Model learning method and program

Abstract

This invention provides a low-cost learning method for creating training data to train an AI model that estimates the type of equipment installed on the ceiling. [Solution] A model learning method performed by a learning system for an equipment estimation device that estimates the type of equipment placed on the ceiling of a space includes: a first acquisition step of acquiring RGB data of the space; an estimation step of using a pre-trained AI model for estimating the type of equipment based on the RGB data of the space acquired in the first acquisition step to estimate the types of multiple equipment included in the RGB data and generate equipment estimation results; a second acquisition step of presenting the types of multiple equipment estimated in the estimation step to the user and acquiring correction instruction information indicating the content of the corrections to the equipment estimation results received from the user; and a learning step of retraining the AI ​​model based on the correction instruction information acquired in the second acquisition step.

Description

Technical Field

[0001] The present disclosure relates to a model learning method and a program.

Background Art

[0002] Patent Document 1 discloses a system that supports a user in identifying the product number of a lighting fixture based on an image obtained by imaging the lighting fixture.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] During work to replace lighting equipment or air conditioning equipment installed on a ceiling, a constructor performs estimation work. In general, in the estimation work, the constructor performs the task of creating a ceiling plan drawing to present to the customer, and the task of creating this ceiling plan drawing incurs costs. The ceiling plan drawing includes information indicating the type of equipment installed on the ceiling.

[0005] In order to reduce worker costs, it is conceivable to use an AI (Artificial Intelligence) model trained to estimate the type of equipment installed on a ceiling based on an image captured targeting the ceiling.

[0006] However, in order to train the AI model in this manner, a large amount of training data needs to be prepared in advance. Generating a training dataset incurs costs, and accordingly, there is a problem that training the model also incurs costs.

[0007] This disclosure aims to provide a model learning method, etc., that can create training data for an AI model used to estimate the type of equipment installed on a ceiling at low cost. [Means for solving the problem]

[0008] One embodiment of the model learning method relating to this disclosure is a model learning method performed by a learning system for an equipment estimation device that estimates the types of equipment arranged on the ceiling of a space, comprising: a first acquisition step of acquiring RGB data of the space; an estimation step of using an AI (Artificial Intelligence) model for estimating the types of equipment, which has been previously trained, to estimate the types of a plurality of equipment included in the RGB data, based on the RGB data of the space acquired in the first acquisition step, and generating an equipment estimation result as a result of the estimation; a second acquisition step of presenting the types of the plurality of equipment estimated in the estimation step to a user and acquiring correction instruction information indicating the content of the correction of the equipment estimation result received from the user; and a learning step of retraining the AI ​​model based on the correction instruction information acquired in the second acquisition step.

[0009] Furthermore, this disclosure can be implemented not only as the model learning method described above, but also as a program that causes a computer to execute the model learning method. Moreover, it can also be implemented as a computer-readable recording medium that stores the program. [Effects of the Invention]

[0010] The model learning method described herein allows for the creation of training data for an AI model used to estimate the type of equipment installed on the ceiling, at a low cost. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 is a block diagram showing the functional configuration of a ceiling plan creation system according to an embodiment. [Figure 2] Figure 2 is a sequence diagram illustrating the operation by which the ceiling plan creation system according to the embodiment creates ceiling plan data. [Figure 3] Figure 3 shows an example of ceiling plan data according to the embodiment. [Figure 4] Figure 4 is a sequence diagram illustrating the operation in which the learning system according to the embodiment retrains the device estimation model. [Figure 5] Figure 5 shows the process of receiving input of modification instruction information according to the embodiment. [Modes for carrying out the invention]

[0012] (Embodiment) The following describes in detail, with reference to drawings, embodiments of the equipment estimation method, ceiling plan data creation method, and program related to this disclosure. The embodiments described below are all preferred examples of this disclosure. The numerical values, components, arrangement and connection configurations of components, steps, and step order shown in the following embodiments are examples and are not intended to limit this disclosure. Furthermore, components in the following embodiments that are not described in an independent claim are described as any component constituting a preferred configuration.

[0013] Please note that each figure is a schematic diagram and not necessarily a strictly accurate representation. Furthermore, in each figure, substantially identical components are denoted by the same reference numerals, and redundant explanations may be omitted or simplified.

