WO2019229789A1

WO2019229789A1 - Trained model suggestion system, trained model suggestion method, and program

Info

Publication number: WO2019229789A1
Application number: PCT/JP2018/020288
Authority: WO
Inventors: 俊二菅谷
Original assignee: 株式会社オプティム
Priority date: 2018-05-28
Filing date: 2018-05-28
Publication date: 2019-12-05
Also published as: JPWO2019229789A1; JP7068745B2

Abstract

[Problem] To provide a trained model suggestion system, a trained model suggestion method, and a program which suggest a trained model capable of outputting accurate image analysis results of an image desired to be newly subjected to image analysis, without spending time on learning. [Solution] This trained model suggestion system comprises: a trained model database 23 which associates and stores trained models for image analysis with purposes and environments; a purpose ascertaining module 211 which ascertains the purpose for which an image is to be analyzed; an environment ascertaining module 212 which ascertains the environment in which the image is captured; and a trained model suggestion module 241 which refers to the trained model database 23 and suggests a trained model matching the purpose and the environment.

Description

Learned model proposal system, learned model proposal method, and program

The present invention relates to a learned model proposal system, a learned model, which can propose an existing learned model similar to that by acquiring and using the purpose of image analysis and the environment in which the image is captured. The present invention relates to a proposed method and a program.

There has been proposed a method of providing a mechanism for discriminating who is captured by performing image analysis processing on a person image and automatically categorizing the person image (Patent Document 1).

As a machine learning technique for artificial intelligence to perform image analysis, supervised learning is a well-known technique, and a method for generating a learned model suitable for the purpose has also been proposed (Patent Literature). 2).

JP2015-69580A Patent 6216024

However, when supervised learning is performed for image analysis, a large number of images, typically tens of millions to millions or more, are prepared, the correct teacher data is added to the images, It is necessary to create a learned model by learning a classifier or a neural network for classification. In order to create an optimal learned model for each of a plurality of cameras, it takes time to prepare an image for learning, and it takes a long time for learning.

In response to this problem, the present inventor uses an existing learned model and has high accuracy without spending learning time if the purpose of image analysis and the environment in which the image is captured are compatible. We focused on the possibility of obtaining image analysis results.

The present invention proposes an existing learned model similar to that by acquiring and using the purpose of image analysis and the environment in which the image is captured, and provides accurate image analysis without taking up learning time. It is an object of the present invention to provide a learned model proposal system, a learned model proposal method, and a program capable of obtaining results.

The present invention provides the following solutions.

The invention according to the first feature is
A system that proposes a trained model suitable for image analysis,
A learned model database that stores a learned model for image analysis in association with a purpose and an environment;
Purpose acquisition means for acquiring the purpose of image analysis;
An environment acquisition means for acquiring an environment in which an image for the purpose is captured;
With reference to the learned model database, a learned model proposing means for proposing a learned model suitable for the purpose and the environment;
Provided is a learned model proposal system characterized by comprising:

According to the first aspect of the invention, in a system that proposes a learned model suitable for image analysis, a learned model database that stores a learned model for performing image analysis in association with a purpose and an environment; A purpose acquisition means for acquiring the purpose of image analysis, an environment acquisition means for acquiring an environment in which an image for the purpose is photographed, and learning adapted to the purpose and the environment with reference to the learned model database Learned model proposing means for proposing a completed model.

The invention according to the first feature is a category of the learned model suggestion system, but the learned model suggestion method and program have the same actions and effects.

The invention according to the second feature is a learned model proposal system that is the invention according to the first feature,
The learned model includes a learned classifier trained with predetermined learning data consisting of past images and correct answer data,
The learned model proposal means provides a learned model proposal system that proposes the learned classifier as a learned model.

According to the second aspect of the invention, in the learned model proposal system according to the first aspect of the invention, the learned model is learned by learning with predetermined learning data including past images and correct data. The learned model proposing means proposes the learned classifier as a learned model.

The invention according to the third feature is a learned model proposal system which is the invention according to the second feature,
The learned model is provided with a learned model suggestion system comprising: a classifier type when an image is classified by a classifier; and a conversion method for converting the image into a feature vector.

