Recognition device, recognition processing method, and program

Abstract

To facilitate performing the setting of recognition processing so as to perform the recognition processing in a recognition environment in which a recognition result similar to that of past recognition processing can be reproduced.SOLUTION: A recognition device manages a model used for recognition processing, acquires a result of first recognition processing performed using a first model on first recognition target information, performs recognition processing on the first recognition target information according to each of a plurality of settings using a model different from the managed first model to generate a result of the recognition processing corresponding to each of the plurality of settings, and outputs the setting corresponding to the result of the recognition processing selected by referring to the result of the first recognition processing from the result of the recognition processing corresponding to each of the plurality of settings.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a recognition device, a recognition processing method, and a program, and relates to, for example, a deformation detection process for an image of a structure. [Background technology]

[0002] Recognition devices and recognition services are available that recognize targets, such as predetermined objects, from information such as images. These recognition devices use large amounts of training data to train models such as CNNs (Convolutional Neural Networks), and can recognize targets from input information using the trained models.

[0003] For example, Patent Document 1 discloses a technology for detecting cracks using a learning model from images captured by an unmanned aerial vehicle. In Patent Document 1, cracks are detected using two types of models: a specialized learning model that can identify specific types of cracks, and a non-specialized learning model that can identify multiple types of cracks other than the specific types of cracks. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6807093 Summary of the Invention [Problem to be solved by the invention]

[0005] Users of recognition devices sometimes want to compare the results of recognition processing on past information (e.g., past images of a structure) with the results of recognition processing on current information (e.g., current images of a structure). For example, when inspecting a structure, the progression of deformation may be analyzed by comparing images. Specifically, the degree of progression of deformation can be evaluated by analyzing the extension or amplification of cracks in the structure. In this case, since the recognition results will differ if the model or processing parameters used in the recognition processing are different, it is desirable to perform the recognition processing in a recognition environment that will produce similar recognition results.

[0006] On the other hand, recognition device providers may update their models and discontinue older models, which may make it difficult to compare the results of recognition processing for information at different points in time, since the previously used models can no longer be used for recognition processing.

[0007] An object of the present invention is to facilitate the setting of recognition processing so that recognition processing can be performed in a recognition environment that can reproduce recognition results similar to those of past recognition processing. [Means for solving the problem]

[0008] A recognition device according to an embodiment of the present invention has the following configuration: a model management means for managing models used in recognition processing; an acquisition means for acquiring a result of a first recognition process performed on the first recognition target information using a first model; a recognition means for performing a recognition process on the first recognition target information in accordance with each of a plurality of settings using a model different from the first model managed by the model management means, and generating a recognition process result corresponding to each of the plurality of settings; an output means for outputting a setting corresponding to the result of the recognition processing selected from the results of the recognition processing corresponding to each of the plurality of settings by referring to the result of the first recognition processing; Equipped with. [Effects of the Invention]

[0009] It is possible to easily set up the recognition process so that the recognition process can be performed in a recognition environment in which a recognition result similar to that of a past recognition process can be reproduced. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram illustrating an example of the functional configuration of a recognition device according to an embodiment. [Figure 2] FIG. 1 is a block diagram illustrating an example of the hardware configuration of a recognition device according to an embodiment. [Figure 3] 1 is a flowchart of a recognition processing method according to an embodiment. [Figure 4] FIG. 10 is a diagram showing an example of a display screen showing the history of recognition processing. [Figure 5] FIG. 10 is a diagram showing an example of a user interface for setting a recognition process. [Figure 6] FIG. 10 is a diagram showing an example of a screen for proposing a search process. [Figure 7] FIG. 10 is a diagram showing an example of a screen displaying search results. [Figure 8] FIG. 3 is a diagram showing an example of history information managed by a history management unit 122. [Figure 9] 10 is a flowchart of a search process. [Figure 10] FIG. 4 is a diagram showing an example of classification information of models managed by a model management unit 106. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention claimed. Although multiple features are described in the embodiments, not all of these multiple features are necessarily essential to the invention, and multiple features may be combined arbitrarily. Furthermore, in the accompanying drawings, the same reference numerals are used to designate the same or similar components, and redundant explanations will be omitted.