[0014] Furthermore, unless otherwise specified, ordinal numbers such as "first," "second," etc., in this disclosure do not refer to the number or order of components, etc., but are used for the purpose of avoiding confusion and distinguishing similar components, etc.

[0015] [Background leading to this disclosure] When performing construction work for replacing lighting equipment or air conditioning equipment in a tenant space or office, estimation work is carried out in advance. For example, estimation work when performing construction to change fluorescent lamps to LEDs (Light Emitting Diodes) can be considered.

[0016] In such estimation work, it is usual for on-site representatives, site supervisors, and the like of electrical contractors who perform construction work handling wiring for lighting and outlets to actually go to the site and conduct on-site surveys.

[0017] In the on-site survey, the site supervisor visually confirms the type, position, and number of devices arranged on the ceiling, and takes notes by hand. Thereafter, estimation work is performed by creating a ceiling layout drawing based on these notes, and a construction proposal is made to the customer (for example, the owner of the tenant space or office). The ceiling layout drawing is a drawing that includes information indicating the type, position, and number of devices arranged on the ceiling.

[0018] The work of creating this ceiling layout drawing is a high-load task for site supervisors. This is because there are many work sites, and the work becomes wasted if the construction order is lost.

[0019] Accordingly, the inventor has conceived a ceiling layout drawing creation system that supports the work of creating such ceiling layout drawings.

[0020] In the ceiling layout drawing creation system according to the embodiment, a trained AI model is used to estimate the type of equipment installed on the ceiling, but training of the AI model requires creating a large amount of training data, which incurs costs.

[0021] Accordingly, the inventor has conceived a method for more efficiently training the AI model.

[0022] [Configuration] First, the configuration of the ceiling layout drawing creation system that executes the device estimation method according to the embodiment will be described. Figure 1 is a block diagram showing the functional configuration of the ceiling layout drawing creation system 10 according to the embodiment.

[0023] The ceiling plan creation system 10 is a system for creating ceiling plan data for an indoor space. An indoor space is, for example, an office or a tenant space. In this specification, "space" means an indoor space. A user of the ceiling plan creation system 10 is, for example, a contractor who performs work related to equipment installed on the ceiling of an office or tenant space.

[0024] The ceiling plan creation system 10 comprises a ceiling plan creation device 20 and a terminal device 50. The ceiling plan creation device 20 and the terminal device 50 are connected to each other so as to be able to communicate via a public network 60.

[0025] The user goes to the indoor space where the ceiling plan data is created, carrying the terminal device 50. By operating the terminal device 50, the user can generate the information necessary for creating the ceiling plan data and verify the ceiling plan data generated by the ceiling plan creation device 20.

[0026] The ceiling plan creation device 20 is a device that creates ceiling plan data. The ceiling plan data is a drawing of the ceiling of an interior space. Specifically, the ceiling plan data includes information showing the outline of the ceiling of the interior space. The ceiling plan data also includes information showing the location and type of equipment installed on the ceiling of the interior space. The ceiling plan creation device 20 comprises a drawing creation device 30 and an equipment estimation device 40. The ceiling plan creation device 20 may be implemented by two server devices, one server device that implements the functions of the drawing creation device 30 and the other server device that implements the functions of the equipment estimation device 40, or it may be implemented by one server device that implements the functions of both the drawing creation device 30 and the equipment estimation device 40.

[0027] The drawing creation device 30 is a device that creates ceiling drawing data and ceiling plan data. The drawing creation device 30 is implemented, for example, by a cloud server. The drawing creation device 30 comprises a communication unit 31, an information processing unit 32, and a storage unit 33.

[0028] The communication unit 31 is a communication circuit (communication module) for the drawing creation device 30 to communicate with the equipment estimation device 40 and the terminal device 50 via the public network 60. The communication performed by the communication unit 31 is, for example, wireless communication. The communication unit 31 may communicate wirelessly with the terminal device 50 and with the equipment estimation device 40 via wired communication. The communication standard for the communication performed by the communication unit 31 is not particularly limited.

