Image selection device, image selection method, and program

Abstract

This image selection device generates teacher data for generating a first machine learning model, and is provided with at least an acquisition unit and a determination unit. The acquisition unit acquires a first processing result and acquires a second processing result. The first processing result is the result of processing a target image using the first machine learning model. The second processing result is the result of processing executed by a high-precision image processing unit capable of processing the target image with higher accuracy than the first machine learning model. The determination unit uses the first processing result and the second processing result to determine whether to use the target image as teacher data in the first machine learning model.

Description

Image selection device, image selection method, and program

[0001] This application claims priority based on Japanese Patent Application No. 2023-092663, filed on June 5, 2023, the disclosure of which is incorporated herein by reference in its entirety.

[0002] The present invention relates to an image selection device, an image selection method, and a program.

[0003] When generating a model by machine learning, training data is often required. Patent Literature 1 describes that labels of training data are determined based on classification results obtained by a model different from the model to which the training data is applied.

[0004] International Publication No. 2019 / 215780

[0005] To improve the performance of a machine learning model, it is necessary to prepare appropriate training data. Preparing such training data manually requires a great deal of effort. However, it is difficult to reduce this effort with the above-mentioned Patent Document 1.

[0006] In view of the above-mentioned problems, an example of an object of the present invention is to provide an image selection device, an image selection method, and a program that can reduce the effort required to prepare training data.

[0007] The image selection device of the present disclosure includes an acquisition means for acquiring a first processing result that is the result of processing a target image using a first machine learning model, and acquiring a second processing result that is the processing result by a high-precision image processing means that can process the target image with higher accuracy than the first machine learning model; and a determination means for determining whether or not to use the target image as training data for the first machine learning model using the first processing result and the second processing result.

[0008] The image processing method of the present disclosure is an image selection method in which at least one computer acquires a first processing result that is the result of processing a target image using a first machine learning model, and acquires a second processing result that is the processing result by a high-precision image processing means that can process the target image with higher accuracy than the first machine learning model, and uses the first processing result and the second processing result to determine whether or not to use the target image as training data for the first machine learning model.

[0009] The program in the present disclosure has at least one computer having: an acquisition means for acquiring a first processing result that is the result of processing a target image using a first machine learning model, and a second processing result that is the processing result by a high-precision image processing means that can process the target image with higher accuracy than the first machine learning model; and a determination means for determining whether or not to use the target image as training data for the first machine learning model using the first processing result and the second processing result.

[0010] According to the present disclosure, an image selection device, an image selection method, and a program can be provided that can reduce the effort required to prepare training data.

[0011] FIG. 1 is a block diagram illustrating an example of a configuration of an image selection device according to the present disclosure. FIG. 2 is a diagram for explaining an example of a first processing result and a second processing result. FIG. 3 is a flowchart illustrating an example of a processing operation of an image selection device according to the present disclosure. FIG. 4 is a diagram illustrating an example of a usage environment of an image selection device according to the present disclosure. FIG. 5 is a block diagram illustrating an example of a configuration of an image selection device according to the present disclosure. FIG. 6 is a block diagram illustrating an example of a hardware configuration of an image selection device according to the present disclosure. FIG. 7 is a flowchart illustrating an example of a processing operation of an image selection device according to the present disclosure. FIG. 8 is a block diagram illustrating an example of a screen output by an image selection device according to the present disclosure. FIG. 9 is a flowchart illustrating an example of a processing operation of an image selection device according to the present disclosure. FIG. 10 is a block diagram illustrating an example of a configuration of an image selection device according to the present disclosure.

[0012] Hereinafter, in this disclosure, the drawings relate to one or more embodiments. In addition, in all drawings, similar components are given similar reference numerals and descriptions thereof will be omitted as appropriate.

[0013] First Embodiment An image selection device 20 is a device that generates training data for generating a first machine learning model, and includes at least an acquisition unit 220 and a determination unit 230 as shown in FIG.

[0014] The acquisition unit 220 acquires a first processing result and a second processing result. The first processing result is a result of processing the target image using a first machine learning model. The second processing result is a processing result by a high-precision image processing unit that can process the target image with higher accuracy than the first machine learning model.

[0015] As shown in Fig. 2, the first processing result and the second processing result indicate, for example, whether or not at least one of a specific event and an object is included in the target image. Examples of objects include, but are not limited to, at least one of a vehicle such as an automobile, a person, and a foreign object on a road. Examples of specific events include, but are not limited to, a traffic accident or a traffic jam. The first processing result and the second processing result may further include the positions of the specific event and the object in the target image.