[0012] FIG. 1 shows the functional configuration of a recognition device according to an embodiment of the present invention. Information 102 to be recognized (recognition target information) is input to the recognition device 101, and a recognition result 103 obtained by performing the recognition process on the recognition target information 102 is output. The type of object of the recognition process is not particularly limited, and may be, for example, an image, text information, audio information, or signal information. The type of recognition process is also not particularly limited, and may include, for example, a process for recognizing a recognition target from an image, a process for recognizing a summary from text information, or a process for recognizing an abnormality from a knocking sound of a structure or a sensor signal from an instrument. In the following embodiment, a case will be described in which the recognition device 101 recognizes a defect, such as a crack, that has occurred in a structure in an image. The structure may be, for example, an infrastructure structure such as a tunnel or a bridge.

[0013] The recognition device 101 according to this embodiment can be realized by a computer including a processor and a memory. FIG. 2 shows an example of the hardware configuration of the recognition device 101 according to one embodiment. A CPU (Central Processing Unit) 201 executes a control program for the recognition device 101. A ROM 202 stores the control program and the like. A RAM 203 provides a work area for the CPU 201. A network IF 204 is used to communicate with applications outside the recognition device 101. A storage device 205 is, for example, a hard disk, and stores data or programs. In this way, a processor such as the CPU 201 executes a program stored in a memory such as the ROM 202, the RAM 203, or the storage device 205, thereby realizing the functions of each unit shown in FIG. 1 and the like.

[0014] Next, the functional configuration of the recognition device 101 according to this embodiment will be described with reference to FIG. 1 The recognition device 101 according to this embodiment can easily search for settings for recognition processing that can obtain similar results when a model that was previously used for recognition processing has been discontinued.

[0015] figure 1As shown in the figure, the recognition device 101 includes a recognition unit 105, a model management unit 106, an output unit 108, and a search unit 120. The model management unit 106 manages models used in the recognition process.

[0016] The search unit 120 acquires the result of a first recognition process performed on the first recognition target information using a first model.

[0017] The recognition unit 105 performs recognition processing on the recognition target information. The recognition unit 105 can perform the recognition processing using a model managed by the model management unit 106. Furthermore, the recognition unit 105 can perform this recognition processing in accordance with set processing parameters. In this embodiment, the recognition unit 105 performs recognition processing on the first recognition target information in accordance with each of a plurality of settings using a model different from the first model managed by the model management unit 106, and generates recognition processing results corresponding to each of the plurality of settings. Furthermore, the recognition unit 105 can perform recognition processing on the second recognition target information. At this time, the recognition unit 105 can perform recognition processing in accordance with the recognition processing settings that are set by referring to the settings output by the output unit 108.

[0018] The output unit 108 outputs a setting corresponding to the result of the recognition processing selected from the results of the recognition processing corresponding to each of the multiple settings based on a comparison with the result of the first recognition processing. For example, the output unit 108 can output the model and processing parameters set by the search unit 120 corresponding to the result of the recognition processing selected by the determination unit 121 or in accordance with a user input. The output unit 108 can also output the result of the recognition processing managed by the history management unit 122.

[0019] The recognition device 101 may further include an information input unit 104, a setting unit 107, a search unit 120, a determination unit 121, and a history management unit 122. The information input unit 104 may acquire second recognition target information.

[0020] The setting unit 107 can set the recognition process. For example, the setting unit 107 can set the recognition process in accordance with a user input.

[0021] The search unit 120 can further create a plurality of settings to be used in the recognition process performed on the first recognition target information by the recognition unit 105. The search unit 120 can create a plurality of settings by combining models and processing parameters to be used based on a search method.

[0022] The determination unit 121 can determine the similarity of the results of the recognition processing. Specifically, the determination unit 121 can determine the similarity between the result of the recognition processing corresponding to each of the multiple settings and the result of the first recognition processing. Furthermore, the determination unit 121 can select a recognition processing result from the results of the recognition processing corresponding to each of the multiple settings according to the similarity. In this embodiment, the determination unit 121 calculates the similarity based on the difference between the two recognition results, as will be described later.

[0023] The history management unit 122 manages recognition target information and the results of the recognition process on the recognition target information. The history management unit 122 may manage the settings of the recognition process in association with the results of the recognition process. The settings of the recognition process may include information for identifying a model, such as a model ID, which is identification information for the model. The settings of the recognition process may also include processing parameters used in the recognition process using the model. The settings of the recognition process may also include information indicating the recognition target. In this way, the settings of the recognition process may indicate the contents of the recognition process, and may be called model information. The history management unit 122 may also manage the recognition target information that was the subject of the recognition process.