[0029] The information processing unit 32 performs information processing for the drawing creation device 30 to create ceiling drawing data and ceiling plan data. The information processing unit 32 is implemented by, for example, a microcomputer, but may also be implemented by a processor or dedicated circuit. The functions of the information processing unit 32 are realized by the execution of a computer program (software) stored in the storage unit 33 by the hardware, such as a microcomputer or processor, that constitutes the information processing unit 32. Specifically, the information processing unit 32 includes an acquisition unit 321, a creation unit 322, and an output unit 323.

[0030] The acquisition unit 321 acquires RGB data and Depth data of the indoor space from the terminal device 50. The RGB data is an RGB image captured with respect to the ceiling surface of the indoor space. The RGB data shows multiple pieces of equipment installed on the ceiling surface of the indoor space. The Depth data is data that includes distance values ​​to the ceiling surface, walls, and equipment of the indoor space. The Depth data may also be three-dimensional point cloud data showing the outline of the indoor space, generated by converting the distance values ​​to the ceiling surface, walls, and equipment of the indoor space into point cloud data.

[0031] The creation unit 322 creates drawing data of the ceiling of the interior space based on the RGB data and depth data of the interior space acquired by the acquisition unit 321. The ceiling drawing data is data that shows the outer shape of the ceiling of the interior space, and the location and number of equipment installed on the ceiling of the interior space.

[0032] Furthermore, the creation unit 322 creates ceiling plan data based on the ceiling drawing data and the equipment estimation results created by the equipment estimation device 40. The ceiling plan data is data that associates the location and type of equipment installed on the ceiling of the room space.

[0033] The output unit 323 outputs the ceiling plan data created by the creation unit 322.

[0034] The memory unit 33 is a storage device that stores information necessary for information processing, such as computer programs executed by the information processing unit 32. The memory unit 33 is implemented, for example, by semiconductor memory. The computer programs stored in the memory unit 33 include a program for creating ceiling drawing data based on RGB data and depth data of the interior space.

[0035] The equipment estimation device 40 is a device that estimates the type of equipment installed on the ceiling. The equipment estimation device 40 is implemented, for example, by a cloud server. The equipment estimation device 40 comprises a communication unit 41, an information processing unit 42, and a storage unit 43. The type of equipment is, for example, if the equipment is lighting equipment, a classification based on the shape of the equipment, such as downlights or base lights. The type of equipment is, for example, if the equipment is air conditioning equipment, exhaust vents and intake vents, etc.

[0036] The communication unit 41 is a communication circuit (communication module) for the equipment estimation device 40 to communicate with the drawing creation device 30 and the terminal device 50 via the public network 60. The communication performed by the communication unit 41 is, for example, wireless communication. The communication unit 41 may communicate wirelessly with the terminal device 50 and with the drawing creation device 30 via wired communication. The communication standard for the communication performed by the communication unit 41 is not particularly limited.

[0037] The information processing unit 42 performs information processing for the equipment estimation device 40 to estimate the type of equipment. The information processing unit 42 is implemented, for example, by a microcomputer, but may also be implemented by a processor or dedicated circuit. The functions of the information processing unit 42 are realized by the execution of a computer program (software) stored in the storage unit 43 by the hardware, such as a microcomputer or processor, that constitutes the information processing unit 42. Specifically, the information processing unit 42 includes an acquisition unit 421, an estimation unit 422, a correction unit 423, an output unit 424, and a learning unit 425.

[0038] The acquisition unit 421 acquires RGB data of the indoor space. For example, the acquisition unit 421 acquires RGB data of the indoor space captured by the terminal device 50.

[0039] The estimation unit 422 estimates the type of equipment included in the RGB data based on the RGB data of the indoor space acquired by the acquisition unit 421. The estimation unit 422 estimates the type of equipment using the trained equipment estimation model 100 stored in the storage unit 43.

[0040] The correction unit 423 corrects the type of equipment estimated by the estimation unit 422. The correction unit 423 corrects the type of equipment using the learned equipment estimation model 100 stored in the storage unit 43. In this embodiment, it is not essential that the equipment estimation device 40 includes the correction unit 423.

[0041] The output unit 424 outputs the equipment estimation result as a result of the equipment type estimation performed by the estimation unit 422.

[0042] The learning unit 425 trains the device estimation model 100. The learning unit 425 trains the device estimation model 100 using supervised learning with training data.

