JP6474946B1 - Image analysis result providing system, image analysis result providing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image analysis result providing system, an image analysis result providing method, and a program capable of outputting an accurate image analysis result without taking learning time for an unknown image to be newly analyzed. provide.
A storage module 231 for storing a machine-learned learned model obtained by image analysis of a known image, an acquisition module 211 for acquiring an unknown image for which a learned model has not yet been created, and a stored learned model A selection module 212 that selects a learned model of a known image similar in imaging condition to the acquired unknown image, and an image analysis module 213 that performs image analysis of the unknown image using the selected learned model The providing module 241 for providing the result of the image analysis is provided.
[Selection] Figure 1

Description

  The present invention selects a learned model of a known image whose imaging condition is similar to an unknown image to be newly analyzed from among a plurality of patterns of machine-learned learned models obtained by image analysis of known images by artificial intelligence The present invention relates to an image analysis result providing system, an image analysis result providing method, and a program that can output an accurate image analysis result without taking a learning time.

  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).

JP2015-69580

  Supervised learning is a well-known technique as a machine learning technique for artificial intelligence to perform image analysis.

  However, for this supervised learning, it is generally necessary to prepare a large number of images of tens of thousands to millions of images and add the correct teacher data to the images before learning. In order to improve the accuracy of image analysis by supervised learning, it takes time to prepare an image for learning and also requires a long time for learning.

  For example, when image detection is used to determine whether a person has entered a railroad crossing or not, since the position of the surveillance camera is usually fixed, there is a teacher at each railroad crossing such as railroad crossing A, railroad crossing B, and railroad crossing C. It is possible to improve the accuracy of approach detection for each of the crossing A, crossing B, and crossing C by performing machine learning, and in the subsequent image analysis processing, the learned model of the crossing A and the learned level of the crossing B have been learned. Use the trained model of the model and level crossing C, respectively. However, when a new approach detection is to be made at the level crossing D, a large number of tagged images serving as supervised learning teacher data for the level crossing D are prepared, and machine learning is performed from the beginning based on the images. Therefore, it takes time and effort to introduce approach detection.

  In response to this problem, the present inventor selects and uses a trained model of a level crossing similar to that of the level crossing D from the level crossing A, level crossing B, and level crossing C already machine-learned. Therefore, attention was paid to the fact that the image analysis accuracy of approach detection of a certain degree or more can be obtained from the beginning even at the crossing D.

  The present invention selects a learned model of a known image whose imaging condition is similar to an unknown image to be newly analyzed from among a plurality of patterns of machine-learned learned models obtained by image analysis of known images by artificial intelligence It is an object of the present invention to provide an image analysis result providing system, an image analysis result providing method, and a program capable of outputting an accurate image analysis result without taking a learning time.

  The present invention provides the following solutions.

The invention according to the first feature is
Storage means for storing a machine-learned learned model obtained by image analysis of a known image;
An acquisition means for acquiring an unknown image for which a trained model has not yet been created;
Selecting means for selecting a learned model of the known image that has a similar imaging condition to the acquired unknown image from the stored learned model;
Image analysis means for image analysis of the unknown image using the selected learned model;
Providing means for providing a result of the image analysis;
An image analysis result providing system is provided.

  According to the first aspect of the invention, a storage unit that stores a machine-learned learned model obtained by image analysis of a known image, an acquisition unit that acquires an unknown image for which a learned model has not yet been created, A selection unit that selects a learned model of the known image that has an imaging condition similar to that of the acquired unknown image from the stored learned model, and uses the selected learned model to convert the unknown image Image analysis means for image analysis, and providing means for providing a result of the image analysis.

  The invention according to the first feature is a category of the image analysis result providing system, but the image analysis result providing method and the program have the same operations and effects.

The invention according to the second feature is an image analysis result providing system which is the invention according to the first feature,
An image analysis result providing system is provided in which the imaging conditions are an imaging position and an imaging angle with respect to an imaging target.

  According to the second aspect of the invention, in the image analysis result providing system according to the first aspect of the invention, the imaging conditions are an imaging position and an imaging angle with respect to an imaging target.

The invention according to the third feature is an image analysis result providing system which is the invention according to the first feature or the second feature,
There is provided an image analysis result providing system, wherein the learned model is a mathematical expression and parameters used for the image analysis calculated by the machine learning and an image used for the machine learning.

  According to the third aspect of the invention, in the image analysis result providing system according to the first or second aspect of the invention, the learned model is used for the image analysis calculated by the machine learning. It is a picture used for a mathematical formula and a parameter, and the machine learning.

The invention according to the fourth feature is an image analysis result providing system which is the invention according to any one of the first feature to the third feature,
Provided is an image analysis result providing system comprising a creation unit that creates a new machine-learned model that has undergone image analysis for the unknown image.

  According to the invention related to the fourth feature, in the image analysis result providing system according to any one of the first feature to the third feature, a new machine-learned image analyzed for the unknown image A creation means for creating a learned model is provided.

The invention according to the fifth feature is an image analysis result providing system which is the invention according to the fourth feature,
The image analysis means performs image analysis of the unknown image using the selected learned model during a period until the new machine-learned learned model is created. Providing a provision system.