[0016] Here, an example of the high-precision image processing unit "being able to process target images with higher accuracy than the first machine learning model" is when at least one of the following conditions is satisfied: (A) The high-precision image processing unit can detect events or objects that could not be detected by the first machine learning model. (B) Both the first machine learning model and the high-precision image processing unit pre-process the target image, but the high-precision image processing unit performs pre-processing with a greater processing load than the first machine learning model. Examples of pre-processing used by the high-precision image processing unit are deep learning, SVM (Support Vector Machine), or cascade separator. SVM requires less computational effort than deep learning, but tends to have lower accuracy. Furthermore, when used alone, the cascade separator also requires less computational effort than deep learning, but tends to have lower accuracy. (C) The first machine learning model does not pre-process the target image, but the high-precision image processing unit performs pre-processing. An example of pre-processing used by the high-precision image processing unit is as illustrated in (B). (D) When the high-precision image processing unit and the first machine learning model are machine learning, the number of parameters used, for example, the number of features, is greater in the high-precision image processing unit than in the first machine learning model. (E) When the high-precision image processing unit and the first machine learning model both use neural networks such as deep learning, the number of intermediate layers and / or nodes of the neural network is greater in the high-precision image processing unit than in the first machine learning model. (F) When the high-precision image processing unit and the first machine learning model both use bagging, the number of algorithms used in the high-precision image processing unit is greater than that in the first machine learning model. (G) When the high-precision image processing unit and the first machine learning model both use bagging, the accuracy of the algorithm used in the high-precision image processing unit is equal to or greater than that of the algorithm used in the first machine learning model.

[0017] In addition, both the high-precision image processing unit and the first machine learning model may calculate a confidence level that an event or object is captured in the target image, and if this confidence level is equal to or greater than a reference value, the target image may be determined to contain an event or object. In this case, the reference value used by the high-precision image processing unit and the reference value used by the first machine learning model may be different from each other or may be the same. Furthermore, at least one of the reference value used by the high-precision image processing unit and the reference value used by the first machine learning model may be set by a user or developer of the image selection device 20 via a user interface.

[0018] Furthermore, the amount of calculation performed by the high-precision image processing unit is usually greater than the amount of calculation when the first machine learning model is used, and therefore the processing time by the high-precision image processing unit is longer than the processing time when the first machine learning model is used.

[0019] The determination unit 230 determines whether to use the target image as training data for the first machine learning model using the first processing result and the second processing result. For example, if the target image contains a specific event or object according to the second processing result, but does not contain an event or object according to the first processing result, the determination unit 230 determines to use the target image as training data. The target image determined by the determination unit 230 to be used as training data is used as training data for updating the first machine learning model. Note that the training data may consist of the target image alone, or may be data in which the target image is associated with an annotation result, for example, the second processing result, as a label.

[0020] 3, the acquisition unit 220 of the image selection device 20 acquires the first processing result and the second processing result (step S10). Next, the determination unit 230 of the image selection device 20 determines whether or not to use the target image as training data for the first machine learning model using the first processing result and the second processing result (step S20). These processes are performed, for example, in a batch format.

[0021] According to this image selection device 20, the determination unit 230 determines whether or not to use the target image as training data for the first machine learning model, using the results of the first and second processes, thereby reducing the effort required to prepare training data.

[0022] Next, a detailed example of the environment in which the image selection device 20 is used will be described.

[0023] 4, the image selection device 20 is used together with the image processing device 10, at least one imaging device 30, a first storage unit 40, a second storage unit 50, and a learning device 60. The image selection device 20 is often used together with multiple imaging devices 30. Although only one image processing device 10 is shown in FIG. 4, multiple image processing devices 10 may be used for one image selection device 20.

[0024] The imaging device 30 generates images at a predetermined frame rate. As an example, the imaging device 30 is a surveillance camera. The imaging range of the imaging device 30 may be a road, a building entrance, or indoors, for example, indoors. When the imaging device 30 captures an image of a road, multiple imaging devices 30 are placed along the road. When multiple imaging devices 30 are present, the imaging devices 30 link the generated images with the identification information and generation date and time of the imaging devices 30 and transmit them to the image processing device 10. The identification information of the imaging devices 30 is used, for example, to identify the location where the image was captured. Note that the generation date and time of the image may be generated by the image processing device 10. In this case, for example, the image processing device 10 sets the date and time when the image was acquired from the imaging device 30 as the generation date and time of the image.