[0024] In the following example, the first recognition target information and the second recognition target information are the same type of information or information about the same type of object, such as a first image and a second image of a structure. The first image and the second image may be images obtained by capturing images of the same structure at different times. For example, the second image may be captured several years after the first image.

[0025] In the following example, it is possible to obtain recognition processing settings using a model managed by the model management unit 106, which can reproduce recognition results similar to those of a first recognition processing performed in the past using a first model on a first image. Such settings can indicate the model and processing parameters used in the recognition processing, which will be referred to hereinafter as similar models and similar parameters. Then, recognition processing is performed on a second image according to the recognition processing settings set with reference to the obtained recognition processing settings. This processing allows recognition processing on the second image to be performed in a recognition environment that can reproduce recognition results similar to those of the first recognition processing performed in the past. Such similar models and similar parameters can be used for recognition processing on images for the purpose of comparison over time.

[0026] Next, the recognition process performed by the recognition device according to this embodiment will be described with reference to the flowchart in Fig. 3. In S301, the information input unit 104 acquires recognition target information. In this example, the information input unit 104 acquires a second image.

[0027] In S302, the setting unit 107 checks whether the first model previously used in the recognition process for the first image is available. In one embodiment, the output unit 108 can present the user with the history of the recognition process managed by the history management unit 122. For example, the output unit 108 can present the user with a list of the execution history of the recognition process as shown in FIG. 4. The output unit 108 may also present the user with the settings for each recognition process. In FIG. 4, a list 401 indicates the execution date and time of the recognition process, the input image as the recognition target information, the identification information of the model used in the recognition process, and the processing parameters used in the recognition process. The processing parameters may include, for example, a detection amount parameter that controls the amount of the recognition target detected in the recognition process (e.g., making the recognition target more or less detectable). The processing parameters may also include a noise reduction parameter that indicates the strength of the noise reduction process used as preprocessing for the recognition target information. However, the types of processing parameters are not limited to these.

[0028] This list 401 further includes a notification 402 indicating that a model previously used in recognition processing has been discontinued and is no longer available. In this way, the model management unit 106 can notify the user of whether a model is available on the execution history list via the output unit 108. The model management unit 106 may also notify the user of whether a model is available on another screen, such as the top screen of the recognition processing system or a model management screen.

[0029] The user can identify from the execution history the history of a recognition process that recognizes the same recognition target for an image of the same structure as the second image. The recognition process identified in this way is the first recognition process. The user can also check whether the model used in the first recognition process for the first image is usable based on the model identification information and the model's obsolete status displayed in the list of execution history.

[0030] In this embodiment, the user checks whether the model used in the first recognition process is available, and if the first model is available, the setting unit 107 acquires a user instruction indicating that recognition process should be performed using the first model, and then the process proceeds to S303. On the other hand, if the first model is unavailable, the setting unit 107 acquires a user instruction indicating that model search process should be performed, and then the process proceeds to S304.

[0031] On the other hand, the setting unit 107 may acquire a user instruction to select a first recognition process from the history of presented recognition processes. In this case, the setting unit 107 refers to the model management unit 106 and selects the first recognition process designated by the user. For use It can be checked whether the entered first model is available. If the first model is available, the process proceeds to S303, and if the first model is not available, the process proceeds to S304.

[0032] In S303, the setting unit 107 acquires the settings for the recognition process for the second recognition target information, which have been set with reference to the settings for the first recognition process. In one embodiment, the user can set the recognition process with reference to the settings for the first recognition process. FIG. 5 shows an example of a user interface (UI) for setting the model and processing parameters. This UI includes a list 501 for selecting a recognition target, a list 502 for selecting a model to be used in the recognition process, and process and a slider 503 for setting parameters. One or more models exist for each recognition target.

[0033] However, the setting unit 107 may set the recognition process for the second recognition target information so that the same process as the first recognition process is performed. Also, the initial values ​​in the UI as shown in Fig. 5 may be set according to the settings in the first recognition process. Furthermore, the setting unit 107 may automatically set the recognition process in consideration of the attributes of the recognition target information (for example, the acquisition time period, the brightness of the image, or GPS information) and the type of the recognition target.

[0034] From S304 onwards, a search for similar models and similar parameters is performed. Fig. 6 shows an example of a UI for suggesting to the user the search for similar models and similar parameters and for accepting user instructions before proceeding to S304. Such a UI can be displayed in response to the user selecting a recognition processing history using an unavailable model in S302. Alternatively, such a UI may be displayed in response to the user selecting an unavailable model in S303.