  According to the fifth aspect of the invention, in the image analysis result providing system according to the fourth aspect of the invention, the image analysis unit is configured to create a new machine-learned learned model, The unknown image is subjected to image analysis using the selected learned model.

An invention according to a sixth feature is an image analysis result providing system that is an invention according to any one of the first feature to the fifth feature,
Using the selected learned model for the unknown image, a new machine-learned learned model is generated by analyzing the unknown image using the analysis result obtained by image analysis of the unknown image as teacher data. Provided is an image analysis result providing system characterized by comprising creation means.

  According to the sixth aspect of the invention, in the image analysis result providing system according to any one of the first to fifth aspects, the selected learned model is used for the unknown image. And creating means for creating a new machine-learned model obtained by image-analyzing the unknown image using the analysis result obtained by image-analyzing the unknown image as teacher data.

The invention according to the seventh feature is
Storing a machine-learned learned model obtained by image analysis of a known image;
Obtaining an unknown image for which a trained model has not yet been created;
Selecting the learned model of the known image that is similar in imaging condition to the acquired unknown image from the stored learned model;
Using the selected learned model to image-analyze the unknown image;
Providing a result of the image analysis;
An image analysis result providing method is provided.

The invention according to the eighth feature is
In the image analysis result providing system,
Storing a machine-learned learned model obtained by image analysis of a known image;
Obtaining an unknown image for which a trained model has not yet been created,
Selecting from the stored learned models a learned model of the known image whose imaging conditions are similar to the acquired unknown image;
Image analysis of the unknown image using the selected learned model;
Providing a result of the image analysis;
Provide a program to execute.

  According to the present invention, a learned model of a known image whose imaging condition is similar to an unknown image to be newly analyzed from a plurality of patterns of a machine-learned learned model obtained by image analysis of a known image by artificial intelligence By selecting and using, it is possible to provide an image analysis result providing system, an image analysis result providing method, and a program capable of outputting an accurate image analysis result without taking a learning time. Become.

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 when an unknown image is acquired from the camera 100, image analysis processing is performed by the computer 200, and an image analysis result is provided. FIG. 4 is a diagram illustrating the relationship between the functional blocks of the camera 100 and the computer 200 and the functions when the learned image creation process of the unknown image is performed. FIG. 5 is a flowchart of the camera 100 and the computer 200 when the unknown image learned model creation process is performed. FIG. 6 is a diagram illustrating the relationship between the functional blocks of the camera 100 and the computer 200 and the respective functions when the image analysis processing is switched depending on whether the learned image creation processing for the unknown image is completed. FIG. 7 is a process corresponding to A in the flowchart of FIG. 6, and an unknown image learned model performed by the computer 200 when machine learning of an unknown image is possible when an unknown image learned model creation process is not completed. It is a flowchart figure of a creation process. FIG. 8 is a process corresponding to B in the flowchart of FIG. 6 and a learned model selection process performed by the computer 200 when machine learning of an unknown image is impossible when the learned image learning model creation process is not completed. FIG. FIG. 9 is a flowchart of the computer 200 when the unknown image learning model creation process is performed and the image analysis result of the unknown image is assigned as teacher data to perform machine learning of the unknown image. FIG. 10 shows examples of mathematical formulas and parameters calculated by machine learning, examples of images used for machine learning, and images of level crossings D, which are unknown images, as learned models of level crossings A, B, and C. FIG. FIG. 11 is a diagram for schematically explaining the relationship between the camera 100, the computer 200, and the subject 400. FIG. 12 is an example of a table showing the data structure of the learned model for each camera. FIG. 13 is an example of a table showing a learned model used for image analysis for each camera when there is no learned model of an unknown image captured by the camera D. FIG. 14 is an example of a table showing learned models used for image analysis for each camera when a learned model of an unknown image captured by the camera D is created.

  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 image analysis result providing 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 image analysis result providing 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 illustrated in FIG. 2, the camera 100 includes an imaging unit 10, a control unit 110, a communication unit 120, and a storage unit 130. As shown in FIG. 2, 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 acquisition module 211 in cooperation with the communication unit 220 and the storage unit 230. Further, the control unit 210 implements a selection module 212 and an image analysis module 213 in cooperation with the storage unit 230. The storage unit 230 implements the storage module 231 in cooperation with the control unit 210. The input / output unit 240 implements the providing 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 that includes an imaging device such as an imaging element and a lens that can perform data communication with the computer 200 and can measure the distance to the subject 400. 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 mounted on an unmanned aircraft, 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.

  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.

  In the image analysis result providing system of FIG. 1, first, the storage module 231 of the computer 200 stores a plurality of learned models in the storage unit 230 (step S01). The learned model may be acquired from another computer or a storage medium, or may be created by the computer 200. Further, a dedicated database may be provided in the storage unit 230.

  FIG. 10 is a diagram illustrating examples of mathematical formulas and parameters calculated by machine learning as trained models of level crossings A, level crossings B, and level crossings C, and examples of images used for machine learning. Moreover, it is a figure which shows an example of the unknown image of the level crossing D in which the learned model is not created yet.