According to the third aspect of the invention, in the learned model suggestion system according to the second aspect of the invention, the learned model includes a classifier type when an image is classified by a classifier, and an image. A conversion method for converting into a feature vector.

The invention according to the fourth feature is a learned model proposal system which is the invention according to the first feature,
Provided is a learned model suggestion system, wherein the learned model is a learned convolutional neural network learned with predetermined learning data including past images and correct answer data.

According to the invention related to the fourth feature, in the learned model proposal system that is the invention related to the first feature, the learned model learns with predetermined learning data composed of past images and correct data. This is a convolutional neural network.

The invention according to the fifth feature is a learned model proposing system that is an invention according to any one of the second feature to the fourth feature,
Image acquisition means for acquiring an image of an environment to be subjected to the image analysis;
Image comparison means for determining whether or not the acquired image and the image of the predetermined learning data are similar,
When the images are similar, the learned model proposing means proposes the learned model proposing system in which the learned model is proposed.

According to the fifth feature of the invention, in the learned model proposing system that is the invention of any one of the second feature to the fourth feature, an image acquisition unit that acquires an image of the environment to be subjected to the image analysis And the image comparison means for determining whether or not the acquired image and the image of the predetermined learning data are similar, and when the images are similar, the learned model suggestion means, Propose the learned model.

An invention according to a sixth feature is a learned model proposal system that is an invention according to any one of the first feature to the fifth feature,
The environment acquisition means provides a learned model proposal system characterized in that an answer input to a presented question is acquired as data relating to the environment.

According to the sixth aspect of the invention, in the learned model suggestion system according to any one of the first to fifth aspects, the environment acquisition means is input to the presented question Obtain answers as environmental data.

An invention according to a seventh feature is a learned model proposal system that is an invention according to any one of the first feature to the fifth feature,
The environment acquisition means provides a learned model proposal system characterized by acquiring data detected by a sensor or a camera.

According to the seventh aspect of the invention, in the learned model suggestion system according to any one of the first to fifth aspects, the environment acquisition unit acquires data detected by a sensor or a camera To do.

The invention according to the eighth feature is
In a system that proposes a trained model suitable for image analysis,
A learned model database that stores a learned model for image analysis in association with a purpose and an environment;
Obtaining the purpose of image analysis;
Obtaining an environment in which an image for that purpose is taken;
Referring to the learned model database and proposing a learned model suitable for the purpose and the environment;
Provided is a learned model proposal method comprising:

The invention according to the ninth feature is
In a learned model proposal system having a learned model database that stores a learned model for performing image analysis in association with a purpose and an environment,
Obtaining the purpose of image analysis;
Obtaining an environment in which an image for that purpose is taken;
Referring to the learned model database and proposing a learned model suitable for the purpose and the environment;
Provide a program to execute.

According to the present invention, by acquiring and using the purpose of image analysis and the environment in which the image is captured, an existing learned model similar to that is proposed, and the accuracy is high without taking the learning time. It is possible to provide a learned model proposal system, a learned model proposal method, and a program that can obtain an image analysis result.

FIG. 1 is a schematic diagram of a preferred embodiment of the present invention. FIG. 2 is a diagram illustrating the functional blocks of the camera 100 and the computer 200 and the relationship between the functions. FIG. 3 is a flowchart of the learned model proposal process. FIG. 4 is a diagram illustrating the functional blocks of the camera 100 and the computer 200 and the relationship between the functions when performing image comparison. FIG. 5 is a flowchart of learned model proposal processing when image comparison is performed. FIG. 6 is an example of a question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment. FIG. 7 is another example of a question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment. FIG. 8 is an example of a learned model proposal screen. FIG. 9 is an example of the configuration of the learned model database.

Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings. This is merely an example, and the technical scope of the present invention is not limited to this.

[Outline of learned model proposal system]
FIG. 1 is a schematic diagram of a preferred embodiment of the present invention. The outline of the present invention will be described with reference to FIG. The learned model proposal system includes a camera 100, a computer 200, and a communication network 300.

In FIG. 1, the number of cameras 100 is not limited to one and may be plural. Further, the computer 200 is not limited to a real device, and may be a virtual device.