[0025] The image processing device 10 includes a first image processing unit 110. The first image processing unit 110 processes a target image generated by the imaging device 30 to determine whether the target image contains a specific event or object. In this process, the first image processing unit 110 uses a first machine learning model. The machine learning used in the first machine learning model is, for example, a neural network such as deep learning, bagging, boosting, or stacking, but is not limited to these. The first image processing unit 110 associates the processing result by the first machine learning model, i.e., the first processing result, with the target image on which the first processing result was based, its generation date and time, and identification information of the imaging device 30 that generated the target image, and stores the result together with the target image in the first storage unit 40.

[0026] The imaging device 30 may transmit target images to the image processing device 10 in real time, and the first image processing unit 110 of the image processing device 10 may process the target images in real time. In this case, the first image processing unit 110 generates a first processing result each time a target image is generated, so the target images need to be processed quickly, i.e., in a short time. For this reason, it is difficult to increase the processing load of the first image processing unit 110 in order to improve processing accuracy.

[0027] The image selection device 20 reads out a target image and a first processing result from the first storage unit 40. The image selection device 20 also includes a high-precision image processing unit 210. The high-precision image processing unit 210 processes the target image to generate a second processing result. In this manner, the second processing result is generated after the first processing result. Furthermore, the amount of processing required to generate the second processing result is greater than the amount of processing required to generate the first processing result. The image selection device 20 then uses the read-out first processing result and the generated second processing result to determine whether or not to use the target image as training data, and stores information capable of identifying the target image determined to be used as training data in the second storage unit 50. As an example, the image selection device 20 stores the target image in the second storage unit 50 as at least a part of the training data.

[0028] The learning device 60 includes a learning unit 610. The learning unit 610 updates the first machine learning model by using the target image selected by the image selection device 20 as at least a part of the training data, and stores the updated first machine learning model in the image processing device 10. The image processing device 10 then processes the target image using the updated first machine learning model.

[0029] The image selection device 20 may include the first image processing unit 110, may include the learning unit 610, or may include both the first image processing unit 110 and the learning unit 610. The image selection device 20 may be integrated with at least one of the first storage unit 40 and the second storage unit 50. The image selection device 20 may also acquire target images directly from the imaging device 30.

[0030] Next, a detailed example of the configuration of the image selection device 20 will be described.

[0031] In the example shown in FIG. 5, the image selection device 20 includes an acquisition unit 220 and a determination unit 230, as well as a high-precision image processing unit 210 and a storage processing unit 240.

[0032] The high-precision image processing unit 210 processes the target image to generate a second processing result. The high-precision image processing unit 210 generates the second processing result by processing the target image using, for example, a second machine learning model. The machine learning used in the second machine learning model is, for example, a neural network such as deep learning, bagging, boosting, or stacking, but is not limited to these. The high-precision image processing unit 210 may also generate the second processing result without using the second machine learning model. As an example, the high-precision image processing unit 210 may generate the second processing result by feature extraction or pattern matching.

[0033] The acquisition unit 220 and the determination unit 230 are as described with reference to FIG.

[0034] The storage processing unit 240 stores information capable of identifying a target image determined by the determination unit 230 to be used as training data for the first machine learning model in the second storage unit 50. This makes it possible to generate training data using the second storage unit 50. The information capable of identifying the target image may be the target image itself, or may be information indicating the imaging device 30 that generated the target image and the date and time of generation.

[0035] The storage processing unit 240 may also generate training data and store it in the second storage unit 50. For example, the storage processing unit 240 generates training data using a target image and the second processing result of the target image. An example of this training data includes the target image and at least a portion of the second processing result of the target image. At least a portion of the second processing result of the target image is used as at least a portion of the label of the target image. An example of at least a portion of the second processing result may be at least one of the position of an object and the position of an event in the target image, at least one of the type of object and the type of event, or both the position and the type. Note that the "position" used here does not need to be a point and may have a certain area. The label may also include information input by a user or developer of the image selection device 20. For example, the storage processing unit 240 acquires this information via a user interface of the image selection device 20.

[0036] As shown in FIG. 6, the image selection device 20 includes, as hardware, a bus 1010, a processor 1020, a memory 1030, a storage device 1040, an input / output interface 1050, and a network interface 1060, for example.

[0037] The bus 1010 is a data transmission path for transmitting and receiving data among the processor 1020, memory 1030, storage device 1040, input / output interface 1050, and network interface 1060. However, the method of connecting the processor 1020 and the like to each other is not limited to bus connection.

[0038] The processor 1020 is implemented by a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or the like.

[0039] The memory 1030 is a main storage device realized by a RAM (Random Access Memory) or the like.