  FIG. 12 is an example of a table showing the data structure of the learned model for each camera. In the present invention, the learned model is obtained by associating mathematical formulas and parameters for image analysis of a subject for each camera. In addition, an image file with teacher data used for machine learning for calculating a learned model may be associated. Further, as an imaging condition for each camera, an imaging angle and an imaging position may be stored in association with each other. The learned model created using the supervised data from the camera A that images the railroad crossing A in FIG. 10 is the learned model A, and the learned model created using the supervised data from the camera B that images the railroad crossing B. A learned model C created by using supervised data from a camera C that captures a learned model B and a railroad crossing C is a learned model C. In addition, a learned model has not been created for an image obtained by capturing the railroad crossing D with the camera D.

  FIG. 11 is a diagram for schematically explaining the relationship between the camera 100, the computer 200, and the subject 400. Assume that the camera 100 and the computer 200 can communicate with each other via the communication network 300. The camera 100 according to the present invention is an imaging device capable of measuring a distance to a subject. Regarding the method of measuring the distance to the subject, in addition to obtaining from the sensor of the camera 100 or the like, if the subject can be imaged simultaneously from a plurality of different directions, the difference between the images captured by each of the plurality of cameras is determined. It is also possible to measure the distance by learning the length and the actual distance. Also, it is possible to calculate the imaging angle using the measured distance. Furthermore, when the location of the camera 100 is fixed, the distance to the imaging location may be explicitly specified. As for the imaging angle, the imaging angle is how many times the camera 100 is tilted from the horizontal direction. In the example of FIG. 11, for the subject 400, the imaging angle of the camera 100 is 30 degrees, and the imaging position, that is, the imaging distance is 5-6 m. In this example, the imaging angle and the imaging position are given as examples of imaging conditions. However, in addition to this, if an approach to a railroad crossing is detected, the presence / absence of an alarm, the presence / absence and shape of a breaker, whether the track is a single line or multiple lines Information useful for determining the similarity between an unknown image and a learned model, such as, etc., may be included.

  As a pre-stage for storing a plurality of learned models in the storage unit 230, for machine learning by supervised learning, it is necessary to prepare a large number of images with appropriate labels as teacher data and perform machine learning. Also, a long time is required for learning. For this reason, it is preferable to prepare a learned model with a wide range of imaging conditions for each camera, assuming that a new unknown image is analyzed later. In addition, with regard to the label for supervised learning, assuming that it is actually necessary to provide a detailed image analysis result, if there is a crossing entry detection, only “with / without entry” is detected. "No entry, entry (adult), entry (child), entry (old man), entry (vehicle), entry (bicycle), entry (animal), possibility of entry" Considering whether detailed classification is necessary, it is necessary to add according to the purpose of the system.

  Returning to FIG. 1, the camera 100 transmits imaging data that is an unknown image to the computer 200 (step S02), and the acquisition module 211 of the computer 200 acquires the unknown image (step S03). Together with the unknown image, the acquisition module 211 acquires the imaging conditions such as the imaging angle and the imaging position from the camera 100 together. Here, the flow of transmitting imaging data that is an unknown image from the camera 100 is described. However, the acquisition module 211 instructs the camera 100 to transmit imaging data, and the camera 100 transmits imaging data in response to the instruction. May be performed. Further, the acquisition module 211 may acquire not only an image captured by the camera 100 in real time but also an image captured by the camera 100 in the past and stored in the storage unit 130. .

  Next, the selection module 212 of the computer 200 selects a learned model whose imaging condition is similar to that of the unknown image acquired in step S03 from the learned models stored in step S01 (step S04). Here, for example, when the imaged unknown image is the crossing D in FIG. 10, the imaging condition is determined from the learned model from any of the crossings A, B, and C where the learned model exists. Choose something similar. Assuming that the imaging angle of the camera D capturing the level crossing D is 20 degrees and the imaging distance is 4-5 m, and further analyzing the composition of the image capturing the level crossing D, the learned model B is selected here. And

  Next, the image analysis module 213 of the computer 200 performs image analysis of an unknown image captured by the camera D using the learned model B (step S05).

  FIG. 13 is an example of a table showing a learned model used for image analysis for each camera when there is no learned model of an unknown image captured by the camera D. In step S04, the learned model B is selected as the learned model whose imaging conditions are similar to those of the railroad crossing D imaged by the camera D. Therefore, in the camera D column of FIG. The table is filled in using yyyyyy as a mathematical expression and BBB, b, and β as parameters. Here, it is desirable that the imaging conditions of the actual camera D are filled in the imaging angle and the imaging position, which are imaging conditions. In addition, since supervised learning is not performed on the image captured by the camera D, the teacher data column may be left blank. Thereafter, until the learned model of the camera D is created, or until the selection of the learned model most suitable for the camera D is performed again by the selection module 212 due to the increase of other learned models, the process of FIG. It is possible to perform image analysis of the camera D using the table.

