KR102755547B1 - Image detection system and method for improving detection performance - Google Patents

Abstract

The present invention relates to an image detection system and method with improved detection performance, and more particularly, to a technology capable of improving detection accuracy for a specific image without increasing search time due to big data in an AI image analysis model based on big data.

Description

Image detection system and method for improving detection performance {Image detection system and method for improving detection performance}

The present invention relates to an image detection system and method with improved detection performance, and more particularly, to an image detection system and method with improved detection performance capable of improving the accuracy of image detection and minimizing the increase in detection time due to big data by applying a plurality of AI image analysis models.

As artificial intelligence (AI) technology advances, various services utilizing these AI analysis/detection models are increasing.

In particular, it is widely used as a service that detects specific objects based on image recognition of objects such as human face images or car license plates.

This method of detecting a specific object using AI image recognition technology detects the object by comparing the characteristic data of the specific object and comparing the degree of matching.

For example, for a method of detecting a specific object, first, an AI analysis model suitable for each object to be detected is used to analyze the characteristics of each object existing in various forms, and the analyzed characteristics are stored in a database in the form of metadata. After that, the characteristic data of the specific object to be found is compared with the characteristic data of the objects stored in the database, and the similarity (correspondence) between them is analyzed, and the specific object is detected based on the high similarity.

However, there is a problem that it is difficult to say that it has high reliability because the time to detect specific image data in the database varies depending on the number of feature data of objects existing in the database and the AI image analysis model used, and the accuracy of whether the detected data is actually the specific object being searched for also varies.

Accordingly, in order to improve accuracy and reliability, which are problems when using a single AI image analysis model in the past, various AI image analysis models were combined and used. In this case, although there is a clear advantage of higher accuracy than when applying a single AI image analysis model, there is a problem in that expensive analysis equipment is required and analysis time increases because the amount of analysis increases.

Domestic Publication Patent No. 10-2019-0093799 (Publication Date 2019.08.12.)

The present invention has been made to solve the problems of the prior art as described above, and an object of the present invention is to provide an image detection system and method with improved detection performance that can improve image recognition and detection performance without improving computing performance or increasing analysis time while applying a plurality of AI image analysis models.

In an image detection system with improved detection performance according to one embodiment of the present invention, the system comprises an image management unit (100) that receives a plurality of image data from the outside in advance, stores and manages each image data by assigning unique identification information to each image data, and includes one pre-stored AI image analysis model, and a first analysis unit (200) that analyzes the characteristics of each image data by the image management unit (100) and specific image data to be analyzed input from the outside, a first detection unit (300) that compares and detects the characteristic data of each image data with the characteristic data of the specific image data using the analysis result of the first analysis unit (200), an image derivation unit (400) that derives at least one image data from among the plurality of image data by the image management unit (100) based on the detection result of the first detection unit (300), and at least another pre-stored AI image analysis model, and a second analysis unit (500) that analyzes the derived image data by the image derivation unit (400) and the characteristics of the specific image data, and uses the analysis result of the second analysis unit (500) to derive the derived image data. It includes a second detection unit (600) that compares and detects characteristic data with characteristic data of the specific image data, and it is preferable that the first analysis unit (200) and the second analysis unit (500) each include different AI image analysis models.

Furthermore, it is preferable that the first analysis unit (200) includes a basic analysis unit (210) that receives a plurality of image data transmitted by the image management unit (100) using the AI image analysis model and analyzes the characteristics of each image data, a basic storage unit (220) that matches the analysis results of the basic analysis unit (210) with the unique identification information assigned to each image data and stores and manages them, and a basic specific analysis unit (230) that analyzes the characteristics of the specific image data using the AI image analysis model.

Furthermore, it is preferable that the first detection unit (300) apply a pre-stored comparison technique to compare the characteristic data of each image data by the basic storage unit (220) with the characteristic data of the specific image data by the basic specific analysis unit (230), thereby detecting image data having a degree of consistency with the characteristic data of the specific image data according to a predetermined standard, and determining unique identification information corresponding to the detected image data.

Furthermore, it is preferable to include a detailed analysis unit (510) that receives the derived image data by the image derivation unit (400) using one or more other AI image analysis models of the second analysis unit (500) and analyzes the characteristics of each derived image data for each AI image analysis model, a detailed storage unit (520) that matches the analysis results by the detailed analysis unit (510) with the unique identification information assigned to each derived image data and stores and manages them for each AI image analysis model, and a detailed specific analysis unit (530) that analyzes the characteristics of the specific image data for each AI image analysis model using the one or more other AI image analysis models.

