generating a synthetic image by synthesizing the object image with the acquired background image; a label assignment unit that assigns a label included in the object image to the composite image, The object image is a video. The object is a construction machine, A teacher image data creation device, wherein the label is information on the operation of the construction machine and information on the position of the construction machine.

2. The teacher image data creating device according to claim 1 , wherein the object image is an image created by simulation.

3. 3. The teacher image data creating device according to claim 1, wherein the background image is a photographed image.

4. A storage unit that stores teacher image data including a plurality of teacher images created by the teacher image data creation device according to any one of claims 1 to 3; a machine learning unit that learns images containing objects and creates a trained model for processing objects from images; A machine learning program creation device including a learning processing unit that causes the machine learning unit to learn the teacher image data.

5. A storage unit that stores a trained learning model created by the machine learning program creation device according to claim 4; A management device including an image analysis unit that processes images stored by a camera using the trained learning model and analyzes objects contained in the images.

Images