  Finally, the providing module 241 of the computer 200 provides the image analysis result to the input / output unit 240 of the computer 200 (step S06). If it is an entry detection to level crossing D, it will only be “with / without entry” or “no entry / with entry (adult) / with entry (child) / with entry (old) / with entry (vehicle) / with entry Image analysis results, such as whether to display in detail (such as (bicycle), entry (animal), entry possibility), or whether to display the result in warning sound or light Output according to the purpose of the providing system shall be performed. In addition, here, the example of detecting entry to a railroad crossing has been described as an example, but image analysis includes, for example, face recognition for individual determination, determination of the state of pest damage on crops, inventory check in a warehouse, and affected part for medical diagnosis It is assumed that the present invention can be applied to an appropriate one according to the purpose of the system, such as image recognition.

  According to the present invention, a learned model of a known image whose imaging condition is similar to an unknown image to be newly analyzed from a plurality of patterns of a machine-learned learned model obtained by image analysis of a known image by artificial intelligence By selecting and using, it is possible to provide an image analysis result providing system, an image analysis result providing method, and a program capable of outputting an accurate image analysis result without taking a learning time. Become.

[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 acquisition module 211 in cooperation with the communication unit 220 and the storage unit 230. Further, the control unit 210 implements a selection module 212 and an image analysis module 213 in cooperation with the storage unit 230. The storage unit 230 implements the storage module 231 in cooperation with the control unit 210. The input / output unit 240 implements the providing 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 that includes an imaging device such as an imaging element and a lens that can perform data communication with the computer 200 and can measure the distance to the subject 400. 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 mounted on an unmanned aircraft, 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.

  The camera 100 includes a lens, an imaging device, various buttons, an imaging device such as a flash, and the like as the imaging unit 10 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. Further, it may be possible to specify the resolution at the time of imaging, the camera angle, the camera magnification, and the like.

  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 device compatible with IEEE 802.11 (WiFi (Wireless Fidelity)) or a third generation, fourth generation mobile communication system, etc. 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 captured images, necessary data such as imaging conditions, and the like.

  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.

  The control unit 210 includes a CPU, RAM, ROM, and the like. The control unit 210 implements the acquisition module 211 in cooperation with the communication unit 220 and the storage unit 230. Further, the control unit 210 implements a selection module 212 and an image analysis module 213 in cooperation with the storage unit 230.

  As the communication unit 220, a device for enabling communication with other devices, 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.

  The storage unit 230 includes a data storage unit such as a hard disk or a semiconductor memory, and stores data necessary for processing such as captured images, teacher data, and image analysis results. The storage unit 230 implements the storage module 231 in cooperation with the control unit 210. Further, the storage unit 230 may include a database of learned models.

  The input / output unit 240 has functions necessary for using the image analysis result providing system. The input / output unit 240 implements the providing 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.

[Image analysis result provision processing]
FIG. 3 is a flowchart when an unknown image is acquired from the camera 100, image analysis processing is performed by the computer 200, and an image analysis result is provided. Processing executed by each module described above will be described in accordance with this processing.

  First, the storage module 231 of the computer 200 stores a plurality of learned models in the storage unit 230 (step S301). The learned model may be acquired from another computer or a storage medium, or may be created by the computer 200. Further, a dedicated database for storing the learned model may be provided in the storage unit 230. The process of step S301 may be skipped when a plurality of learned models are already stored or when a new learned model does not exist.

  FIG. 10 is a diagram illustrating an example of mathematical formulas and parameters calculated by machine learning as an already learned model of railroad crossing A, railroad crossing B, and railroad crossing C, and an image used for machine learning. Moreover, it is a figure which shows an example of the unknown image of the level crossing D in which the learned model is not created yet.

  FIG. 12 is an example of a table showing the data structure of the learned model for each camera. In the present invention, the learned model is obtained by associating mathematical formulas and parameters for image analysis of a subject for each camera. In addition, an image file with teacher data used for machine learning for calculating a learned model may be associated. Further, as an imaging condition for each camera, an imaging angle and an imaging position may be stored in association with each other. The learned model created using the supervised data from the camera A that images the railroad crossing A in FIG. 10 is the learned model A, and the learned model created using the supervised data from the camera B that images the railroad crossing B. A learned model C created by using supervised data from a camera C that captures a learned model B and a railroad crossing C is a learned model C. In addition, a learned model has not been created for an image obtained by capturing the railroad crossing D with the camera D.

  FIG. 11 is a diagram for schematically explaining the relationship between the camera 100, the computer 200, and the subject 400. Assume that the camera 100 and the computer 200 can communicate with each other via the communication network 300. The camera 100 according to the present invention is an imaging device capable of measuring a distance to a subject. Regarding the method of measuring the distance to the subject, in addition to obtaining from the sensor of the camera 100 or the like, if the subject can be imaged simultaneously from a plurality of different directions, the difference between the images captured by each of the plurality of cameras is determined. It is also possible to measure the distance by learning the length and the actual distance. Also, it is possible to calculate the imaging angle using the measured distance. Furthermore, when the location of the camera 100 is fixed, the distance to the imaging location may be explicitly specified. As for the imaging angle, the imaging angle is how many times the camera 100 is tilted from the horizontal direction. In the example of FIG. 11, for the subject 400, the imaging angle of the camera 100 is 30 degrees, and the imaging position, that is, the imaging distance is 5-6 m. In this example, the imaging angle and imaging distance are given as examples of imaging conditions, but in addition to this, if an approach to a railroad crossing is detected, the presence / absence of an alarm device, the presence / absence and shape of a breaker, whether the track is a single wire or a double track Information useful for determining the similarity between an unknown image and a learned model, such as, etc., may be included.