Furthermore, it is preferable that the second detection unit (600) apply a pre-stored comparison technique to compare the characteristic data of each derived image data by the detailed storage unit (520) with the characteristic data of the specific image data by the detailed specific analysis unit (530) for each AI image analysis model, analyze the degree of agreement between the specific image data and the derived image data, determine a unique identification number of each specific image data, and match the analyzed degree of agreement with the determined unique identification number for each specific image data to generate final detection result information.

In another embodiment of the present invention, a method for image detection with improved detection performance, wherein each step is performed by a computer-implemented image detection system with improved detection performance, comprises: a first input step (S100) in which an image management unit receives a plurality of image data in advance, stores and manages each image data by assigning unique identification information to each image data, a second input step (S200) in which a first analysis unit receives specific image data to be analyzed from the outside, a first analysis step (S300) in which the first analysis unit analyzes the characteristics of each image data by the first input step (S100) and the specific image data by the second input step (S200) using any one of the pre-stored AI image analysis models, a first comparison step (S400) in which the first detection unit compares and detects the degree of correspondence between each image data and the specific image data using the analysis result of the first analysis step (S300), and an image derivation unit detects at least one of the plurality of image data by the first input step (S100) based on the comparison detection result of the first comparison step (S400). A method for producing image data, comprising: a derivation step (S500) for deriving one image data; a second analysis step (S600) for analyzing, in a second analysis unit, the characteristics of the derived image data by the derivation step (S500) and the specific image data by the second input step (S200) for each AI image analysis model using one or more other pre-stored AI image analysis models; and a second comparison step (S700) for comparing and detecting the degree of consistency between the derived image data and the specific image data using the analysis result of the second analysis step (S600) in a second detection unit, wherein the AI image analysis model by the first analysis step (S300) and the AI image analysis model by the second analysis step (S600) are preferably different AI image analysis models.

Furthermore, it is preferable that the derivation step (S500) derives image data having a certain standard of consistency among a plurality of image data by the first input step (S100) based on the comparison detection result by the first comparison step (S400).

Furthermore, it is preferable that the second analysis step (S600) analyzes the characteristics of the image data derived by the derivation step (S500) and the specific image data derived by the second input step (S200) for each AI image analysis model using one or more AI image analysis models.

Furthermore, it is preferable that the image detection method with improved detection performance further includes a final result step (S800) in which, after performing the second comparison step (S700), the second detection unit determines a unique identification number of each specific image data, and matches the unique identification number determined for each specific image data with the degree of correspondence of each specific image data according to the comparison detection result by the second comparison step (S700) to generate final detection result information.

The image detection system and method of the present invention with improved detection performance by the above configuration have the advantage of improving detection accuracy without improving computing performance or increasing analysis time/amount of analysis by sequentially applying multiple AI image analysis models in stages while utilizing multiple AI image analysis models.

Figure 1 is a configuration example diagram showing a conventional image detection system that applies multiple AI image analysis modules.
FIG. 2 is a configuration example diagram showing an image detection system with improved detection performance according to one embodiment of the present invention.
FIG. 3 is a flowchart illustrating an image detection method with improved detection performance according to one embodiment of the present invention.

Hereinafter, with reference to the attached drawings, the image detection system and method of the present invention with improved detection performance will be described in detail. The drawings introduced below are provided as examples so that those skilled in the art can sufficiently convey the spirit of the present invention. Accordingly, the present invention is not limited to the drawings presented below and may be embodied in other forms. In addition, the same reference numerals represent the same components throughout the specification.

In this case, if there is no other definition for the technical and scientific terms used, they have the meaning commonly understood by a person of ordinary skill in the art to which this invention belongs, and the description of well-known functions and configurations that may unnecessarily obscure the gist of the present invention in the following description and attached drawings are omitted.

In addition, a system refers to a set of components including devices, mechanisms, and means that are organized and interact regularly to perform necessary functions.

The method of recognizing and detecting images of specific objects using artificial intelligence models is widely used. For example, as shown in Fig. 1a), a plurality of image data extracted for facial image recognition are given independent identification information for each identification, and characteristic information for each image data is extracted through an applied AI image analysis model, and this is stored and managed in a database. After extracting characteristic information for specific image data to be detected using the same AI image analysis model, it is compared and analyzed with the characteristic information being stored and managed, and similar characteristic information is derived and the corresponding image data is detected.