As shown in FIG. 2, the camera 100 includes an imaging unit 10, a control unit 110, a communication unit 120, and a storage unit 130. The computer 200 includes a control unit 210, a communication unit 220, a storage unit 230, and an input / output unit 240, as shown in FIG. The storage unit 230 includes a learned model database 23. The control unit 210 implements the purpose acquisition module 211 and the environment acquisition module 212 in cooperation with the communication unit 220, the storage unit 230, and the input / output unit 240. Also, the input / output unit 240 implements the learned model proposal module 241 in cooperation with the control unit 210 and the storage unit 230. The communication network 300 may be a public communication network such as the Internet or a dedicated communication network, and enables communication between the camera 100 and the computer 200.

The camera 100 is an imaging device including an imaging device such as an imaging element and a lens that can perform data communication with the computer 200, and captures an image to be analyzed. Here, a WEB camera is illustrated as an example, but an imaging apparatus having necessary functions such as a digital camera, a digital video, a camera installed in a drone, a wearable device camera, a security camera, an in-vehicle camera, and a 360-degree camera. It may be. The captured image may be stored in the storage unit 130. Further, the camera 100 may be a stereo camera, in which case the distance to the subject group can be measured. Further, the camera 100 may be provided with a light intensity sensor, and in that case, the ambient light intensity can be measured.

The computer 200 is a computing device capable of data communication with the camera 100. Here, a desktop computer is illustrated as an example, but in addition to a mobile phone, a portable information terminal, a tablet terminal, a personal computer, electrical appliances such as a netbook terminal, a slate terminal, an electronic book terminal, and a portable music player Or a wearable terminal such as a smart glass or a head-mounted display.

1, first, a plurality of learned models are stored in the learned model database 23 of the computer 200 (step S01). The learned model may be acquired from another computer or a storage medium, or may be created by the computer 200. Further, this step S01 can be omitted when a sufficiently learned model is already stored in the learned model database 23.

FIG. 9 shows an example of the configuration of the learned model database. In the present invention, the learned model refers to predetermined learning data (teacher data) composed of past images and correct data, a learned classifier learned from the predetermined learning data, and a learned convolutional neural network ( And machine learning methods such as Convolutional (Neural Network: CNN). When there is a conversion method for converting an image into a feature vector, the conversion method is also included in the learned model together with the machine learning method. In the learned model database 23, for each learned model, the purpose of image analysis and the environment in which the image was captured are stored in association with each other. Here, for the purpose of image analysis, for example, detecting the entry of a suspicious person (suspicious person detection), appropriately detecting the harvest time of crops (crop detection), and detecting the occurrence of pests ( Pest detection) is considered. In addition, regarding the environment in which an image is taken, conditions such as location, size, camera position, lighting, etc. can be considered. For example, for location, whether indoors, outdoors (city), outdoors (farm), for area, how many square meters or how ha, for camera position, in the corner of the ceiling, in the center of the ceiling, on the desk / shelf As for the top, the outside of the building, the utility pole, the drone, the lighting, it is possible to select whether it is a fluorescent lamp, an LED, natural light, or not, etc. By using the purpose of the image analysis and the information of the environment where the image is captured, it is possible to propose a learned model having the same purpose and environment.

Returning to FIG. 1, the purpose acquisition module 211 of the computer 200 acquires the purpose of what the image analysis is desired to be performed for (Step S02). Here, as a method of acquiring the purpose, as shown in FIG. 1, the purpose may be transmitted from the camera 100 and acquired, or the user may input using the input / output unit 240 of the computer 200. Or may be obtained by causing the user to input via another terminal (not shown).

Next, the environment acquisition module 212 of the computer 200 acquires an environment for capturing an image to be subjected to image analysis (step S03). Here, as an environment acquisition method, the environment may be transmitted from the camera 100 as illustrated in FIG. 1 to acquire the environment, or the user may input using the input / output unit 240 of the computer 200. Or may be obtained by causing the user to input via another terminal (not shown).