[0043] The memory unit 43 is a storage device that stores information necessary for information processing, such as computer programs executed by the information processing unit 42. The memory unit 43 is implemented, for example, by semiconductor memory. The computer program stored in the memory unit 43 includes a device estimation model 100 for estimating the type of device.

[0044] The device estimation model 100 is a pre-trained AI (Artificial Intelligence) model. The device estimation model 100 is pre-trained by the learning unit 425.

[0045] The method for creating the equipment estimation model 100 is, for example, supervised learning using a training dataset. Specifically, for example, the person setting up the system prepares training data, which is RGB data (image data) captured on the ceiling surface of a room, associating the area in which equipment installed on the ceiling is visible with the type of equipment. The learning unit 425 then uses this training data to machine-learn the equipment estimation model 100. This training data is an example of the first training data. The equipment estimation model 100 is trained to output the area in which equipment is visible and the type of equipment when RGB data captured on the ceiling surface of a room is input. The equipment estimation model 100 trained by the learning unit 425 is stored in the storage unit 43.

[0046] The terminal device 50 is an information terminal used by the user. The user can use the terminal device 50 to check the ceiling plan data of the room space. The terminal device 50 is implemented, for example, by a dedicated terminal of the ceiling plan creation system 10. The terminal device 50 may also be implemented by a general-purpose terminal such as a smartphone or tablet terminal. Specifically, the terminal device 50 comprises a communication unit 51, an information processing unit 52, a storage unit 53, an operation reception unit 54, a display unit 55, and an imaging unit 56.

[0047] The communication unit 51 is a communication circuit (communication module) for the terminal device 50 to communicate with the drawing creation device 30 and the equipment estimation device 40 via a public network 60 such as the Internet. The communication performed by the communication unit 51 is, for example, wireless communication. The communication standard for the communication performed by the communication unit 51 is not particularly limited.

[0048] The information processing unit 52 performs information processing to allow the user to acquire ceiling plan data using the terminal device 50. The information processing unit 52 is implemented by, for example, a microcomputer, but may also be implemented by a processor or dedicated circuit. The functions of the information processing unit 52 are realized by the execution of a computer program (software) stored in the storage unit 53 by the hardware, such as the microcomputer or processor that constitutes the information processing unit 52.

[0049] The memory unit 53 is a storage device that stores computer programs and the like executed by the information processing unit 52. The computer programs stored in the memory unit 53 include a predetermined application program for presenting ceiling plan data to the user. The memory unit 53 is implemented, for example, by semiconductor memory.

[0050] The operation reception unit 54 receives user input. The operation reception unit 54 is implemented, for example, by a touch panel.

[0051] The display unit 55 is a display for presenting images to the user, for example, by presenting the user with an image of the ceiling plan data created by the ceiling plan creation device 20. The display unit 55 is implemented by a display panel such as a liquid crystal panel or an organic EL (Electronic Luminescence) panel.

[0052] The imaging unit 56 captures an image of the indoor space. The user uses the imaging unit 56 of the terminal device 50 to perform imaging targeting the ceiling surface of the indoor space. Specifically, the imaging unit 56 captures an RGB image and depth data of the space. The imaging unit 56 is implemented, for example, by a camera capable of capturing visible light images and LiDAR (Light Detection And Ranging).

[0053] [Action to create ceiling diagram data] The operation of the ceiling plan creation device 20 according to the embodiment in which ceiling plan data is created will be described. Figure 2 is a sequence diagram illustrating the operation of the ceiling plan creation system 10 according to the embodiment in which ceiling plan data is created.

[0054] First, the terminal device 50 captures RGB data of the space (S101). Specifically, the user uses the imaging unit 56 of the terminal device 50 to capture RGB data of the ceiling surface of the room.

[0055] Next, the terminal device 50 captures depth data of the space (S102). The user uses the imaging unit 56 of the terminal device 50 to capture depth data targeting the ceiling surface of the room space.

[0056] When spatial RGB data and depth data are captured, the information processing unit 52 of the terminal device 50 transmits the spatial RGB data and depth data to the drawing creation device 30 via the public network 60 using the communication unit 51 (S103). The spatial RGB data and depth data transmitted by the terminal device 50 are received by the communication unit 31 of the drawing creation device 30.