In this case, it is difficult to have high reliability for the derived similarity by using a single AI image analysis model.

In order to solve these problems, conventionally, as shown in b) of Fig. 1, image data detection is performed by combining two or more AI image analysis models rather than a single AI image analysis model. In this case, the accuracy can naturally be improved compared to a single AI image analysis model, but there is a problem in that the amount of analysis and analysis time increase accordingly and an expensive analysis device is required.

Accordingly, an image detection system and method with improved detection performance according to one embodiment of the present invention solve the problems of the above-described conventional art, and relate to a technique for applying a plurality of AI image analysis models in two stages, so as to improve image recognition and detection performance without improving computing performance or increasing analysis time while applying a plurality of AI image analysis models.

FIG. 2 is a configuration example diagram showing an image detection system with improved detection performance according to one embodiment of the present invention. Referring to FIG. 2, the image detection system with improved detection performance according to one embodiment of the present invention will be described in detail.

An image detection system with improved detection performance according to one embodiment of the present invention is preferably configured to include an image management unit (100), a first analysis unit (200), a first detection unit (300), an image derivation unit (400), a second analysis unit (500), and a second detection unit (600), as shown in FIG. 2. It is preferable that each of the components is individually or integrally included in at least one operation processing means including a computer to perform operations.

Let's take a closer look at each component:

It is preferable that the above image management unit (100) store and manage a large number of various image data input from outside in advance.

The above image data is data extracted and input through image information or video information, etc., in order to detect a facial image, and it is preferable for the image management unit (100) to store and manage each image data by assigning unique identification information (such as an ID).

That is, it is desirable for the image management unit (100) to store and manage each input image data and unique identification information by matching them.

It is preferable that the above first analysis unit (200) be configured to include one pre-stored AI image analysis model.

It is preferable that the first analysis unit (200) analyzes the characteristics of each image data by the image management unit (100) using the AI image analysis model. In addition, the first analysis unit (200) receives specific image data to be analyzed from the outside and analyzes the characteristics thereof.

In detail, the first analysis unit (200) is preferably configured to include a basic analysis unit (210), a basic storage unit (220), and a basic specific analysis unit (230), as illustrated in FIG. 2.

The above basic analysis unit (210) uses the AI image analysis model to receive a plurality of image data from the image management unit (100) and analyzes the characteristics of each image data.

The above basic storage unit (220) stores and manages the analysis results by the above basic analysis unit (210), but it is preferable to store and manage the analysis results by matching them with the unique identification information assigned to each image data by the image management unit (100).

That is, the basic storage unit (220) stores and manages the characteristic information for each image data by the basic analysis unit (210) by matching it with the unique identification information provided by the image management unit (100).

The above basic specific analysis unit (230) uses the AI image analysis model to analyze the characteristics of the input specific image data.

The technology of inputting image data into an AI image analysis model and outputting characteristics through the above-mentioned basic analysis unit (210) and the above-mentioned basic specific analysis unit (230) is a widely used conventional technology, and a detailed description thereof is omitted.

In addition, the basic analysis unit (210) and the basic specific analysis unit (230) analyze the characteristics of image data using the same AI image analysis model, so that the image data being managed and the specific image data to be analyzed undergo characteristic analysis through the same analysis engine.

It is preferable that the first detection unit (300) use the analysis results of the first analysis unit (200) to compare and detect the characteristic data of each image data with the characteristic data of the specific image data.

In detail, it is preferable that the first detection unit (300) compares the characteristic data of each image data by the basic storage unit (220) with the characteristic data of the specific image data by the basic specific analysis unit (230) by applying a pre-stored comparison technique.

That is, it is preferable that the first detection unit (300) compares each characteristic data extracted from the plurality of image data by the image management unit (100) with the characteristic data of specific image data to be analyzed to calculate the degree of agreement (similarity). At this time, the first detection unit (300) applies the CUBE Extension technique as an example of the comparison technique, but this is only an example and is not limited thereto.

Through this, it is preferable that the first detection unit (300) detects image data having a degree of consistency with the characteristic data of the specific image data among the plurality of image data at a predetermined standard by using the results of the consistency analysis calculated as the number of the plurality of image data. In addition, it is preferable that the detection result is determined as the corresponding unique identification information together with the degree of consistency.