FIG. 6 is an example of a question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment. This screen may be displayed on the input / output unit 240 of the computer 200 or may be displayed on the input / output unit of another terminal (not shown) used by the user. The purpose of the image analysis and the question about the shooting environment of the image are displayed to the user, and the user is selected or input to specify the purpose and environment. The example of FIG. 6 shows an example in which suspicious person detection is selected by presenting suspicious person detection, crop detection, and pest detection as the purpose of image analysis. In addition, as an image capturing environment, an indoor, outdoor (city), outdoor (farm) is presented for a place, an indoor is selected, and an area is input as 20 square meters. For the camera position, the ceiling corner, the center of the ceiling, the top of the desk / shelf were presented, and the center of the ceiling was selected. For lighting, the fluorescent lamp, LED, and natural light were presented and the LED was selected. An example is shown. Here, by selecting the search button 601, the input to the question is confirmed, and the purpose acquisition module 211 and the environment acquisition module 212 complete the acquisition. Here, an example has been shown in which questions about the purpose of image analysis and the shooting environment of an image are presented on one screen, but different screens may be used.

FIG. 7 is another example of a question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment. The example of FIG. 7 shows an example in which crop detection is selected by presenting suspicious person detection, crop detection, and pest detection as the purpose of image analysis. In addition, as an image capturing environment, for example, indoor, outdoor (city), and outdoor (farm) are presented for the location, and the outdoor (farm) is selected. For example, 5ha is input for the area. As for the camera position, an example in which the outside of the building, the utility pole, and the drone are presented and the utility pole is selected is shown. For the lighting, none and yes are presented, and no is selected. FIG. 7 shows an example in which the options presented as the image capturing environment are changed and displayed because the purpose of image analysis is different from the example of FIG. In this way, by changing the shooting environment options in accordance with the selected items such as the purpose of image analysis and the location, the user can more easily input according to the purpose and location. .

Returning to FIG. 1, finally, the learned model proposal module 241 of the computer 200 refers to the learned model database 23 to determine which purpose of the learned model is the purpose acquired in step S02 and the environment acquired in step S03. It is checked whether it is compatible with the environment, and an appropriate learned model is proposed (step S04). The proposal of the learned model here may be output to the input / output unit 240 of the computer 200 or may be output to the input / output unit of another terminal (not shown) used by the user.

FIG. 8 is an example of a screen for learning model proposal. This screen may be displayed on the input / output unit 240 of the computer 200 or may be displayed on the input / output unit of another terminal (not shown) used by the user. As a search result in the example of FIG. 6, when a search is performed on the condition that the purpose is suspicious person detection, the environment is indoors, the area is approximately 20 square meters, the center of the camera position ceiling, and the lighting LED, the learned model database 23 of FIG. Is retrieved, the learned model “Bunruki002” matches. Therefore, here, an example is shown in which “Bunruki002” is proposed as a learned model. As shown in the link 801 in FIG. 8 on the proposal screen, the download URL may be displayed and the user may select it so that the proposed learned model can be downloaded immediately. Further, the learned model proposal system may be terminated by selecting the end button 802. Further, by selecting the button 803 to the search screen, the screen may return to the question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment shown in FIGS. In addition, when a suitable learned model is not found, a learned model whose purpose matches and whose environment is close may be output as a proposal or reference. By using these learned models, the user can obtain an accurate image analysis result suitable for the purpose without spending learning time.

[Description of each function]
FIG. 2 is a diagram illustrating the functional blocks of the camera 100 and the computer 200 and the relationship between the functions. The camera 100 includes an imaging unit 10, a control unit 110, a communication unit 120, and a storage unit 130. The computer 200 includes a control unit 210, a communication unit 220, a storage unit 230, and an input / output unit 240. The control unit 210 implements the purpose acquisition module 211 and the environment acquisition module 212 in cooperation with the communication unit 220, the storage unit 230, and the input / output unit 240. Further, the input / output unit 240 implements the learned model proposal module 241 in cooperation with the control unit 210 and the storage unit 230. The communication network 300 may be a public communication network such as the Internet or a dedicated communication network, and enables communication between the camera 100 and the computer 200.

The camera 100 includes, as the imaging unit 10, an imaging device such as a lens, an imaging device, various buttons, and a flash, and captures images as captured images such as moving images and still images. An image obtained by imaging is a precise image having an amount of information necessary for image analysis.

The control unit 110 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and the like.