[0040] The storage device 1040 is an auxiliary storage device realized by removable media such as a hard disk drive (HDD), a solid state drive (SSD), or a memory card, or a read-only memory (ROM), and has a recording medium. The recording medium of the storage device 1040 stores program modules that realize each function of the image selection device 20 (e.g., the high-precision image processing unit 210, the acquisition unit 220, the determination unit 230, and the storage processing unit 240). The processor 1020 loads each of these program modules into the memory 1030 and executes them, thereby realizing each function corresponding to the program module. The storage device 1040 may also function as at least one of the first storage unit 40 and the second storage unit 50.

[0041] The input / output interface 1050 is an interface for connecting the image selection device 20 to various input / output devices.

[0042] The network interface 1060 is an interface for connecting the image selection device 20 to a network. This network may be, for example, a local area network (LAN) or a wide area network (WAN). The network interface 1060 may connect to the network wirelessly or via a wired connection. The image selection device 20 may communicate with at least one of the image processing device 10, the imaging device 30, the first storage unit 40, the second storage unit 50, and the learning device 60 via the network interface 1060.

[0043] The image selection device 20 may be realized by a plurality of devices each having the configuration shown in FIG.

[0044] The hardware configurations of the image processing device 10 and the learning device 60 are similar to the hardware configuration of the image selection device 20 shown in FIG.

[0045] Next, a detailed example of the processing performed by the image selection device 20 will be described. In the example described below, the operation timing of the image selection device 20 is not synchronized with the operation timing of the image processing device 10. The image selection device 20 operates periodically to process unprocessed target images.

[0046] 7 , for example, the acquisition unit 220 of the image selection device 20 acquires an unprocessed target image and a first processing result associated with the target image from the first storage unit 40 (step S110). Next, the high-precision image processing unit 210 processes the target image acquired by the acquisition unit 220 to generate a second processing result (step S120). The acquisition unit 220 acquires this second processing result. Next, the determination unit 230 determines whether or not to use the currently processed target image as training data using the first processing result and the second processing result acquired by the acquisition unit 220 (step S130).

[0047] The determination unit 230 then updates the second storage unit 50 using the target image determined to be used as training data in step S130 (step S140). A specific example of the processing performed here is as described with reference to FIG. 5.

[0048] As described above, when using the image selection device 20, the determination unit 230 determines whether to use the target image as training data for the first machine learning model using the first processing result and the second processing result. This reduces the effort required to prepare the training data. Furthermore, when the storage processing unit 240 of the image selection device 20 stores the target image and at least a portion of the second processing result in the second storage unit 50 as at least a portion of the training data, the effort required to prepare the training data can be further reduced.

[0049] Second Embodiment The image selection device 20 shown in Fig. 8 includes a selection unit 250 in addition to the components of the image selection device 20 shown in Fig. 5. The selection unit 250 uses auxiliary information to select a target image from images generated by the imaging device 30. The high-precision image processing unit 210, acquisition unit 220, and determination unit 230 perform processing on the target image selected by the selection unit 250. In this manner, the processing load of the high-precision image processing unit 210, acquisition unit 220, and determination unit 230 can be reduced while maintaining the quality of the training data generated using the image selection device 20.

[0050] In this case, it is preferable that the first image processing unit 110 of the image processing device 10 processes all images generated by the imaging device 30.

[0051] Specific examples of auxiliary information are as follows: (1) The auxiliary information includes at least one of the location where the target image was taken, the timing of taking the image, and weather information at the time of taking the image. (2) The auxiliary information includes information input via a user interface. Note that the type of information input via a user interface may be, for example, at least one of the location where the target image was taken, the timing of taking the image, and weather information at the time of taking the image as shown in (1) above. (3) If the imaging device 30 generates an image by taking an image of a road, the auxiliary information includes information about the condition of this road.

[0052] The "photography location" in the auxiliary information described in (1) above is, for example, a photography location that satisfies a specific condition, i.e., the photography range of the imaging device 30. When multiple imaging devices 30 are arranged along a road, an example of the "photography location" used as auxiliary information is a location that satisfies at least one of the criteria: the number of accidents per unit period, the number of traffic jams per unit period, the presence of stopped vehicles, the number of detections of vehicles traveling in the wrong direction, the number of detections of vehicles traveling at or below a predetermined speed, the number of detections of vehicles traveling at or above the legal speed limit, the number of detections of vehicles attempting to avoid traffic, and the volume of moving objects such as vehicles. At least one of the number of accidents per unit period, the number of traffic jams per unit period, and the volume of moving objects such as vehicles used here is generated, for example, by processing images generated by the imaging device 30. This processing may be performed by the first image processing unit 110 of the image processing device 10 or the high-precision image processing unit 210 of the image selection device 20. An example of a specific moving object is at least one of a pedestrian, a bicycle, a motorcycle, and a specific vehicle such as a truck. The selection unit 250 then selects an image that satisfies the "photography location" as a target image. The photography location is specified using the identification information of the image processing device 10 that generated the image. This identification information is stored in the first storage unit 40, for example.