Here, the predetermined standard of matching degree for deriving some image data from among the plurality of image data by the image management unit (100) may be used to detect a preset number of images by sorting the matching degree in descending order, or a threshold value may be set to detect all image data having a matching degree higher than the threshold value.

This may vary depending on the settings of external users (administrators, etc.), and is not limited thereto.

However, the first detection unit (300) detects some image data among the plurality of image data by the image management unit (100) according to set criteria, and at this time, the detection criteria uses the results of the consistency through comparative analysis between characteristic data.

The above first detection unit (300) is similar to the operation of a conventional specific image detection module. However, the image detection system with improved detection performance according to one embodiment of the present invention performs the following operations in order to detect images with higher accuracy by applying an additional AI image analysis model without improving computing performance or increasing analysis time/analysis volume.

It is preferable that the image derivation unit (400) derives at least one image data among a plurality of image data by the image management unit (100) based on the detection result of the first detection unit (300).

In detail, it is preferable that the image derivation unit (400) derives image data in which the degree of agreement between specific image data to be analyzed and characteristic data among a plurality of image data by the image management unit (100) meets a predetermined standard.

At this time, it is clear that the number of image data derived is one or more, but is a very small number of data compared to the large number of image data input in advance.

It is preferable that the above second analysis unit (500) be configured to include one or more other AI image analysis models that are stored in advance.

At this time, it is preferable that the first analysis unit (200) and the second analysis unit (500) are each configured to include different AI image analysis models.

For example, AI image analysis models include the VGG model, the Facenet model, the Arcface model, the Deepface model, etc., and it is preferable that the first analysis unit (200) is configured to include any one selected from the AI image analysis models described above, and it is preferable that the second analysis unit (500) is configured to include at least one of the remaining models excluding the AI image analysis model composed by the first analysis unit (200) among the AI image analysis models described above.

Through this, it is preferable that the second analysis unit (500) analyzes the characteristic data of the image data derived by the image derivation unit (400) and the characteristic data of the specific image data using the AI image analysis model.

Specifically, the second analysis unit (500) is preferably configured to include a detailed analysis unit (510), a detailed storage unit (520), and a detailed specific analysis unit (530), as illustrated in FIG. 2.

When the second analysis unit (500) includes an AI image analysis model that is different from the AI image analysis model included in the first analysis unit (200), the detailed analysis unit (510) uses the AI image analysis model to receive the derived image data from the image derivation unit (400) and analyze the characteristics of each derived image data.

The above detailed storage unit (520) stores and manages the analysis results by the above detailed analysis unit (510), but it is preferable to store and manage the analysis results by matching them with the unique identification information assigned to each image data by the image management unit (100).

Through this, the detailed storage unit (520) stores and manages the characteristic information for each derived image data by the detailed analysis unit (510) by matching it with the unique identification information provided by the image management unit (100).

The above detailed specific analysis unit (530) uses the AI image analysis model to analyze the characteristics of the input specific image data.

The technology of inputting image data into an AI image analysis model and outputting characteristics through the detailed analysis unit (510) and detailed specific analysis unit (530) is a widely used conventional technology, and a detailed description thereof is omitted.

In addition, the detailed analysis unit (510) and the detailed specific analysis unit (530) analyze the characteristics of image data using the same AI image analysis model, so that the image data being managed and the specific image data to be analyzed undergo characteristic analysis through the same analysis engine.

It is preferable that the second detection unit (600) use the analysis results of the second analysis unit (500) to compare and detect the characteristic data of the derived image data with the characteristic data of the specific image data.

In detail, it is preferable that the second detection unit (600) compares the characteristic data of each derived image data by the detailed storage unit (520) with the characteristic data of the specific image data by the detailed specific analysis unit (530) by applying a pre-stored comparison technique.

That is, it is preferable that the second detection unit (600) compares the characteristic data extracted from at least one derived image data by the image derivation unit (400) with the characteristic data of specific image data to be analyzed to calculate the degree of agreement (similarity). At this time, the second detection unit (600) applies the Cosine or Euclidian method as an example of the comparison technique, but this is only an example and is not limited thereto.

It is preferable that the second detection unit (600) determines the unique identification number of at least one derived image data by the image derivation unit (400), matches the determined unique identification number with the calculated matching result, and generates final detection result information.