As the communication unit 120, a device for enabling communication with other devices, for example, a WiFi (Wireless Fidelity) compatible device compliant with IEEE 802.11 or an IMT-2000 standard such as a third generation or fourth generation mobile communication system. Compliant wireless device etc. It may be a wired LAN connection.

The storage unit 130 includes a data storage unit such as a hard disk or a semiconductor memory, and stores necessary data such as captured images. The purpose of image analysis and the shooting environment of the image may be stored together.

The control unit 210 includes a CPU, RAM, ROM, and the like. The control unit 210 realizes the purpose acquisition module 211 and the environment acquisition also by the communication unit 220, the storage unit 230, and the input / output unit 240.

A device for enabling communication with other devices as the communication unit 220, for example, a WiFi compatible device compliant with IEEE802.11 or a wireless device compliant with the IMT-2000 standard such as a third generation or fourth generation mobile communication system Etc. It may be a wired LAN connection. In addition, it is assumed that communication with another terminal (not shown) used by the user is performed via the communication unit 220 as necessary.

The storage unit 230 includes a data storage unit using a hard disk or a semiconductor memory, and stores data necessary for processing such as captured images, teacher data, and image analysis results. Further, the storage unit 230 includes a learned model database 23.

Suppose that the input / output unit 240 has functions necessary to use the learned model proposal system. The input / output unit 240 implements the learned model proposal module 241 in cooperation with the control unit 210 and the storage unit 230. As an example for realizing the input, a liquid crystal display that realizes a touch panel function, a keyboard, a mouse, a pen tablet, a hardware button on the apparatus, a microphone for performing voice recognition, and the like can be provided. Further, as an example for realizing the output, forms such as a liquid crystal display, a PC display, a projection on a projector, and an audio output can be considered. The function of the present invention is not particularly limited by the input / output method.

[Learned model proposal process]
FIG. 3 is a flowchart of the learned model proposal process. Processing executed by each module described above will be described in accordance with this processing.

First, a plurality of learned models are stored in the learned model database 23 of the computer 200 (step S301). The learned model may be acquired from another computer or a storage medium, or may be created by the computer 200. Further, this step S301 can be omitted when a sufficiently learned model is already stored in the learned model database 23.

FIG. 9 shows an example of the configuration of the learned model database. In the present invention, the learned model refers to predetermined learning data (teacher data) composed of past images and correct data, a learned classifier learned from the predetermined learning data, a learned convolutional neural network, and the like And machine learning techniques. When there is a conversion method for converting an image into a feature vector, the conversion method is also included in the learned model together with the machine learning method. In the learned model database 23, for each learned model, the purpose of image analysis and the environment in which the image was captured are stored in association with each other. Here, for the purpose of image analysis, for example, detecting the entry of a suspicious person (suspicious person detection), appropriately detecting the harvest time of crops (crop detection), and detecting the occurrence of pests ( Pest detection) is considered. In addition, regarding the environment in which an image is taken, conditions such as location, size, camera position, lighting, etc. can be considered. For example, for location, whether indoors, outdoors (city), outdoors (farm), for area, how many square meters or how ha, for camera position, in the corner of the ceiling, in the center of the ceiling, on the desk / shelf As for the top, the outside of the building, the utility pole, the drone, the lighting, it is possible to select whether it is a fluorescent lamp, an LED, natural light, or not, etc. By using the purpose of the image analysis and the information of the environment where the image is captured, it is possible to propose a learned model having the same purpose and environment.

When creating a learned model by the computer 200, machine learning is performed using predetermined learning data (teacher data) composed of past images and correct answer data. The machine learning method used here is preferably suitable for image analysis. Machine learning techniques include convolutional neural networks (CNN), perceptrons, recurrent neural networks (RNN), residual networks (ResNet), and other neural networks, support vector machines (SVN), and naive Bayes classifiers. It is done. Examples of the conversion method for converting an image into a feature vector include Bug of Visual Words, HOG (Histgram of Oriented Gradients), ORB, SURF, and the like. Further, when the learned model is stored in the learned model database 23, predetermined learning data composed of past images and correct answer data, a learned classifier learned from the predetermined learning data, and a learned classifier And machine learning techniques such as convolutional neural networks. When there is a conversion method for converting an image into a feature vector, the conversion method is also included in the learned model together with the machine learning method. A plurality of machine learning methods may be tried for certain learning data, and only the method having the best image analysis result may be stored in the learned model database 23. In addition, when a learned model is created by the computer 200, it takes time for learning. Therefore, it is performed when sufficient time and CPU can be spent for learning, such as before operation of the learned model proposal system. It is desirable.