[0057] The acquisition unit 321 of the drawing creation device 30 acquires spatial RGB data and depth data (S104).

[0058] Once RGB data and depth data are acquired, the creation unit 322 of the drawing creation device 30 creates drawing data for the ceiling (S105). Specifically, the drawing data for the ceiling is drawing data that shows the outer shape of the ceiling of the room space, and the location and number of equipment installed on the ceiling of the room space.

[0059] Furthermore, the information processing unit 52 of the terminal device 50 transmits spatial RGB data to the device estimation device 40 via the public network 60 using the communication unit 51 (S106). The spatial RGB data transmitted by the terminal device 50 is received by the communication unit 41 of the device estimation device 40.

[0060] The acquisition unit 421 of the equipment estimation device 40 acquires spatial RGB data (S107).

[0061] The information processing unit 42 of the equipment estimation device 40 estimates the type of equipment installed on the ceiling of the room space based on the acquired RGB data of the space (S108). For example, the estimation unit 422 identifies the area of ​​equipment in the RGB data and estimates the type of equipment reflected in the RGB data using the trained equipment estimation model 100. In other words, the estimation unit 422 estimates the type of each of the one or more equipment whose areas have been identified.

[0062] Next, the information processing unit 42 of the equipment estimation device 40 transmits the equipment estimation result to the drawing creation device 30 using the communication unit 41 (S109). The equipment estimation result includes information indicating the type of equipment estimated by the equipment estimation device 40 and information indicating the position of the equipment in RGB data. The equipment estimation result transmitted by the equipment estimation device 40 is received by the drawing creation device 30.

[0063] Upon receiving the equipment estimation results, the creation unit 322 of the drawing creation device 30 creates ceiling plan data (S110). Specifically, the drawing creation device 30 creates ceiling plan data by applying the received equipment estimation results to the ceiling drawing data created in step S105. In other words, the creation unit 322 associates the equipment estimation results with the equipment whose positions are indicated in the ceiling drawing data. The ceiling plan data is data that associates the positions and types of equipment installed on the ceiling of an indoor space. Figure 3 shows an example of ceiling plan data 200 according to an embodiment.

[0064] As shown in Figure 3, the ceiling plan data 200 indicates that the ceiling is a rectangle with dimensions of Xm × Ym. It also indicates that nine pieces of equipment are installed on the ceiling, eight of which are base lights and one of which is an air conditioning unit. In other words, the ceiling plan data 200 shown in Figure 3 indicates the location and type of equipment installed on the ceiling of the interior space.

[0065] Once the ceiling plan data is created, the output unit 323 of the drawing creation device 30 outputs the ceiling plan data (S111). The output unit 323 of the drawing creation device 30 transmits the ceiling plan data to the terminal device 50 using the communication unit 31 (S112). The ceiling plan data transmitted by the drawing creation device 30 is received by the terminal device 50.

[0066] Upon receiving the ceiling plan data, the terminal device 50 presents the ceiling plan data to the user (S113). For example, the information processing unit 52 presents the ceiling plan data to the user by displaying it on the display unit 55.

[0067] Thus, the ceiling plan creation system 10 according to this embodiment can create and present ceiling plan data of an indoor space based on RGB data and depth data captured of the indoor space by the user.

[0068] This ceiling plan creation system 10 reduces the effort required for users to draft ceiling plans.

[0069] [The process of retraining the device estimation model] As described above, the ceiling plan creation system 10 can create ceiling plan data for an indoor space. The types of equipment included in the ceiling plan data are those estimated by the equipment estimation model 100. However, the equipment estimation model 100 may make errors. As a result, the ceiling plan data created by the ceiling plan creation system 10 may contain errors.

[0070] To improve the accuracy of the equipment estimation model 100 in estimating equipment types, it is necessary to train it using more training data, but creating training data is costly.

[0071] The learning method according to this embodiment can reduce the cost required to create training data. This learning method will be described below. Figure 4 is a sequence diagram illustrating the operation in which the learning system according to this embodiment retrains the device estimation model 100. The learning system comprises a device estimation device 40 and a terminal device 50.