As another example, when the second analysis unit (500) includes two or more AI image analysis models that are different from the AI image analysis models included in the first analysis unit (200), the detailed analysis unit (510) uses each of the two or more AI image analysis models to receive the derived image data by the image derivation unit (400) and analyze the characteristics of each derived image data.

The above detailed storage unit (520) stores and manages the analysis results by the above detailed analysis unit (510) for each AI image analysis model, and it is preferable to store and manage the analysis results by matching them with the unique identification information assigned to each image data by the image management unit (100).

Through this, the detailed storage unit (520) stores and manages, for each AI image analysis model, the characteristic information for each derived image data by the detailed analysis unit (510) by matching it with the unique identification information provided by the image management unit (100).

In addition, the detailed specific analysis unit (530) analyzes the characteristics of the input specific image data using each of the two or more AI image analysis models.

Accordingly, it is preferable that the second detection unit (600) use the analysis results of the second analysis unit (500) for each AI image analysis model.

That is, it is preferable that the second detection unit (600) apply a pre-stored comparison technique to compare the characteristic data of each derived image data by the detailed storage unit (520) with the characteristic data of the specific image data by the detailed specific analysis unit (530) for each AI image analysis model.

Through this, it is preferable that the second detection unit (600) compares the characteristic data extracted from at least one derived image data by the image derivation unit (400) for each AI image analysis model with the characteristic data of specific image data to be analyzed to calculate the degree of agreement (similarity). At this time, the second detection unit (600) applies the Cosine or Euclidian method as an example of the comparison technique, but this is only an example and is not limited thereto.

It is preferable that the second detection unit (600) determines a unique identification number of at least one derived image data by the image derivation unit (400), and, together with the determined unique identification number, calculates an average value of the degree of agreement calculated for each AI image analysis model, and the result of the highest or lowest degree of agreement among the degrees of agreement calculated for each AI image analysis model, and matches them to generate final detection result information.

That is, even if the image data is the same, since the criteria for analyzing the characteristic data are different for each AI image analysis model, the results of the consistency of the analyzed characteristic data may also be different. Considering this, the consistency results used to generate the final detection result information can be generated by matching the average value of the consistency calculated for each of two or more AI image analysis models, the highest consistency result, or the lowest consistency result.

Here, as an example, regarding the image detection system with improved detection performance according to one embodiment of the present invention, in the first step, when 100 image data are input in advance through the image management unit (100), 100 characteristic data are derived using one AI image analysis model, and 1 characteristic data corresponding to specific image data to be analyzed is derived, and the top 5 images with high degree of agreement are derived through each consistency comparison (at least 100 comparisons are performed). Thereafter, in the second step, 5 characteristic data for images with high degree of agreement derived from at least one AI image analysis model different from the AI image analysis model are derived, and 1 characteristic data is derived again for specific image data to be analyzed, and then the consistency for the images with high degree of agreement derived from the previous AI image analysis model is verified through each consistency comparison (at least 5 comparisons are performed).

At this time, when applying two or more AI image analysis models applied in step 2, it is desirable to perform characteristic derivation and consistency comparison for each AI image analysis model and reconfirm the consistency by making maximum use of the different analysis engines of each AI image analysis model.

Through this, there is an advantage in that image detection can be performed with higher accuracy without improving computing performance or increasing analysis time/amount by sequentially applying multiple AI image analysis models.

FIG. 3 is a flowchart illustrating an image detection method with improved detection performance according to one embodiment of the present invention. Referring to FIG. 3, an image detection method with improved detection performance according to one embodiment of the present invention will be described in detail.

An image detection method with improved detection performance according to one embodiment of the present invention is preferably configured to include a first input step (S100), a second input step (S200), a first analysis step (S300), a first comparison step (S400), a derivation step (S500), a second analysis step (S600), and a second comparison step (S700), as illustrated in FIG. 3. In addition, each step is performed by an image detection system with improved detection performance implemented by a computer.

Let's take a closer look at each step:

In the above first input step (S100), the image management unit (100) receives a plurality of different image data from the outside in advance, assigns unique identification information (such as ID) to each image data, and stores and manages the data.

The above image data is data extracted and input through video information or image information, etc., to detect a facial image, and the first input step (S100) matches each of the input image data with unique identification information and stores and manages them.

In the above second input step (S200), specific image data to be analyzed is input from the outside into the first analysis unit (200).