Returning to FIG. 3, the purpose acquisition module 211 of the computer 200 transmits a purpose transmission request in order to acquire the purpose of what the image analysis is to be performed for (Step S302). FIG. 3 illustrates a case where the purpose is acquired by transmitting the purpose from the camera 100 as a method for acquiring the purpose. The destination for transmitting the target transmission request may be another terminal or the like (not shown). Further, instead of transmitting the purpose transmission request, a question may be presented to the input / output unit 240 of the computer 200.

The camera 100 receives the purpose transmission request from the computer 200 and transmits the purpose through the communication unit 120 (step S303). When the destination to which the purpose transmission request is transmitted is another terminal or the like (not shown), the other terminal (not shown) transmits the purpose to the computer 200. In addition, when a question is presented to the input / output unit 240 of the computer 200 instead of transmitting the purpose transmission request, the timing when the user confirms the input to the question corresponds to the purpose transmission.

The purpose acquisition module 211 of the computer 200 acquires the purpose (step S304). The acquisition source may be the camera 100, another terminal (not shown), or the input / output unit 240 of the computer 200 in accordance with the purpose transmission request destination in step S <b> 302.

Next, the environment acquisition module 212 of the computer 200 transmits an environment transmission request in order to acquire an environment for capturing an image to be subjected to image analysis (step S305). FIG. 3 illustrates a case where an environment is transmitted from the camera 100 and acquired as an environment acquisition method. The destination for transmitting the target transmission request may be another terminal or the like (not shown). Further, instead of transmitting the environment transmission request, a question may be presented to the input / output unit 240 of the computer 200.

The camera 100 receives the environment transmission request from the computer 200 and transmits the environment via the communication unit 120 (step S306). Here, when the camera 100 is a stereo camera, the size or the like may be determined by analyzing the distance to the subject and transmitted as an environment. In addition, when the camera 100 includes a special sensor such as a light intensity sensor, the location, illumination, and the like may be determined from the sensor value and transmitted as an environment. When the destination of sending the environment transmission request is another terminal or the like (not shown), the other terminal (not shown) transmits the environment to the computer 200. Further, when a question is presented to the input / output unit 240 of the computer 200 instead of transmitting the environment transmission request, the timing when the user confirms the input to the question corresponds to environment transmission.

Next, the environment acquisition module 212 of the computer 200 acquires the environment (step S307). The acquisition destination may be the camera 100, another terminal (not shown), or the input / output unit 240 of the computer 200, in accordance with the environment transmission request destination in step S305.

Here, the environment acquired by the environment acquisition module 212 may directly acquire the location, size, camera position, illumination, etc. For example, when the camera is equipped with a light intensity sensor, the location is obtained by using it. It may be determined whether the camera is indoors or outdoors, and when the camera is a stereo camera, the distance and area to the subject may be determined using the camera.

FIG. 6 is an example of a question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment. This screen may be displayed on the input / output unit 240 of the computer 200 or may be displayed on the input / output unit of another terminal (not shown) used by the user. The purpose of the image analysis and the question about the shooting environment of the image are displayed to the user, and the user is selected or input to specify the purpose and environment. The example of FIG. 6 shows an example in which suspicious person detection is selected by presenting suspicious person detection, crop detection, and pest detection as the purpose of image analysis. In addition, as an image capturing environment, indoors, outdoors (city), and outdoors (farm) are presented for places, and indoors are selected. For example, 20 square meters is entered for the area. For the position, the ceiling corner, the center of the ceiling, and the desk / shelf are presented and the center of the ceiling is selected. For the lighting, the fluorescent lamp, LED, and natural light are presented and the LED is selected. Is shown. Here, by selecting the search button 601, the input to the question is confirmed, and the purpose acquisition module 211 and the environment acquisition module 212 complete the acquisition. Here, an example has been shown in which questions about the purpose of image analysis and the shooting environment of an image are presented on one screen, but different screens may be used.