[0072] First, the ceiling plan creation system 10 presents the created ceiling plan data to the user (S113). The ceiling plan data is displayed on the display unit 55 of the terminal device 50.

[0073] Next, the information processing unit 52 of the terminal device 50 receives a correction from the user regarding the estimated results of the equipment type in the ceiling plan data (S201). For example, the user can correct the equipment type by tapping on the equipment that is displayed with an incorrect type in the ceiling plan data displayed on the display unit 55. Figure 5 shows the process of receiving correction instruction information according to the embodiment.

[0074] As shown in Figure 5, the estimated ceiling plan data 70 is displayed on the display unit 55 of the terminal device 50. The user checks whether there are any errors in the estimated ceiling plan data 70 by comparing it with the actual ceiling of the room. Figure 5 shows how, if there is an error in the estimated result of the type of equipment 71, the user can correct the estimated ceiling plan data 70 by tapping to select the equipment 71.

[0075] In this manner, the information processing unit 52 receives correction instruction information from the user. The correction instruction information is information indicating the content of the correction to the device estimation result received from the user. The correction instruction information includes information indicating the type of device after correction and information indicating the position of the corrected device in the RGB data.

[0076] Next, the information processing unit 52 creates training data from the correction details and RGB data (S202). This training data is an example of second training data. Specifically, the information processing unit 52 creates training data based on the RGB data captured in step S101 of Figure 2 and the correction details included in the correction instruction information received in step S201. More specifically, the information processing unit 52 creates training data by labeling the area in which the equipment is visible in the RGB data captured in step S101 of Figure 2 based on the correction details included in the correction instruction information received in step S201.

[0077] Furthermore, the training data may include information indicating the type of equipment that was not corrected in the processing of step S201. In other words, the training data may include information indicating equipment whose type was correctly estimated by the equipment estimation model 100.

[0078] Once training data is created, the information processing unit 52 transmits the training data to the device estimation device 40 using the communication unit 51 (S203). The training data transmitted by the terminal device 50 is received by the device estimation device 40.

[0079] When training data is received, the learning unit 425 of the equipment estimation device 40 retrains the equipment estimation model 100 using the received training data (S204). Specifically, the learning unit 425 updates the training data by adding the received training data to the training data used for prior training, and then retrains the equipment estimation model 100 using the updated training data.

[0080] This process allows for the creation of training data to be used to retrain the equipment estimation model 100 using the user's correction results. Since the users are workers nationwide, a large amount of training data tailored to the specific field can be created.

[0081] Therefore, this operation allows the learning system to create training data at low cost for training an equipment estimation model 100 to estimate the type of equipment installed on the ceiling. Furthermore, since the equipment estimation model 100 can be retrained using such a large amount of on-site training data, the accuracy of the equipment type estimation results can be improved.

[0082] [Effects, etc.] The technologies derived from the disclosures in this specification include, for example, the following. The technologies derived from the disclosures in this specification will be described below, along with the effects obtained by such technologies.

[0083] Technology 1 is a model learning method performed by a learning system for an equipment estimation device that estimates the types of equipment placed on the ceiling of a space, and includes: a first acquisition step of acquiring RGB data of the space; an estimation step of performing the estimation of multiple equipment types included in the RGB data using a pre-trained AI model for estimating equipment types based on the RGB data of the space acquired in the first acquisition step, and generating equipment estimation results as a result of the estimation; a second acquisition step of presenting the multiple equipment types estimated in the estimation step to the user and acquiring correction instruction information indicating the content of the correction of the equipment estimation results received from the user; and a learning step of retraining the AI ​​model based on the correction instruction information acquired in the second acquisition step.

[0084] This model learning method allows for the retraining of the equipment estimation model 100 using user correction results. Since the users are workers nationwide, a large amount of correction results tailored to the field are obtained. Therefore, with this model learning method, the learning system can create training data for the equipment estimation model 100, which estimates the type of equipment installed on the ceiling, at low cost.

[0085] Technology 2 is a model learning method of Technology 1, in which the AI ​​model is pre-trained by supervised learning using first training data, and in the learning step, second training data is generated based on the spatial RGB data acquired in the first acquisition step and the equipment estimation result corrected based on the correction instruction information, and the AI ​​model is retrained by supervised learning using the first training data and the second training data.