The first analysis step (S300) analyzes the characteristics of each image data by the first input step (S100) using one of the AI image analysis models stored in advance in the first analysis unit (200). In addition, the first analysis step (S300) receives the specific image data by the second input step (S200) using the AI image analysis model and analyzes the characteristics thereof.

In detail, the first analysis step (S300) analyzes the characteristics of each image data by the first input step (S100), stores and manages the analysis results, and stores and manages the analysis results by matching them with the unique identification information assigned to each image data.

Through the first analysis step (S300), the image data from the first input step (S100) and the specific image data from the second input step (S200) are analyzed using the same AI image analysis model to analyze the characteristics of the image data, so that the image data being managed and the specific image data to be analyzed are analyzed through the same analysis engine.

In the above first comparison step (S400), the first detection unit (300) uses the analysis results of the first analysis step (S300) to compare and detect the characteristic data of each image data with the characteristic data of the specific image data.

In detail, the first comparison step (S400) applies a pre-stored comparison technique to compare the characteristic data of each image data with the characteristic data of the specific image data, and calculates the degree of agreement (similarity). At this time, in the first comparison step (S400), the CUBE Extension technique is applied as an example of the comparison technique, but this is only an example and is not limited thereto.

In addition, the first comparison step (S400) detects image data having a degree of consistency with characteristic data of the specific image data among the plurality of image data according to a predetermined standard by using the calculated consistency analysis result.

A predetermined standard of matching degree for deriving some image data from among the above plurality of image data may be used to detect a preset number of images by sorting the matching degree in descending order, or a threshold value may be set to detect all image data having a matching degree higher than the threshold value.

This may vary depending on the settings of external users (administrators, etc.), and is not limited thereto.

The above derivation step (S500) derives at least one image data from among the plurality of image data from the first input step (S100) based on the detection result from the first comparison step (S400) in the image derivation unit (400).

That is, among the plurality of image data by the first input step (S100), image data is derived in which the degree of agreement between the specific image data to be analyzed and the characteristic data meets a predetermined standard. At this time, it is clear that the derived image data is at least one in number and is a very small number of data compared to the plurality of image data input in advance.

The second analysis step (S600) uses at least one other AI image analysis model stored in advance in the second analysis unit (500) to analyze the characteristic data of the image data derived by the derivation step (S500) and the characteristic data of the specific image data.

At this time, AI image analysis models include VGG model, Facenet model, Arcface model, Deepface model, etc., and in the first analysis step (S300), it is preferable to configure including one selected from among the AI image analysis models described above, and in the second analysis step (S600), it is configured including at least one of the remaining models excluding the AI image analysis model configured by the first analysis step (S300) among the AI image analysis models described above.

Additionally, the second analysis step (S600) may include one AI image analysis model different from the AI image analysis model included in the first analysis step (S300), or may include two or more AI image analysis models.

For example, when an AI image analysis model different from the AI image analysis model included in the first analysis step (S300) is included, the second analysis step (S600) uses the AI image analysis model to receive the derived image data by the derivation step (S500) and analyze the characteristics of each derived image data.

In addition, the specific image data is input through the second input step (S200) and its characteristics are analyzed.

In detail, the second analysis step (S600) analyzes the characteristics of the image data derived by the derivation step (S500), and stores and manages the analysis results, while matching the analysis results with the unique identification information assigned to each derived image data and storing and managing them. Through this, the image data derived by the derivation step (S500) and the specific image data by the second input step (S200) analyze the characteristics of the image data using the same AI image analysis model, so that the image data being managed and the specific image data to be analyzed undergo characteristic analysis through the same analysis engine.

In the second comparison step (S700), the second detection unit (600) uses the analysis results of the second analysis step (S600) to compare and detect the characteristic data of the derived image data with the characteristic data of the specific image data.

In detail, the second comparison step (S700) calculates the degree of agreement (similarity) by comparing the characteristic data of each derived image data with the characteristic data of the specific image data by applying a pre-stored comparison technique. At this time, the second comparison step (S700) applies the Cosine or Euclidian method as an example of the comparison technique, but this is only an example and is not limited thereto.

In addition, it is preferable that the image detection method with improved detection performance according to one embodiment of the present invention further includes a final result step (S800) in order to transmit the final detection result by the second comparison step (S700) to the outside, as illustrated in FIG. 3.