FIG. 7 is another example of a question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment. The example of FIG. 7 shows an example in which crop detection is selected by presenting suspicious person detection, crop detection, and pest detection as the purpose of image analysis. In addition, as an image capturing environment, for example, indoors, outdoors (city), and outdoors (farm) are presented as places, and the outdoors (farm) is selected. As for the camera position, the outside of the building, the utility pole, and the drone are presented to show the example where the utility pole is selected, and the lighting is shown as none, yes, and no is selected. FIG. 7 shows an example in which the options presented as the image capturing environment are changed because the purpose of image analysis is different from the example of FIG. As described above, by changing the shooting environment options in accordance with the selected items such as the purpose of image analysis and the location, it is possible to easily perform input according to the purpose and location.

Returning to FIG. 3, finally, the learned model proposal module 241 of the computer 200 refers to the learned model database 23 to determine which purpose of the learned model is the purpose acquired in step S304 and the environment acquired in step S307. It searches for suitability with the environment and proposes an appropriate learned model (step S308). The proposal of the learned model here may be output to the input / output unit 240 of the computer 200 or may be output to the input / output unit of another terminal (not shown) used by the user.

FIG. 8 is an example of a screen for learning model proposal. This screen may be displayed on the input / output unit 240 of the computer 200 or may be displayed on the input / output unit of another terminal (not shown) used by the user. As a search result in the example of FIG. 6, when a search is performed on the condition that the purpose is suspicious person detection, the environment is indoors, the area is approximately 20 square meters, the center of the camera position ceiling, and the lighting LED, the learned model database 23 of FIG. And “Bunruki002” is proposed as a suitable learned model. As shown by a link 801 in FIG. 8 on the proposal screen, a download URL may be indicated and the proposed learned model may be immediately downloaded by selecting it. Moreover, the learned model suggestion system may be terminated by selecting the end button 802. Further, by selecting the button 801 to the search screen, the screen may return to the question presentation and input screen for acquiring the purpose of image analysis and the image capturing environment shown in FIGS. In addition, when a suitable learned model is not found, a learned model whose purpose matches and whose environment is close may be output as a proposal or reference. By using the proposed learned model, the user can obtain an accurate image analysis result suitable for the purpose without spending learning time.

[Image comparison processing]
FIG. 4 is a diagram illustrating the functional blocks of the camera 100 and the computer 200 and the relationship between the functions when performing image comparison. In addition to the configuration of FIG. 2, the control unit 210 of the computer 200 implements the image acquisition module 213 in cooperation with the communication unit 220 and the storage unit 230. Further, the control unit 210 implements the image comparison module 214 in cooperation with the storage unit 230. FIG. 5 is a flowchart of learned model proposal processing when image comparison is performed. Processing executed by each module described above will be described in accordance with this processing. Since the processing from step S501 to step S507 in FIG. 5 corresponds to the processing from step S301 to step S307 in FIG. 3, only step S508 and subsequent steps will be described. Note that the processing in step S501 can be omitted when a sufficiently learned model is already stored in the learned model database 23, as in step S301.

The image acquisition module 213 of the computer 200 transmits an image transmission request to the camera 100 in order to acquire an image to be subjected to image analysis (step S508).

The camera 100 receives an image transmission request from the computer 200 and transmits an image via the communication unit 120 (step S509). The camera 100 may not only transmit an image captured in real time but also transmit an image captured by the camera 100 in the past and stored in the storage unit 130.

The image acquisition module 213 of the computer 200 acquires an image to be analyzed from the camera 100 (step S510).

Here, although not described in the flowchart, for example, when the environment information acquired in step S507 is insufficient, the environment acquisition module 212 analyzes the image acquired in step S510, and analyzes the location, area, and size. The camera position, lighting, etc. may be determined.