[0053] Furthermore, the "photography timing" of the auxiliary information shown in (1) above includes, for example, at least one of a specific month, a specific day of the week, a specific season, and a specific time period. An example of a specific day of the week is at least one of a weekday, a holiday, and a national holiday. An example of a specific time period is at least one of morning, evening, and night. The selection unit 250 then selects an image that satisfies this "photography timing" as the target image. The date and time of image generation is stored, for example, in the first storage unit 40.

[0054] Furthermore, among the auxiliary information described in (1) above, the "weather information at the time of shooting" indicates that, for example, an image captured under a specific weather condition is selected as the target image. Here, an example of the specific weather condition is at least one of rain, fog, snow, and strong wind. In this example, the selection unit 250 may acquire, as weather information for an image generated by the imaging device 30, weather information corresponding to the date and time of shooting and the location of shooting, from a server that stores past weather conditions. This server may acquire and store weather information corresponding to the date and time of shooting and the location of shooting from a sensor installed together with the imaging device 30. In this case, this server may be managed by an organization that manages the image processing device 10, the image selection device 20, and the learning device 60. This server may also be managed by a public institution such as the Japan Meteorological Agency.

[0055] In addition, information generated by a sensor that is less related to weather, such as a distance measurement sensor such as a LiDAR or ultrasonic sensor, an infrared sensor, or an infrared camera, may be used as auxiliary information.

[0056] The user interface described in (2) above also includes a field that displays auxiliary information in a selectable state to the user of the image selection device 20, such as a field that allows the user to select auxiliary information in a pull-down format, as shown in the screen of Fig. 9. The screen of Fig. 9 allows the user to select season, time zone, and weather information in a pull-down format.

[0057] Furthermore, the "information regarding road conditions" shown in (3) above includes, for example, at least one of the following: the degree of road congestion, the presence and type of traffic restrictions, the presence or absence of foreign objects such as fallen objects, and the type of road. The degree of road congestion at the time the image was generated may be determined, for example, using traffic congestion history information stored in an external server, or may be determined by processing the image generated by the imaging device 30. The type of road may include at least one of general roads, roads that only certain vehicles are permitted to use, such as expressways, and toll roads. This type of road may be linked in advance to, for example, the identification information of the imaging device 30.

[0058] The image selection device 20 shown in Fig. 8 operates, for example, as shown in Fig. 10. First, the selection unit 250 acquires auxiliary information and uses this auxiliary information to select a target image from the images stored in the first storage unit 40 (step S102). Subsequent processing, for example, steps S110 to S140, is as described using Fig. 7.

[0059] In addition, the storage processing unit 240 may display the auxiliary information used to select the target image on the display device viewed by the user of the image selection device 20, together with at least one of the image selected as the target image and the image not selected as the target image.

[0060] The first machine learning model used by the image processing device 10 may have reduced processing accuracy for images captured under specific conditions. In contrast, the image selection device 20 can select images that satisfy these specific conditions as target images, thereby increasing the amount of training data that satisfies these conditions. Therefore, using the image selection device 20 can reduce the weaknesses of the first machine learning model.

[0061] Furthermore, if you want to prioritize improving the processing accuracy of images that meet certain conditions, for example, if you want to improve the accuracy of vehicle detection in locations where the probability of accidents is high, you can use the image selection device 20 to prioritize training data that meets these specific conditions into the first machine learning model.

[0062] Third Embodiment In the image selection device 20 shown in Fig. 11, the determination unit 230 further uses auxiliary information when determining whether or not to use a target image as training data. Examples of the auxiliary information that can be used include (1) to (3) described with reference to Figs. 8 and 9.

[0063] In this case, the determination unit 230 determines, for example, whether the target image can be used as teacher data using the auxiliary information, and performs a determination process using the first and second processing results on the target image that is determined to be usable. As another example, the determination unit 230 determines, for example, whether the target image can be used as teacher data using the first and second processing results, and performs a determination process using the auxiliary information on the target image that is determined to be usable.

[0064] The "judgment process using auxiliary information" is similar to the process in which the selection unit 250 shown in Figure 8 selects a target image, and the conditions for selecting an image as a target image here are the conditions for judging that the image can be used as training data.