The final result step (S800) above determines the unique identification number of the image data derived by the derivation step (S500) when one AI image analysis model is used in the second analysis step (S600), and matches the determined unique identification number with the calculated match result to generate the final detection result information.

As another example, when two or more AI image analysis models different from the AI image analysis models included in the first analysis step (S300) are included, the second analysis step (S600) uses each of the AI image analysis models to receive the derived image data by the derivation step (S500) and analyze the characteristics of each derived image data.

In addition, the specific image data is input by the second input step (S200) and its characteristics are analyzed. In this case, the characteristics of the specific image data input are analyzed using each of the two or more AI image analysis models.

In detail, the second analysis step (S600) analyzes the characteristics of the image data derived by the derivation step (S500), and stores and manages the analysis results for each AI image analysis model. The analysis results are stored and managed by matching them with the unique identification information assigned to each derived image data.

The second comparison step (S700) also uses the analysis results of the second analysis step (S600) for each AI image analysis model to compare and detect the characteristic data of the derived image data and the characteristic data of the specific image data.

That is, for each AI image analysis model, the characteristic data of each derived image data and the characteristic data of the specific image data are compared to calculate the degree of agreement (similarity).

At this time, in the final result step (S800), even if the image data is the same, since the criteria for analyzing the characteristic data are different for each AI image analysis model, the consistency results of the analyzed characteristic data may also be different. Considering this, the consistency results used to generate the final detection result information may be generated by matching the average value of the consistency calculated for each of two or more AI image analysis models, the highest consistency result, or the lowest consistency result.

In detail, the final result step (S800) uses the degree of agreement between the characteristic data of the derived image data calculated for each AI image analysis model and the characteristic data of the specific image data to calculate the average of the degrees of agreement calculated for each AI image analysis model, the highest degree of agreement or the lowest degree of agreement among the degrees of agreement calculated for each AI image analysis model, and matches them with a corresponding unique identification number to generate the final detection result information.

Through the image detection method with improved detection performance according to one embodiment of the present invention, there is an advantage in that image detection can be performed with higher accuracy without improving computing performance or increasing analysis time/analysis volume by sequentially applying a plurality of AI image analysis models.

As described above, the present invention has been described by specific matters such as specific components and limited example drawings, but this has only been provided to help a more general understanding of the present invention, and the present invention is not limited to the above-described embodiment, and those skilled in the art to which the present invention pertains can make various modifications and variations from this description.

Therefore, the idea of the present invention should not be limited to the described embodiments, and all things that are equivalent or equivalent to the scope of the patent claims as well as the scope of the patent claims are included in the scope of the idea of the present invention.

100 : Image Management Department
200: 1st Analysis Department
210: Basic Analysis Section 220: Basic Storage Section
230: Basic Specific Analysis Section
300: 1st Detection Unit
400 : Image derivation section
500: 2nd Analysis Department
510: Detailed Analysis Section 520: Detailed Storage Section
530: Detailed Specific Analysis Department
600: 2nd Detection Unit

Claims (9)