[0035] In S304, the output unit 108 presents to the user the history of the recognition processes managed by the history management unit 122. In addition, in S305, the setting unit 107 acquires a user instruction to select the first recognition process from the presented history of the recognition processes. 4 The processing in S304 and S305 can be performed as described for S302. Note that if the first recognition processing has already been selected in S302 or the like, S304 and S305 can be omitted.

[0036] In S306, the search unit 120 acquires the result of the first recognition processing performed on the first recognition target information using the first model from the history management unit 122. That is, the search unit 120 can acquire the recognition result managed by the history management unit 122 in association with the history selected in S305. Furthermore, the search unit 120 can acquire the first recognition target information from the history management unit 122. For example, the search unit 120 can acquire the first image managed by the history management unit 122 in association with the history selected in S305. Note that in another embodiment, the search unit 120 may acquire the first recognition target information and the result of the first recognition processing that have been exported.

[0037] In S307, the search unit 120 and the determination unit 121 search for similar models and similar parameters based on the result of the first recognition processing and the first recognition target information acquired in S306. Specifically, a result of the recognition processing corresponding to each of the multiple settings is selected based on a comparison with the result of the first recognition processing. As will be described later, the determination unit 121 can select a result of the recognition processing based on the similarity with the result of the first recognition processing. Then, the model and processing parameters used to obtain the selected result of the recognition processing are determined to be similar models and similar parameters. Detailed processing will be described later.

[0038] In S308, the output unit 108 outputs the search result in S307. In this example, the output unit 108 presents to the user the result of the recognition process selected by the determination unit 121 from the results of the recognition processes corresponding to each of the multiple settings.

[0039] FIG. 7 shows an example of a UI used to present search results to the user. Area 701 shows the result of the first recognition processing acquired in S306, and area 702 shows the result of the recognition processing selected by the determination unit 121. The output unit 108 can output settings corresponding to the result of the recognition processing selected in S307. The settings output by the output unit 108 may include information identifying a model used in the recognition processing. The settings output by the output unit 108 may also include processing parameters used in the recognition processing. For example, area 702 also shows the similar model and similar parameters selected by the determination unit 121. The output unit 108 may also output the similarity between the result of the recognition processing and the result of the first recognition processing, determined in S307, together with the result of the recognition processing.

[0040] In S309, the settings for the recognition process are performed with reference to the settings output by the output unit 108. For example, the setting unit 107 accepts a user input indicating the settings for the recognition process to be performed on the second recognition target information. Here, the user can set the recognition process with reference to the similar model and similar parameters output in S308. The setting for the recognition process can be performed in the same manner as in S303. For example, the user can set the recognition process using the UI shown in FIG. 5. The user may set the similar model and similar parameters output in S308. Meanwhile, the user may further adjust the processing parameters. Furthermore, the output unit 108 may present to the user a UI including an area in which the output unit 108 presents the settings and an area in which the output unit 108 accepts a user input indicating the settings for the recognition process to be performed on the second recognition target information. In this case, the user can set the recognition process on such a UI.

[0041] As another example, the setting unit 107 may automatically set the recognition process to be performed on the second recognition target information in accordance with the setting output by the output unit 108 in S307. For example, the setting unit 107 can set the second recognition process so as to use the similar model and similar parameters output in S308.

[0042] In S310, the recognition unit 105 performs recognition processing on the second recognition target information in accordance with the settings of the recognition processing set in S303 or S309. For example, the recognition unit 105 performs recognition processing on the image acquired in S301 using the set model and processing parameters.

[0043] In S311, the history management unit 122 stores the history of the recognition process for the second recognition target information in S310. Here, the history management unit 122 can store the model and processing parameters used in the recognition process, the second recognition target information, and the recognition results. Fig. 8 shows an example of history information stored in the history management unit 122. As shown in Fig. 8, the recognition date and time, identification information of the image which is the recognition target information, identification information of the model, and processing parameters are stored in association with an execution ID which identifies the history. Note that the information stored in the history management unit 122 history The information may include information indicating the recognition target or identification information of the user who performed the recognition process.

[0044] In S312, the output unit 108 10 The output unit 108 outputs the results of the recognition process obtained by the above process. In this embodiment, the output unit 108 can superimpose the positions of recognized defects such as cracks on the image, which is the recognition target information, and output the superimposed image. The output unit 108 may also output vector data indicating cracks. However, as described above, the recognition process performed by the recognition device according to the present invention is not particularly limited, and the output of the output unit 108 can take various forms. The results of such recognition process can be displayed on a display device such as a monitor, so that the user can check them.