  As a pre-stage for storing a plurality of learned models in the storage unit 230, for machine learning by supervised learning, it is necessary to prepare a large number of images with appropriate labels as teacher data and perform machine learning. Also, a long time is required for learning. For this reason, it is preferable to prepare a learned model with a wide range of imaging conditions for each camera, assuming that a new unknown image is analyzed later. In addition, with regard to the label for supervised learning, assuming that it is actually necessary to provide a detailed image analysis result, if there is a crossing entry detection, only “with / without entry” is detected. "No entry, entry (adult), entry (child), entry (old man), entry (vehicle), entry (bicycle), entry (animal), possibility of entry" Considering whether detailed classification is necessary, it is necessary to add according to the purpose of the system.

  Returning to FIG. 3, the acquisition module 211 of the computer 200 requests the camera 100 to transmit an image (step 302). If there is no learned model for the image of the camera 100 at the time of this transmission request, the image acquired from the camera 100 is an unknown image.

  Upon receiving an image transmission request from the computer 200, the camera 100 captures an image with the imaging unit 10 (step S303).

  Then, the camera 100 transmits imaging data that is an unknown image to the computer 200 via the communication unit 120 (step S304).

  The acquisition module 211 of the computer 200 acquires an unknown image (step S305). Here, it is assumed that imaging conditions such as an imaging angle and an imaging position are acquired from the camera 100 together with the unknown image. The acquisition module 211 may acquire not only an image captured by the camera 100 in real time but also an image captured by the camera 100 in the past and stored in the storage unit 130.

  Next, the selection module 212 of the computer 200 selects a learned model whose imaging condition is similar to the unknown image acquired in step S305 from the learned models stored in step S301 (step S306). Here, for example, when the imaged unknown image is the crossing D in FIG. 10, the imaging condition is determined from the learned model from any of the crossings A, B, and C where the learned model exists. Choose something similar. Assuming that the imaging angle of the camera D capturing the level crossing D is 20 degrees and the imaging distance is 4-5 m, and further analyzing the composition of the image capturing the level crossing D, the learned model B is selected here. And

  Next, the image analysis module 213 of the computer 200 performs image analysis of the unknown image captured by the camera D using the learned model B (step S307).

  FIG. 13 is an example of a table showing a learned model used for image analysis for each camera when there is no learned model of an unknown image captured by the camera D. In step S306, the learned model B is selected as the learned model whose imaging conditions are similar to those of the railroad crossing D imaged by the camera D. Therefore, in the column of the camera D in FIG. The table is filled in using yyyyyy as a mathematical expression and BBB, b, and β as parameters. Here, it is desirable that the imaging conditions of the actual camera D are filled in the imaging angle and the imaging position, which are imaging conditions. In addition, since supervised learning is not performed on the image captured by the camera D, the teacher data column may be left blank. Thereafter, until the learned model of the camera D is created, or until the selection of the learned model most suitable for the camera D is performed again by the selection module 212 due to the increase of other learned models, the process of FIG. It is possible to perform image analysis of the camera D using the table.

  Finally, the providing module 241 of the computer 200 provides the image analysis result to the input / output unit 240 of the computer 200 (step S308). If it is an entry detection to level crossing D, it will only be “with / without entry” or “no entry / with entry (adult) / with entry (child) / with entry (old) / with entry (vehicle) / with entry Image analysis results, such as whether to display in detail (such as (bicycle), entry (animal), entry possibility), or whether to display the result in warning sound or light Output according to the purpose of the providing system shall be performed. In addition, here, the example of detecting entry to a railroad crossing has been described as an example, but image analysis includes, for example, face recognition for individual determination, determination of the state of pest damage on crops, inventory check in a warehouse, and affected part for medical diagnosis It is assumed that the present invention can be applied to an appropriate one according to the purpose of the system, such as image recognition. Further, the provision of the image analysis result is not limited to the output to the input / output unit 240 of the computer 200, and the output according to the system is performed, for example, output to another device via the communication unit 220. And

  As described above, according to the present invention, a known image whose imaging condition is similar to an unknown image to be newly subjected to image analysis from among a plurality of patterns of machine-learned learned models obtained by image analysis of known images by artificial intelligence. An image analysis result providing system, an image analysis result providing method, and a program capable of outputting an accurate image analysis result without spending learning time by selecting and using a learned model of It becomes possible.

[Learned model creation process]
FIG. 4 is a diagram illustrating the relationship between the functional blocks of the camera 100 and the computer 200 and the functions when the learned image creation process of the unknown image is performed. In addition to the configuration of FIG. 2, the control unit 210 of the computer 200 implements the creation module 214 in cooperation with the storage unit 230. FIG. 5 is a flowchart of the camera 100 and the computer 200 when the unknown image learned model creation process 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 S503 in FIG. 5 corresponds to the processing from step S301 to step S303 in FIG. 3, step S504 and subsequent steps will be described. As in step S301, the process of step S501 may be skipped when a plurality of learned models are already stored or when a new learned model does not exist.