Next, the image comparison module 214 of the computer 200 obtains the image to be analyzed in step S510 and the image of the predetermined learning data (teacher data) of the learned model stored in the learned model database 23. Are compared (step S511). Here, as the image of the predetermined learning data of the learned model used for comparison, not all the images of one learned model may be used, but one or a plurality of images may be picked up and used. By this comparison, when the image to be analyzed is similar to the image of the predetermined learning data of the learned model, the accuracy when performing the image analysis using the learned model is improved, Shall be proposed. In addition, this comparison work may be performed to confirm that the image is similar to the image of the learned model that matches the purpose and the environment, and when there are multiple learned models that match the purpose and the environment, This may be done to suggest or refine a completed model, or it may be used to select a learned model to be proposed from a learned model that matches only the purpose when there is no learned model that matches the purpose and environment. Good. When a suitable learned model is not found, there may be no suitable learned model, or a learned model that matches the purpose and has the closest environment may be proposed.

Finally, the learned model proposal module 241 of the computer 200 proposes an appropriate learned model based on the image comparison result in step S511 (step S512). The proposal of the learned model here may be output to the input / output unit 240 of the computer 200 or may be output to the input / output unit of another terminal (not shown) used by the user. FIG. 8 shows an example of the learned model proposal screen as described above. By using the proposed learned model, the user can obtain an accurate image analysis result suitable for the purpose without spending learning time.

According to the present invention, in addition to the purpose of image analysis and the environment in which the image is captured, an existing learned model similar to that is proposed by acquiring and using the image to be analyzed. In addition, it is possible to provide a learned model proposal system, a learned model proposal method, and a program that can obtain an accurate image analysis result without taking a learning time.

The means and functions described above are realized by a computer (including a CPU, an information processing apparatus, and various terminals) reading and executing a predetermined program. The program may be, for example, in the form (SaaS: Software as a Service) provided from a computer via a network, or a flexible disk, CD (CD-ROM, etc.), DVD (DVD-ROM, DVD). -RAM, etc.) and a computer-readable recording medium such as a compact memory. In this case, the computer reads the program from the recording medium, transfers it to the internal storage device or the external storage device, stores it, and executes it. The program may be recorded in advance in a storage device (recording medium) such as a magnetic disk, an optical disk, or a magneto-optical disk, and provided from the storage device to a computer via a communication line.

As mentioned above, although embodiment of this invention was described, this invention is not limited to these embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

100 cameras, 200 computers, 300 communication networks

Claims

A system that proposes a trained model suitable for image analysis,
A learned model database that stores a learned model for image analysis in association with a purpose and an environment;
Purpose acquisition means for acquiring the purpose of image analysis;
An environment acquisition means for acquiring an environment in which an image for the purpose is captured;
With reference to the learned model database, a learned model proposing means for proposing a learned model suitable for the purpose and the environment;
A learned model suggestion system characterized by comprising:
The learned model includes a learned classifier trained with predetermined learning data consisting of past images and correct answer data,
The learned model proposing system according to claim 1, wherein the learned model proposing means proposes the learned classifier as a learned model.
The learned model suggestion system according to claim 2, wherein the learned model includes a type of classifier when an image is classified by a classifier and a conversion method for converting the image into a feature vector. .
The learned model suggestion system according to claim 1, wherein the learned model is a learned convolutional neural network learned with predetermined learning data including past images and correct answer data.
Image acquisition means for acquiring an image of an environment to be subjected to the image analysis;
Image comparison means for determining whether or not the acquired image and the image of the predetermined learning data are similar,
The learned model proposal system according to any one of claims 2 to 4, wherein, when the images are similar, the learned model proposal unit proposes the learned model. .
The learned model suggestion system according to any one of claims 1 to 5, wherein the environment acquisition unit acquires an answer input to the presented question as data related to the environment.
The learned model suggestion system according to any one of claims 1 to 5, wherein the environment acquisition unit acquires data detected by a sensor or a camera.
In a system that proposes a trained model suitable for image analysis,
A learned model database that stores a learned model for image analysis in association with a purpose and an environment;
Obtaining the purpose of image analysis;
Obtaining an environment in which an image for that purpose is taken;
Referring to the learned model database and proposing a learned model suitable for the purpose and the environment;
A learned model proposing method characterized by comprising:
In a learned model proposal system having a learned model database that stores a learned model for performing image analysis in association with a purpose and an environment,
Obtaining the purpose of image analysis;
Obtaining an environment in which an image for that purpose is taken;
Referring to the learned model database and proposing a learned model suitable for the purpose and the environment;
A program for running