[0045] Next, the search process performed by the search unit 120 and the determination unit 121 in S307 will be described with reference to the flowchart in Fig. 9. As described above, in S305, the first recognition process is selected, and in S306 the search unit 120 acquires the first recognition result and the first image stored in the history management unit 122. The search unit 120 searches for a similar model and similar parameters based on the first recognition result and the first image.

[0046] In S901, the search unit 120 generates settings for recognition processing. In S902, the recognition unit 105 performs recognition processing on the first recognition target information in accordance with the settings generated in S901. In S903, the determination unit 121 determines the similarity between the recognition result obtained in S902 and the first recognition result. In S904, the search unit 120 determines whether to continue the search. If the search is to be continued, the process returns to S901, and settings for new recognition processing are generated. If the search is to be ended, the process proceeds to S905.

[0047] By repeating the processes of S901 to S904, the search unit 120 can generate multiple settings for recognition processing. Each of the multiple settings can have a different combination of models and processing parameters. For example, the search unit 120 can generate multiple settings that use each of the multiple models managed by the model management unit 106. Furthermore, the search unit 120 can generate multiple settings that use each of different processing parameters.

[0048] The search unit 120 can generate settings including models and processing parameters based on a search method. For example, in S901, the search unit 120 may generate settings using various processing parameters while using one fixed model. As another method, the search unit 120 may generate settings using various models while using one fixed processing parameter (e.g., initial setting parameters). The search method used by the search unit 120 is not particularly limited, and may be, for example, a full search, a grid search, a random search, a Bayesian optimization, or another search method.

[0049] In one embodiment, the search unit 120 generates a setting to use a model associated with the first model among the models managed by the model management unit 106. In this case, the model management unit 106 can manage the association between models. In this way, by narrowing the search range of the setting to use only models associated with the first model, the time required for the search can be reduced. For example, the model management unit 106 can manage information specifying one or more models as successor models of the first model. In this case, the search unit 120 can generate a setting to use the successor model.

[0050] FIG. 10 shows an example of model classification information managed by the model management unit 106. The model management unit 106 can register model identification information such as the model name, the recognition target of the model, and information indicating the model's lineage. The lineage is a value assigned by the model registrant to classify the model. The model management unit 106 may also manage other information related to the model, such as information identifying users who can use the model or the registration date and time of the model. Based on such model classification information, the search unit 120 can generate a setting to use only models associated with the first model. For example, the search unit 120 can search for a setting to use a model that has the same recognition target or lineage as the first model, or both.

[0051] In S905, the search unit 120 determines the recognition result with the highest similarity calculated in S903 from among the recognition results obtained according to each of the multiple settings generated by repeating S902. The model and parameters used to obtain this recognition result are the similar model and similar parameters. The similar model and similar parameters determined in S905 are output by the output unit 108 in S308.

[0052] The method for determining the similarity in S903 will be described. In S903, the determination unit 121 estimates the difference between the first recognition result and the recognition result obtained in S902, and can acquire this estimated value as the similarity. When detecting anomalies from images as in this embodiment, it is conceivable that an image indicating the position of the anomaly will be output as the recognition result. In such a case, the difference between the two images indicating the respective recognition results can be taken, and the similarity can be calculated based on the amount of difference.

[0053] As a specific example, the similarity can be calculated as follows: One image contains a predetermined number of pixels. Therefore, the difference between images is calculated by finding the absolute value of the density difference between corresponding pixels in two images to be compared. absolute In this case, it can be considered that the smaller the difference, the higher the similarity, and the larger the difference, the lower the similarity. For example, if the density of each pixel is expressed as 0 to 1, and the density of each pixel of the recognition result A is A(a 0,0 , a 0,1 , …, a 0,n , a 1,0 , …, a n,n ), the density of each pixel of the recognition result B is B(b 0,0 , b 0,1 , …, b 0,n , b 1,0 , …, b n,n In this case, the similarity between the recognition result A and the recognition result B can be calculated according to the formula (1).

number

[0054] On the other hand, because defects such as cracks are partial features in image data, even if the cracks are represented by the same vector data, a slight misalignment may be detected as a difference. For this reason, a method of calculating the difference that takes misalignment into account, such as expanding the crack, may be adopted. Furthermore, the vector data representing the results of the recognition process may be converted into feature data suitable for comparison, and the similarity may be calculated by calculating the distance between the feature data.