  Upon receiving the image transmission request from the computer 200, the camera 100 transmits the unknown image captured by the imaging unit 10 to the computer 200 via the communication unit 120 (step S504). Here, it is desirable to acquire as many unknown images as possible captured by the camera 100 in order to perform a learned model creation process for unknown images. Therefore, not only images captured by the camera 100 in real time but also images previously captured by the camera 100 and stored in the storage unit 130 may be transmitted.

  The acquisition module 211 of the computer 200 acquires a plurality of unknown images (step S505). Here, it is assumed that imaging conditions such as an imaging angle and an imaging position are acquired from the camera 100 together with each unknown image.

  Next, the creation module 214 of the computer 200 adds teacher data to the unknown image acquired in step S505 (step S506). Here, an operation of adding a label that is a correct answer of the image analysis result to a plurality of acquired unknown images is assigned teacher data. As described above, with regard to the label for supervised learning, assuming that it is actually necessary to provide a detailed image analysis result, if it is a crossing entry detection, “with / without entry” "There is no entry, there is entry (adult), there is entry (child), there is entry (old man), there is entry (vehicle), there is entry (bicycle), there is entry (animal), and there is a possibility of entry. It is necessary to add according to the purpose of the system in consideration of whether detailed classification is necessary.

  The creation module 214 of the computer 200 performs machine learning by supervised learning using an unknown image to which teacher data is added (step S507).

  Next, the creation module 214 creates a learned model of an unknown image based on the result of machine learning in step S507 (step S508).

  Finally, the learned module of the unknown image is stored in the storage unit 230 by the storage module 231 (step S509).

  FIG. 14 is an example of a table showing learned models used for image analysis for each camera when a learned model of an unknown image captured by the camera D is created. The learned model D of the camera D created in step S508 is described in the column of the camera D in FIG. The table is filled in using learned model D as a usage model of camera D, vvvvvvvv as a mathematical expression, and DDD, d, and dD as parameters. In addition, since supervised learning is performed, the teacher data column also includes teacher data. Thereafter, until the teacher D increases the teacher data again and creates the learned model, the image analysis of the camera D can be performed using the table of FIG.

  In this learned model creation process, it is necessary to prepare a large number of images with appropriate labels in order to increase the accuracy of image analysis for new unknown images, and to learn over time. When a new unknown image is accepted after the operation of the image analysis result providing system is started, it becomes a heavy burden. Therefore, when a new learned model creation process cannot be performed, the method described with reference to FIGS. 2 and 3 can provide a somewhat accurate image analysis result without taking a learning time. is suggesting. However, in order to obtain better accuracy, it is desirable to perform machine learning according to an unknown image and create a learned model of a more appropriate mathematical formula or parameter. Therefore, depending on the timing at which a large amount of unknown images necessary for machine learning can be accumulated and the timing at which appropriate correct labels can be added, the system should not be overloaded so as not to affect other image analysis tasks. It is necessary to implement it appropriately in accordance with the operation of the image analysis result providing system.

  As described above, according to the present invention, a known image whose imaging condition is similar to an unknown image to be newly subjected to image analysis from among a plurality of patterns of machine-learned learned models obtained by image analysis of known images by artificial intelligence. By selecting and using the learned model, it is possible to output accurate image analysis results without spending learning time, and to perform machine learning according to the new unknown image by the learned model creation process. By creating a learned model, it is possible to provide an image analysis result providing system, an image analysis result providing method, and a program that can further improve the image analysis result.

[Image analysis switching process]
FIG. 6 is a diagram illustrating the relationship between the functional blocks of the camera 100 and the computer 200 and the respective functions when the image analysis processing is switched depending on whether the learned image creation processing for the unknown image is completed. FIG. 7 is a process corresponding to A in the flowchart of FIG. 6, and an unknown image learned model performed by the computer 200 when machine learning of an unknown image is possible when an unknown image learned model creation process is not completed. It is a flowchart figure of a creation process. FIG. 8 is a process corresponding to B in the flowchart of FIG. 6 and a learned model selection process performed by the computer 200 when machine learning of an unknown image is impossible when the learned image learning model creation process is not completed. FIG. The configurations of the camera 100 and the computer 200 are the same as those in FIG. Further, the processing from step S601 to step S605 in FIG. 6 corresponds to the processing from step S301 to step S305 in FIG.

  The creation module 214 of the computer 200 confirms whether the learned image learned model has been created for the imaging data acquired in step S605 (step S606). Here, if the imaging data acquired in step S605 is acquired data from the camera 100 for the first time, it is an unknown image, so it is considered that a learned model of the unknown image has not been created.

  When the learned model of the unknown image has not been created, the creation module 214 determines whether machine learning is possible using the unknown image acquired in step S605 or the previously stored unknown image ( Step S607). Here, whether there is a large amount of images that will be teacher data with the appropriate labels necessary for machine learning, whether the system is loaded and there is no problem over time for machine learning, It is possible to judge based on the above. Further, the determination may be made according to the operation status of the image analysis result providing system, and the determination corresponding to the system is performed.