An image management unit (100) that receives a large amount of image data from an external source in advance, stores and manages each image data by assigning unique identification information to each image data;
A first analysis unit (200) that includes one AI image analysis model that is stored in advance and analyzes the characteristics of each image data by the image management unit (100) and specific image data to be analyzed that is input from the outside;
A first detection unit (300) that uses the analysis results of the first analysis unit (200) to compare and detect the characteristic data of each image data with the characteristic data of the specific image data, and analyzes the degree of consistency between the characteristic data of the specific image data and each image data;
An image derivation unit (400) that derives at least one image data having a degree of consistency that satisfies a predetermined standard and the specific image data among a plurality of image data by the image management unit (100) based on the detection result of the first detection unit (300);
A second analysis unit (500) that includes at least one other AI image analysis model that is stored in advance, and analyzes the image data derived by the image derivation unit (400) and the characteristics of the specific image data; and
A second detection unit (600) that uses the analysis results by each AI image analysis model of the second analysis unit (500) to compare and detect the characteristic data of the derived image data and the characteristic data of the specific image data, and calculates the degree of consistency between the specific image data and the derived image data;
Including,
The above second detection unit (600)
By determining the unique identification number of the above-mentioned derived image data, and matching the result of the calculated match with the determined unique identification number of the above-mentioned derived image data, the final detection result information for the above-mentioned specific image data is generated.
When two or more matches are calculated, the average of the two or more matches calculated, or the highest match or the lowest match result among the two or more matches calculated, is set as the match result to generate the final detection result information.
The above first analysis unit (200) and the above second analysis unit (500)
An image detection system with improved detection performance, each comprising different AI image analysis models.
In paragraph 1,
The above first analysis unit (200)
A basic analysis unit (210) that receives a plurality of image data from the image management unit (100) using the above AI image analysis model and analyzes the characteristics of each image data;
A basic storage unit (220) that matches the analysis results by the basic analysis unit (210) with the unique identification information assigned to each image data and stores and manages them; and
A basic specific analysis unit (230) that analyzes the characteristics of the specific image data using the above AI image analysis model;
An image detection system having improved detection performance, comprising:
In the second paragraph,
The above first detection unit (300)
An image detection system with improved detection performance, which applies a pre-stored comparison technique to compare the characteristic data of each image data by the basic storage unit (220) with the characteristic data of the specific image data by the basic specific analysis unit (230), detects image data having a degree of consistency with the characteristic data of the specific image data according to a predetermined standard, and determines unique identification information corresponding to the detected image data.
In the third paragraph,
The above second analysis unit (500)
A detailed analysis unit (510) that receives the derived image data by the image derivation unit (400) using one or more other AI image analysis models and analyzes the characteristics of each derived image data for each AI image analysis model;
A detailed storage unit (520) that matches the analysis results by the detailed analysis unit (510) with the unique identification information assigned to each derived image data and stores and manages them for each AI image analysis model; and
A detailed specific analysis unit (530) that analyzes the characteristics of the specific image data for each AI image analysis model using one or more other AI image analysis models;
An image detection system having improved detection performance, comprising:
In paragraph 4,
The above second detection unit (600)
By applying the pre-stored comparison technique, the characteristic data of each derived image data by the detailed storage unit (520) and the characteristic data of the specific image data by the detailed specific analysis unit (530) are compared for each AI image analysis model, and the consistency between the specific image data and the derived image data is analyzed.
An image detection system with improved detection performance, which determines a unique identification number for each specific image data, and generates final detection result information by matching the unique identification number determined for each specific image data with the analyzed degree of agreement.
In a method for image detection with improved detection performance, each step is performed by an image detection system with improved detection performance implemented by a computer,
In the image management section, a first input step (S100) of inputting a plurality of image data in advance, assigning unique identification information to each image data, and storing and managing the same;
In the first analysis section, a second input step (S200) for receiving specific image data to be analyzed from the outside;
In the first analysis unit, a first analysis step (S300) for analyzing the characteristics of each image data by the first input step (S100) and specific image data by the second input step (S200) using any one of the pre-stored AI image analysis models;
In the first detection unit, a first comparison step (S400) for comparing and detecting the degree of correspondence between each image data and specific image data using the analysis result of the first analysis step (S300);
In the image derivation unit, a derivation step (S500) of deriving at least one image data having a matching degree that satisfies a preset standard among a plurality of image data by the first input step (S100) based on the comparison detection result by the first comparison step (S400);
In the second analysis unit, a second analysis step (S600) for analyzing the characteristics of the image data derived by the derivation step (S500) and the specific image data derived by the second input step (S200) for each AI image analysis model using one or more other AI image analysis models that are stored in advance;
In the second detection unit, a second comparison step (S700) for comparing and detecting the degree of consistency between the derived image data and specific image data using each analysis result of the second analysis step (S600); and
In the second detection unit, a final result step (S800) of determining a unique identification number of each derived image data, matching the result of the match between the unique identification information determined for each derived image data and the specific image data detected by comparison through the second comparison step (S700) to generate final detection result information;
Including,
The above final result step (S800)
In the second analysis step (S600) above, when two or more AI image analysis models are used and two or more matches are detected through comparison in the second comparison step (S700), the average value of the two or more matches, the highest match and the lowest match result among the two or more matches are set as the match result to generate the final detection result information.
An image detection method with improved detection performance, characterized in that the AI image analysis model by the first analysis step (S300) and the AI image analysis model by the second analysis step (S600) are different AI image analysis models.
delete In paragraph 6,
The above second analysis step (S600)
An image detection method with improved detection performance, which uses one or more AI image analysis models to analyze the characteristics of the image data derived by the derivation step (S500) and the specific image data derived by the second input step (S200) for each AI image analysis model.
delete

Images