[0055] In the above embodiment, the recognition result with the highest similarity was determined in S905, and the model and parameters used to obtain this recognition result were output as the similar model and similar parameters in S308. On the other hand, the most desirable setting for a user to obtain a recognition result similar to the first recognition processing result may not be the setting that provides the highest similarity. For example, this possibility is likely to arise when the recognition result is expressed as visual information or multiple item values, or when there are multiple settings that provide a high similarity. Therefore, in S905, the search unit 120 may select two or more recognition results. For example, the search unit 120 may select a predetermined number of recognition results that have a higher similarity than other settings.

[0056] In this case, in S308, the search unit 120 may output each of the two or more recognition results selected in S905, instead of outputting the recognition result with the highest similarity. In this case, the search unit 120 may output the combination of models and parameters used to obtain each of the two or more recognition results. With this configuration, the user can select a recognition result that meets the user's intention from multiple recognition results with high similarity, and further, in S309, set similar models and similar parameters that will obtain a recognition result that meets the user's intention. Alternatively, in S309, the user may provide a user input indicating a recognition result that meets the user's intention from the recognition results output in S308. In this case, the setting unit 107 can set the combination of models and parameters used to obtain the recognition result indicated by the user input as the setting for the recognition process to be performed on the second recognition target information.

[0057] In the above embodiment, in S902, recognition processing is performed on the recognition target information used in past recognition processing managed by the history management unit 122. Furthermore, in S903, the results of past recognition processing managed by the history management unit 122 are compared with the recognition result obtained in S902. On the other hand, in S902, recognition processing may be performed on other recognition target information. For example, the first recognition target information used in S902 may be a sample image. This sample image is prepared in advance as a general image to be input into recognition processing. In this case, in S902, the recognition unit 105 can perform recognition processing on the sample image in accordance with the settings generated in S901.

[0058] In such an embodiment, the history management unit 122 can manage the results of recognition processes performed using each of multiple models on common recognition target information such as a sample image. For example, the history management unit 122 can manage the results of a first recognition process performed on a sample image using a first model. Then, in S903, the determination unit 121 can compare the results of the first recognition process performed on the sample image using the first model, which are managed by the history management unit 122, with the recognition result obtained in S902.

[0059] On the other hand, with a configuration that uses sample images in this way, it is possible to search for similar models and similar parameters corresponding to the first model in advance without waiting for a user instruction to search for similar models and similar parameters. Therefore, for example, by performing such a search when a model is discontinued, it is possible to quickly present similar models and similar parameters that generally correspond to the discontinued model.

[0060] Such a configuration can be realized by performing the processing shown in Fig. 9 when the first model is to be discontinued or when it has been decided that the first model will be discontinued. The model management unit 106 can manage whether a model can be used, and discontinuing a model means that the model management unit 106 sets the model to be unusable.

[0061] In this case, before performing the process of S901, the recognition unit 105 first performs a first recognition process using a first model on a sample image. The processing parameters used at this time may be parameters according to initial settings or may be set by the user. Then, the processes of S901 to S905 are performed using the sample image as first recognition target information. Thus, in S905, a recognition result that has a high degree of similarity to the result of the first recognition process using the first model is selected. Then, the output unit 108 can output the settings used to obtain the recognition result selected in S905, and the model management unit 106 can manage these settings in association with the identification information of the first model. For example, the model management unit 106 can store the similar model and similar parameters used to obtain the recognition result selected in S905 in association with the first model to be discontinued.

[0062] As described above, in response to making the first model unavailable, the model management unit 106 can store the settings output by the output unit 108 in association with the identification information of the first model.

[0063] 3 when the model management unit 106 manages similar models and similar parameters associated with the first model, the recognition process according to Fig. 3 can be performed as follows: That is, instead of performing the processes of S306 to S308, the output unit 108 can refer to the model management unit 106 and output similar models and similar parameters associated with the first model used in the first recognition process selected in S305.

[0064] 10, when the model management unit 106 manages model classification information such as the recognition target or family of the model, a sample image may be prepared for each model classification defined by the recognition target or family, etc. In this case, in S902, the recognition unit 105 can refer to the model management unit 106 and acquire a sample image corresponding to the first model classification.