[0065] The auxiliary information used in the image selection device 20 further includes the following (4): (4) The auxiliary information is based on at least a part of the second processing result.

[0066] In this case, for example, the auxiliary information may be that the event or object being detected in the second processing result has been detected. In this case, the determination unit 230 determines whether or not the target image in which the event or object has been detected in the second processing result can be used as training data using the first processing result and the second processing result.

[0067] As another example, the auxiliary information may be a comparison result of the first processing result with the second processing result. For example, if the second processing result indicates that the target image contains multiple events or objects, and the ratio of the number of events or objects indicated by the first processing result to the number of multiple events or objects indicated by the second processing result (i.e., the detection rate of the first image processing unit 110 when the second processing result is considered correct) is equal to or less than a reference value, the determination unit 230 determines that the target image can be used as training data. For example, if the target image contains three objects but the first image processing unit 110 can only detect one object, the detection rate will be 1 / 3.

[0068] Furthermore, although related to the auxiliary information described in (4) above, the following (4-1) may also be used as an auxiliary condition. (4-1) Assume that an image is generated by an imaging device 30 photographing a road, and that multiple imaging devices 30 are arranged along this road as shown in FIG. 12. In a case where the second processing result corresponding to the imaging device 30a that generated the target image does not detect a moving object, the auxiliary information includes whether or not the second processing result corresponding to an image generated at a predetermined timing by an imaging device 30b different from the imaging device 30a has detected a moving object. Note that the predetermined timing used here is a timing determined based on the timing of generation of the target image, and is set, for example, by the administrator of the image selection device 20.

[0069] Specifically, the distance L between the image capturing device 30a and the image capturing device 30b is known. Also, the speed of a moving object between the image capturing device 30a and the image capturing device 30b, for example, the speed of a vehicle, is within a certain range V １ More than V ２ Therefore, the time from when the moving object appears in the image generated by the image capturing device 30a until when it appears in the image generated by the image capturing device 30b is L / V ２ More than L / V １ The following is likely to occur:

[0070] For example, suppose that the imaging device 30b is installed downstream of the imaging device 30a. Then, the high-precision image processing unit 210 determines that the target image generated by the imaging device 30a does not include a moving object. However, if the L / V ２ More than L / V １ Suppose that the high-precision image processing unit 210 determines that the image generated by the imaging device 30b contains a moving object some time later. In this case, the second processing result of the target image generated by the imaging device 30a may be incorrect. Therefore, the judgment unit 230 determines that this target image should be used as training data.

[0071] In this case, the storage processing unit 240 stores the target image in the second storage unit 50, but at this stage, a label for the training data is not generated. This label is set, for example, by a person who views the target image stored in the second storage unit 50.

[0072] In addition, when the imaging device 30b is installed upstream of the imaging device 30a, the image of the imaging device 30b used in this process is L / V from the timing when the target image is generated. ２ More than L / V １ Below is the image generated earlier.

[0073] This image selection device 20 also selects images that satisfy specific conditions as target images, thereby increasing the amount of training data that satisfies these conditions. Therefore, use of the image selection device 20 can reduce the weaknesses of the first machine learning model. Furthermore, use of the image selection device 20 allows the first machine learning model to preferentially learn training data that satisfies the specific conditions.

[0074] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0075] In addition, although the flowcharts used in the above description show multiple steps (processes) in a sequential order, the order of steps executed in each embodiment is not limited to the order shown. In each embodiment, the order of the steps shown in the drawings can be changed as long as it does not cause any problems in terms of content. Furthermore, the above-described embodiments can be combined as long as the content is not contradictory.