  If it is determined that machine learning of an unknown image is possible, the process proceeds to the flowchart of process A in FIG. 7 (step S608).

  In process A of FIG. 7, the creation module 214 of the computer 200 adds teacher data to the unknown image acquired in step S605 and the previously stored unknown image (step S701). Here, an operation of adding a label that is a correct answer of the image analysis result to a plurality of acquired unknown images is assigned teacher data. As described above, with regard to the label for supervised learning, assuming that it is actually necessary to provide a detailed image analysis result, if it is a crossing entry detection, “with / without entry” "There is no entry, there is entry (adult), there is entry (child), there is entry (old man), there is entry (vehicle), there is entry (bicycle), there is entry (animal), and there is a possibility of entry. It is necessary to add according to the purpose of the system in consideration of whether detailed classification is necessary.

  The creation module 214 of the computer 200 performs machine learning by supervised learning using an unknown image to which teacher data is added (step S702).

  Next, the creation module 214 creates a learned model of an unknown image based on the result of machine learning in step S702 (step S703).

  Next, the learned module of the unknown image is stored in the storage unit 230 by the storage module 231 (step S704).

  FIG. 14 is an example of a table showing learned models used for image analysis for each camera when a learned model of an unknown image captured by the camera D is created. The learned model D of the camera D created in step S703 is described in the column of the camera D in FIG. The table is filled in using learned model D as a usage model of camera D, vvvvvvvv as a mathematical expression, and DDD, d, and dD as parameters. In addition, since supervised learning is performed, the teacher data column also includes teacher data. Thereafter, until the teacher D increases the teacher data again and creates the learned model, the image analysis of the camera D can be performed using the table of FIG.

  Next, the image analysis module 213 of the computer 200 performs image analysis of an unknown image captured by the camera D using the created learned model D (step S705).

  Next, the providing module 241 of the computer 200 provides the image analysis result to the input / output unit 240 of the computer 200 (step S706). Thereafter, returning to the flowchart of FIG. 6, the process proceeds to step S614.

  If it is determined in step S607 that machine learning of an unknown image is not possible, the creation module 214 stores the unknown image acquired in step S605 in the storage unit 230 (step S609). This is because it is used as teacher data when machine learning for learned model creation processing for an unknown image is performed later.

  After the storage process in step S609, the process proceeds to the flowchart of process B in FIG. 8 (step S610).

  The selection module 212 of the computer 200 selects a learned model whose imaging condition is similar to the unknown image acquired in step S605 from the learned models stored in step S601 (step S801). Here, for example, when the imaged unknown image is the crossing D in FIG. 10, the imaging condition is determined from the learned model from any of the crossings A, B, and C where the learned model exists. Choose something similar. Assuming that the imaging angle of the camera D capturing the level crossing D is 20 degrees and the imaging distance is 4-5 m, and further analyzing the composition of the image capturing the level crossing D, the learned model B is selected here. And

  Next, the image analysis module 213 performs image analysis of an unknown image captured by the camera D using the learned model B (step S802).

  FIG. 13 is an example of a table showing a learned model used for image analysis for each camera when there is no learned model of an unknown image captured by the camera D. In step S801, the learned model B is selected as the learned model whose imaging conditions are similar to those of the railroad crossing D imaged by the camera D. Therefore, in the column of the camera D in FIG. The table is filled in using yyyyyy as a mathematical expression and BBB, b, and β as parameters. Here, it is desirable that the imaging conditions of the actual camera D are filled in the imaging angle and the imaging position, which are imaging conditions. In addition, since supervised learning is not performed on the image captured by the camera D, the teacher data column may be left blank. Thereafter, until the learned model of the camera D is created, or until the selection of the learned model most suitable for the camera D is performed again by the selection module 212 due to the increase of other learned models, the process of FIG. It is possible to perform image analysis of the camera D using the table.

  Next, the providing module 241 provides the image analysis result to the input / output unit 240 of the computer 200 (step S803). Thereafter, returning to the flowchart of FIG. 6, the process proceeds to step S614.

  If a learned model of an unknown image has been created in step S606 of FIG. 6, it is considered that a learned model of an unknown image has already been created through the flow of process A. In this case, the selection module 212 selects and applies the learned model D created in step S703 of process A (step S611).

  Next, the image analysis module 213 of the computer 200 performs image analysis of an unknown image captured by the camera D using the learned model D (step S612).

  Next, the providing module 241 of the computer 200 provides the image analysis result to the input / output unit 240 of the computer 200 (step S613).