[0065] The above-described recognition device may be configured by a plurality of devices. For example, a recognition device according to an embodiment may be configured by a plurality of information processing devices connected via a network. In particular, when the recognition device 101 is a server that receives requests from user terminals, the function of the recognition unit 105 may be executed by another server. With such a configuration, even if the processing of the recognition unit 105 takes time, the processing of the recognition unit 105 can be performed asynchronously with the request. Furthermore, in FIG. 1 The functions of the information processing device as shown in can be realized by a computer, but some or all of the functions of the information processing device may be realized by dedicated hardware.

[0066] The recognition unit 105 can take various forms, such as a library in the recognition device 101, or a server or cloud service separate from the recognition device 101.

[0067] (Other Examples) The present invention can also be realized by supplying a program that realizes one or more functions of the above-described embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program. It can also be realized by a circuit (e.g., ASIC) that realizes one or more functions.

[0068] The invention is not limited to the above-described embodiments, and various changes and modifications can be made without departing from the spirit and scope of the invention. Accordingly, the following claims are appended to apprise the public of the scope of the invention. [Explanation of symbols]

[0069] 101: Recognition device, 104: Information input unit, 105: Recognition unit, 106: Model management unit, 107: Setting unit, 108: Output unit, 120: Search unit, 121: Determination unit, 122: History management unit

Claims

1. a model management means for managing models used in recognition processing; an acquisition means for acquiring a result of a first recognition process performed on the first recognition target information using a first model; a recognition means for performing a recognition process on the first recognition target information in accordance with each of a plurality of settings using a model different from the first model managed by the model management means, and generating a recognition process result corresponding to each of the plurality of settings; an output means for outputting a setting corresponding to a result of a recognition process selected from the results of the recognition processes corresponding to each of the plurality of settings by referring to the result of the first recognition process; A recognition device comprising:

2. 2. The recognition device according to claim 1, further comprising a determination means for determining a similarity between a result of the recognition processing corresponding to each of the plurality of settings and a result of the first recognition processing, and selecting a result of the recognition processing from the results of the recognition processing corresponding to each of the plurality of settings in accordance with the similarity.

3. 3. The recognition device according to claim 2, wherein said output means outputs said similarity together with the result of said recognition processing.

4. 4. The recognition device according to claim 1, wherein the recognition means performs recognition processing using a model associated with the first model among models managed by the model management means.

5. The recognition device according to claim 1 , wherein the recognition process according to each of the plurality of settings is a recognition process using each of a plurality of models.

6. The recognition device according to claim 1 , wherein the recognition processing according to each of the plurality of settings is a recognition processing using different processing parameters.

7. 7. The recognition device according to claim 1, wherein the settings output by the output means include information for identifying a model used in recognition processing.

8. 8. The recognition device according to claim 1, wherein the settings output by the output means include processing parameters used in recognition processing.

9. further comprising a history management means for managing the recognition target information and the results of the recognition processing for the recognition target information; the acquisition means acquires a result of the first recognition processing managed by the history management means, The recognition means performs a recognition process on the first recognition target information managed by the history management means.

9. A recognition device according to claim 1, wherein the recognition device is a recognition device for detecting a plurality of objects.

10. 10. The recognition device according to claim 9, wherein the result of said first recognition process is designated by a user from the results of the recognition processes managed by said history management means.

11. 11. The recognition device according to claim 1, wherein the recognition means further performs a recognition process on second recognition target information in accordance with settings made by referring to the settings output by the output means.

12. 12. The recognition device according to claim 1, wherein the model management means manages results of recognition processes performed using a plurality of models on common recognition target information.

13. 13. The recognition device according to claim 1, wherein the model management means manages the settings output by the output means in association with the identification information of the first model.

14. 14. The recognition device according to claim 13, wherein the model management means, in response to making the first model unavailable, stores the setting output by the output means in association with identification information of the first model.

15. A recognition processing method performed by a recognition device, acquiring a result of a first recognition process performed on the first recognition target information using the first model; performing a recognition process on the first recognition target information in accordance with each of a plurality of settings using a model different from the first model among models managed as models to be used in the recognition process, and generating a recognition process result corresponding to each of the plurality of settings; outputting a setting corresponding to the result of the recognition processing selected from the results of the recognition processing corresponding to each of the plurality of settings with reference to the result of the first recognition processing; A recognition processing method comprising:

16. A program for causing a computer to function as the recognition device according to any one of claims 1 to 14.

Images