[0076] Some or all of the above embodiments may be described as, but are not limited to, the following supplementary notes. 1. An image selection device comprising: an acquisition means for acquiring a first processing result that is a result of processing a target image using a first machine learning model, and acquiring a second processing result that is a processing result by high-precision image processing means that can process the target image with higher accuracy than the first machine learning model; and a determination means for determining whether or not to use the target image as training data for the first machine learning model using the first processing result and the second processing result. 2. The image selection device described in above 1, comprising: a selection means for selecting the target image using auxiliary information. 3. The image selection device described in above 1, wherein the determination means further determines whether or not to use the target image as the training data using auxiliary information. 4. The image selection device described in above 1, comprising: a storage processing means for storing the target image that the determination means determines to use as the training data, and at least a portion of the second processing result for the target image, in a storage means, as at least a portion of the training data. 5. 5. The image selection device according to any one of claims 1 to 4, wherein the auxiliary information includes at least one of a shooting location of the target image, a shooting timing, and weather information at the time of shooting. 6. The image selection device according to any one of claims 1 to 4, wherein the target image is generated by shooting a road, and the auxiliary information includes information about the condition of the road. 7. The image selection device according to any one of claims 1 to 4, wherein the auxiliary information is based on at least a part of the second processing result. 8. The image selection device according to any one of claims 1 to 7, further comprising: learning means for updating the first machine learning model using the training data.9. The image selection device described in above 7, wherein the target image is generated by photographing a road, and a plurality of photographing means are arranged along the road, and the first processing result and the second processing result include the presence or absence of a moving object, and the auxiliary information includes, when the second processing result corresponding to the photographing means that generated the target image does not detect the moving object, whether the second processing result corresponding to an image generated by a photographing means other than the photographing means that generated the target image at a predetermined timing detects a moving object, and the predetermined timing is a timing determined based on the timing of generation of the target image. 10. The image selection device described in above 2 or 3, wherein the auxiliary information includes information input via a user interface. 11. The image selection device described in above 4, wherein the first processing result and the second processing result include the presence or absence of an object and its position, and at least a portion of the second processing result included in the training data includes the position of the object. 12. 12. The image selection device according to any one of claims 1 to 11, wherein the amount of processing required to generate the second processing result is greater than the amount of processing required to generate the first processing result, the first processing result is generated each time the target image is generated, and the second processing result is generated after the first processing result. 13. The image selection device according to any one of claims 1 to 12, further comprising the high-precision image processing means. 14. The image selection device according to any one of claims 1 to 13, further comprising first image processing means for generating the first processing result. 15. An image selection method, wherein at least one computer acquires a first processing result that is a result of processing a target image using a first machine learning model, and acquires a second processing result that is a processing result obtained by high-precision image processing that can process the target image with higher accuracy than the first machine learning model, and determines whether or not to use the target image as training data for the first machine learning model using the first processing result and the second processing result.16. The image selection method according to claim 15, wherein the at least one computer selects the target image using auxiliary information. 17. The image selection method according to claim 15, wherein the at least one computer further uses auxiliary information to determine whether or not to use the target image as the training data. 18. The image selection method according to claim 15, wherein the at least one computer stores the target image determined to be used as the training data and at least a portion of the second processing result for the target image in a storage means as at least a portion of the training data. 19. The image selection method according to claim 16 or 17, wherein the auxiliary information includes at least one of the shooting location, shooting timing, and weather information at the time of shooting of the target image. 20. The image selection method according to claim 16 or 17, wherein the target image is generated by photographing a road, and the auxiliary information includes information about the road conditions. 21. The image selection method according to claim 16 or 17, wherein the auxiliary information is based on at least a portion of the second processing result. 22. The image selection method according to any one of claims 15 to 21, wherein the at least one computer updates the first machine learning model using the training data. 23. The image selection method according to claim 21, wherein the target image is generated by photographing a road, and a plurality of image capturing means are arranged along the road, and the first processing result and the second processing result include the presence or absence of a moving object, and the auxiliary information includes, when the second processing result corresponding to the image capturing means that generated the target image does not detect the moving object, whether the second processing result corresponding to an image generated by an image capturing means other than the image capturing means that generated the target image at a predetermined timing detects a moving object, and the predetermined timing is a timing determined based on the timing of generation of the target image. 24. The image selection method according to claim 16 or 17, wherein the auxiliary information includes information input via a user interface.25. The image selection method described in any one of claims 15 to 25, wherein the first processing result and the second processing result include the presence or absence of an object and its position, and at least a portion of the second processing result included in the training data includes the position of the object. 26. The image selection method described in any one of claims 15 to 25, wherein the amount of processing required to generate the second processing result is greater than the amount of processing required to generate the first processing result, the first processing result is generated every time the object image is generated, and the second processing result is generated after the first processing result. 27. The image selection method described in any one of claims 15 to 26, wherein the at least one computer performs the high-precision image processing. 28. The image selection method described in any one of claims 15 to 27, wherein the at least one computer generates the first processing result. 29. 30. A program comprising: at least one computer having: acquisition means for acquiring a first processing result that is a result of processing a target image using a first machine learning model, and acquiring a second processing result that is a processing result using high-precision image processing that can process the target image with higher accuracy than the first machine learning model; and determination means for determining whether or not to use the target image as training data for the first machine learning model using the first processing result and the second processing result. 30. A program comprising: the at least one computer having selection means for selecting the target image using auxiliary information. 31. A program comprising: the program recited in claim 29, wherein the determination means further determines whether or not to use the target image as training data using auxiliary information. 32. A program recited in claim 29, wherein the at least one computer has storage processing means for storing the target image that the determination means determines to use as the training data, and at least a portion of the second processing result for the target image, in storage means, as at least a portion of the training data.33. The program described in any one of 30 or 31 above, wherein the auxiliary information includes at least one of the location where the target image was taken, the timing of taking the image, and weather information at the time of taking the image. 34. The program described in 30 or 31 above, wherein the target image is generated by taking an image of a road, and the auxiliary information includes information about the condition of the road. 35. The program described in 30 or 31 above, wherein the auxiliary information is based on at least a part of the second processing result. 36. The program described in any one of 29 to 35 above, wherein the at least one computer is provided with learning means that updates the first machine learning model using the training data. 37. 38. The program described in claim 30 or 31, wherein the auxiliary information includes information input via a user interface. 39. The program described in claim 32, wherein the first processing result and the second processing result include the presence or absence of a moving object, and when the second processing result corresponding to the imaging means that generated the target image does not detect the moving object, the auxiliary information includes whether the second processing result corresponding to an image generated by an imaging means other than the imaging means that generated the target image detects a moving object, and the predetermined timing is a timing determined based on the generation timing of the target image. 39. The program described in claim 32, wherein the first processing result and the second processing result include the presence or absence of a moving object and its position, and at least a portion of the second processing result included in the training data includes the position of the moving object. 40. 40. The program according to any one of claims 29 to 39, wherein the amount of processing required to generate the second processing result is greater than the amount of processing required to generate the first processing result, the first processing result is generated each time the target image is generated, and the second processing result is generated after the first processing result.41. The program according to any one of claims 29 to 40 above, wherein the at least one computer is provided with high-precision image processing means for performing the high-precision image processing. 42. The program according to any one of claims 29 to 41 above, wherein the at least one computer is provided with first image processing means for generating the first processing result. 43. A recording medium storing the program according to any one of claims 29 to 42 above.