  In step S706, step S803, and step S613, the providing module 241 provides the image analysis result to the input / output unit 240 of the computer 200. If the entry to the railroad crossing D is detected, “provided / not entered”. Or "No entry / Entering (adult) / Entering (child) / Entering (old man) / Entering (vehicle) / Entering (bicycle) / Entering (animal) / possible entry" In addition to displaying the image in detail, whether to display the image as well as whether to present the result with a warning sound or light, output corresponding to the purpose of the image analysis result providing system is performed. In addition, here, the example of detecting entry to a railroad crossing has been described as an example, but image analysis includes, for example, face recognition for individual determination, determination of the state of pest damage on crops, inventory check in a warehouse, and affected part for medical diagnosis It is assumed that the present invention can be applied to an appropriate one according to the purpose of the system, such as image recognition. Further, the provision of the image analysis result is not limited to the output to the input / output unit 240 of the computer 200, and the output according to the system is performed, for example, output to another device via the communication unit 220. And

  Finally, it is confirmed whether or not the image analysis result providing process can be ended (step S614). If not, the process returns to step S602 to continue the process. If the process ends, the image analysis result providing process is ended. To do.

  As described above, according to the present invention, a known image whose imaging condition is similar to an unknown image to be newly subjected to image analysis from among a plurality of patterns of machine-learned learned models obtained by image analysis of known images by artificial intelligence. By selecting and using the learned model, it is possible to output accurate image analysis results without spending learning time, and to perform machine learning according to the new unknown image by the learned model creation process. It is possible to create a learned model. Also, until a new trained model is created, select and use a trained model of a known image, and after creating a new trained model, a dedicated trained model with higher accuracy is created. By using, it is possible to provide an image analysis result providing system, an image analysis result providing method, and a program that have high accuracy and can further improve the accuracy from the citation start point of the system.

[Learned model creation process using image analysis result of unknown image]
FIG. 9 is a flowchart of the computer 200 when the unknown image learning model creation process is performed and the image analysis result of the unknown image is assigned as teacher data to perform machine learning of the unknown image. The configurations of the camera 100 and the computer 200 are the same as those in FIG. In FIG. 9, the process is described from step S901, but before this, processing corresponding to step S301 to step S308 of FIG. 3 is performed, and image analysis of the unknown image by the selected learned model is performed. Assume that image analysis results have been obtained.

  The creation module 214 of the computer 200 assigns the image analysis result of step S307 as teacher data to the unknown image acquired in step S305 (step S901). Here, by adding the image analysis result of the unknown image based on the selected learned model as teacher data as it is, the cost of manually adding teacher data that becomes correct data to a large amount of images necessary for machine learning, It becomes possible to reduce significantly.

  The creation module 214 performs machine learning by supervised learning using the unknown image to which the teacher data is added (step S902).

  Next, the creation module 214 creates a learned model of an unknown image based on the result of machine learning in step S902 (step S903).

  Finally, the learned module of the unknown image is stored in the storage unit 230 by the storage module 231 (step S904).

  As described above, according to the present invention, a known image whose imaging condition is similar to an unknown image to be newly subjected to image analysis from among a plurality of patterns of machine-learned learned models obtained by image analysis of known images by artificial intelligence. By selecting and using the learned models, it is possible to output accurate image analysis results without spending learning time. Can be used to significantly reduce the cost of manually adding correct teacher data to a large amount of images required for machine learning, and ultimately improve image analysis results. It is possible to provide an image analysis result providing system, an image analysis result providing method, and a program.

  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. For example, the program may be 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, 400 subjects

Claims (6)

Storage means for storing a machine-learned learned model obtained by image analysis of a known image;
An acquisition means for acquiring an unknown image for which a trained model has not yet been created;
A selection unit that selects, from the stored learned models, a learned model of the known image that has a similar imaging condition including an imaging position and an imaging angle with respect to the acquired unknown image and an imaging target;
Image analysis means for image analysis of the unknown image using the selected learned model;
Providing means for providing a result of the image analysis;
With
The system for providing an image analysis result , wherein the learned model includes at least a mathematical expression and parameters used for the image analysis.   2. The image analysis result providing system according to claim 1, further comprising a creation unit that creates a new machine-learned model that has undergone image analysis for the unknown image.   3. The image analysis unit performs image analysis of the unknown image using the selected learned model during a period until the new machine-learned learned model is created. The image analysis result providing system described in 1.   Using the selected learned model for the unknown image, a new machine-learned learned model is generated by analyzing the unknown image using the analysis result obtained by image analysis of the unknown image as teacher data. The image analysis result providing system according to claim 1, further comprising a creation unit. An image analysis result providing method executed by an image analysis result providing system,
Storing a machine-learned learned model obtained by image analysis of a known image;
Obtaining an unknown image for which a trained model has not yet been created;
Selecting from the stored learned models a learned model of the known image having a similar imaging condition including an imaging position and an imaging angle with respect to the acquired unknown image and an imaging target;
Using the selected learned model to image-analyze the unknown image;
Providing a result of the image analysis;
With
The trained model is at least composed of image analysis results providing method in a formula and parameters used for the image analysis. In the image analysis result providing system, the learned model is configured with at least a mathematical expression and parameters used for image analysis calculated by machine learning.
Storing a machine-learned learned model obtained by image analysis of a known image;
Obtaining an unknown image for which a trained model has not yet been created,
A step of selecting a learned model of the known image having similar imaging conditions including an imaging position and an imaging angle with respect to the acquired unknown image and an imaging target from the stored learned model;
Image analysis of the unknown image using the selected learned model;
Providing a result of the image analysis;
A computer-readable program for executing the program.

Images