[0086] This model learning method allows for the low-cost creation of training data for training an equipment estimation model 100, which is used to estimate the type of equipment installed on the ceiling. Furthermore, since the equipment estimation model 100 can be retrained using a large amount of training data that is relevant to the actual site, the accuracy of the equipment type estimation results can be improved.

[0087] Technique 3 is a program that causes the learning system to execute the model learning method of Technique 1 or 2.

[0088] Such programs can help create training data at low cost for AI models that estimate the type of equipment installed on ceilings.

[0089] (Other embodiments) Although embodiments have been described above, this disclosure is not limited to the embodiments described above.

[0090] For example, in the above embodiment, the ceiling plan creation system was implemented by multiple devices. When the ceiling plan creation system is implemented by multiple devices in this way, the components of the ceiling plan creation system (especially the functional components) may be distributed among the multiple devices in any manner.

[0091] Furthermore, the ceiling plan creation system may be implemented as a single device. For example, the ceiling plan creation system may be implemented as a single device equivalent to a ceiling plan creation device.

[0092] Furthermore, the method of communication between devices in the above embodiment is not particularly limited. In addition, relay devices (such as gateway devices), which are not shown, may be involved in the communication between devices.

[0093] Furthermore, in the above embodiment, a process executed by a specific processing unit may be executed by another processing unit. Also, the order of multiple processes may be changed, or multiple processes may be executed in parallel.

[0094] Furthermore, in the above embodiment, each component may be realized by executing a software program suitable for each component. Each component may also be realized by a program execution unit such as a CPU or processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory.

[0095] Furthermore, each component may be implemented by hardware. For example, each component may be a circuit (or integrated circuit). These circuits may form a single circuit as a whole, or they may be separate circuits. Also, each of these circuits may be a general-purpose circuit or a dedicated circuit.

[0096] Furthermore, the general or specific embodiments of this disclosure may be implemented as a system, apparatus, method, integrated circuit, computer program, or recording medium such as a computer-readable CD-ROM. They may also be implemented in any combination of systems, apparatus, methods, integrated circuits, computer programs, and recording media.

[0097] Furthermore, this disclosure may be implemented as a ceiling plan creation method executed by a computer, such as the ceiling plan creation system (ceiling plan creation device) of the above embodiment, or as a program (in other words, a computer program product) for causing a computer to execute the ceiling plan creation method. Alternatively, this disclosure may be implemented as a computer-readable non-temporary recording medium on which such a program is recorded.

[0098] Furthermore, this disclosure also includes forms obtained by applying various modifications to each embodiment that a person skilled in the art could conceive, or forms realized by arbitrarily combining the components and functions of each embodiment without departing from the spirit of this disclosure. [Explanation of Symbols]

[0099] 20 Ceiling Plan Creation Device 322 Treasury Section 40 Equipment estimation device 421 Acquisition Department 425 Learning Department 50 Terminal devices 52 Information Processing Section 100 device estimation models 200 Ceiling Plan Data

Claims

1. A model learning method performed by a learning system for an equipment estimation device that estimates the type of equipment placed on the ceiling of a space, A first acquisition step of acquiring RGB data of the aforementioned space, An estimation step is performed to estimate the types of multiple devices included in the RGB data, using an AI (Artificial Intelligence) model for estimating the types of devices that have been previously trained based on the RGB data of the space acquired in the first acquisition step, and to generate a device estimation result as a result of the estimation. A second acquisition step involves presenting the user with the types of the multiple devices estimated in the estimation step, and obtaining correction instruction information from the user indicating the content of the correction to the device estimation result. The process includes a learning step in which the AI ​​model is retrained based on the correction instruction information acquired in the second acquisition step, Model learning methods.

2. The aforementioned AI model has been trained in advance using supervised learning with the first training data. In the learning step, second training data is generated based on the RGB data of the space acquired in the first acquisition step and the equipment estimation results corrected based on the correction instruction information, and the AI ​​model is retrained by supervised learning using the first training data and the second training data. The model learning method according to claim 1.

3. A program for causing the learning system to execute the model learning method described in claim 1 or 2.

Images