[0077] REFERENCE SIGNS LIST 10 Image processing device 20 Image selection device 30 Imaging device 40 First storage unit 50 Second storage unit 60 Learning device 110 First image processing unit 210 High-precision image processing unit 220 Acquisition unit 230 Determination unit 240 Storage processing unit 250 Selection unit 610 Learning unit

Claims

1. an acquisition means for acquiring a first processing result that is a result of processing a target image using a first machine learning model, and acquiring a second processing result that is a processing result by a high-precision image processing means that can process the target image with higher accuracy than the first machine learning model; a determination means for determining whether or not to use the target image as training data for a first machine learning model using the first processing result and the second processing result; An image selection device comprising:

2. 2. The image selection device according to claim 1, An image selection device comprising a selection means for selecting the target image using auxiliary information.

3. 2. The image selection device according to claim 1, The determination means further uses auxiliary information to determine whether or not the target image is to be used as the training data.

4. 2. The image selection device according to claim 1, An image selection device comprising a storage processing means for storing the target image that the judgment means has determined to be used as the teacher data and at least a portion of the second processing result of the target image in a storage means as at least a portion of the teacher data.

5. 4. The image selection device according to claim 2, The auxiliary information includes at least one of the location where the target image was taken, the timing of taking the image, and weather information at the time of taking the image.

6. 4. The image selection device according to claim 2, the target image is generated by photographing a road, The auxiliary information includes information about the road conditions.

7. 4. The image selection device according to claim 2, The auxiliary information is based at least in part on the second processing result.

8. 8. The image selection device according to claim 7, the target image is generated by photographing a road, A plurality of photographing means are arranged along the road, the first processing result and the second processing result include the presence or absence of a moving object; the auxiliary information includes, when the second processing result corresponding to the photographing means that generated the target image does not detect the moving object, whether or not the second processing result corresponding to an image generated at a predetermined timing by the photographing means different from the photographing means that generated the target image detects a moving object; The predetermined timing is a timing determined based on a generation timing of the target image.

9. At least one computer Obtain a first processing result that is a result of processing the target image using a first machine learning model, and obtain a second processing result that is a processing result by high-precision image processing that can process the target image with higher accuracy than the first machine learning model; An image selection method that uses the first processing result and the second processing result to determine whether or not to use the target image as training data for a first machine learning model.

10. At least one computer an acquisition means for acquiring a first processing result that is a result of processing a target image using a first machine learning model, and acquiring a second processing result that is a processing result by high-precision image processing that can process the target image with higher accuracy than the first machine learning model; a determination means for determining whether or not to use the target image as training data for a first machine learning model using the first processing result and the second processing result; A program that allows you to have