KR102628456B1 - Customer fashion analysis system using artificial intelligence - Google Patents

Abstract

The present invention relates to a customer fashion analysis system using artificial intelligence. The customer fashion analysis system using artificial intelligence includes a customer fashion information collection unit that collects the customer's fashion information by photographing the customer's fashion; and an AI fashion sensibility analysis unit that receives the customer's fashion video data collected through the customer fashion information collection unit, classifies it into fashion sensibility image clusters according to the customer's pre-learned fashion sensibility, and outputs them.
According to the present invention, the time to classify consumers' emotions can be greatly shortened and costs can be reduced by classifying customer fashion using artificial intelligence. In addition, by quickly identifying consumer sensibility, you can identify changing consumer sensibilities and set targets for fashion products you want to plan and develop. This prevents fashion products that do not properly reflect consumer sensibilities from being stocked up in stores and prevents the waste of raw materials used in fashion products.

Description

Customer fashion analysis system using artificial intelligence {Customer fashion analysis system using artificial intelligence}

The present invention relates to fashion analysis and diagnosis, and in particular to diagnose brand diagnosis, target setting, trend prediction, concept setting, product composition, material selection, brand image, product line, in-store display, directing technology, and sales promotion improvement. This is about a customer fashion analysis system using artificial intelligence that allows you to set targets for products by classifying consumers' emotions into clusters through artificial intelligence.

Fashion strategy analysis is a specialized field in which experts with long experience in the fashion field visit sites offline for long periods of time to provide consulting. When analyzing fashion strategies, consumer emotion is an important factor. For example, you can understand the emotions of consumers by looking at their clothes when they come to the store. However, planning fashion products and analyzing fashion strategies using consumers' emotions is not standardized and relies on the experience of experts.

Specifically, in order to understand the emotions of consumers, we photograph the fashion of customers who visit the store and analyze each photo based on the experience of experts. For example, fashion experts were dispatched to customer stores to collect data, consult with store managers, and analyze customers' fashion to analyze the store's fashion strategy.

Classifying consumers' emotions in this way has the problem of taking a long time and increasing costs. Failure to classify consumer sensibilities in a timely manner can cause fashion products that do not properly reflect consumer sensibilities to accumulate in stores, and can also lead to waste of raw materials.

In addition, recently, due to the coronavirus pandemic, it has become impossible to visit the site and analyze fashion strategies. Accordingly, a system that analyzes fashion on a non-contact basis and experts who diagnose fashion strategies based on fashion analysis data It is necessary to respond to the non-contact era by developing a system.

The problem to be solved by the present invention was created to solve the above-mentioned problems, including brand diagnosis, target setting, trend prediction, concept setting, product composition, material selection, brand image, product line, in-store display, directing technology, In order to improve and diagnose sales promotion, etc., the image data generated by photographing the fashion of customers visiting the store is classified into consumer emotion-based clusters through artificial intelligence, so that targets can be set for fashion products to be planned and developed. The goal is to provide a customer fashion analysis system using artificial intelligence.

A customer fashion analysis system using artificial intelligence according to an example of the present invention to achieve the above technical problem includes a customer fashion information collection unit that collects the customer's fashion information by photographing the customer's fashion; and an AI fashion sensibility analysis unit that receives the customer's fashion video data collected through the customer fashion information collection unit, classifies it into fashion sensibility image clusters according to the customer's pre-learned fashion sensibility, and outputs them.

The customer fashion information collection unit includes a CCTV installed in the store to photograph the customer's fashion. The video captured by the CCTV can be transmitted to the AI fashion sensibility analysis unit through a network, and a large amount of fashion images can be converted to fashion sensibility. Clustering modeling is performed through supervised machine learning using the labeled customer's fashion images as learning data based on the fashion sensibility basic diagram consisting of multiple fashion sensibility image axes to classify according to the input. By analyzing fashion images, you can classify which cluster the input fashion image belongs to based on the learned model. The plurality of fashion emotional image axes are Classic - Modern fashion emotional image axis, Romantic - Masculine fashion emotional image axis, Elegance - Active fashion emotional image axis, Sophisticated - Can include a country sensibility image axis.

In addition, the AI fashion emotion analysis unit is composed of a CNN (Convolutional Netural Network) structure for supervised machine learning, and when receiving an input fashion image, a convolution is performed to find features from the input fashion image, and the convolution is used to find features. Through a pooling process that downsamples data and a flattening process that converts data in a data matrix format into a one-dimensional array, it is connected to a fully connected layer and learned to form a CNN, and the coneball Convolution divides the input image into three channels according to RGB (Red, Green, Blue) colors, finds features in each channel, goes through a filtering process to classify them by feature, and then calculates the values of the three channels. In addition, one feature map can be constructed.

The AI fashion sensibility analysis unit receives the customer's fashion video data collected through the customer fashion information collection unit, anonymizes the customer's face information, and analyzes the store customer's fashion to determine top fashion, bottom fashion, shoe fashion, bag fashion, etc. AI customer fashion image analysis unit that classifies customer fashion images by type; and supervised machine learning using customer fashion images labeled based on a basic fashion sensibility diagram consisting of a plurality of fashion sensibility image axes to classify a large amount of fashion images according to fashion sensibility as learning data. After clustering modeling, it may include an AI fashion sensibility classification unit that analyzes the input fashion image and classifies which cluster the input fashion image belongs to based on the learned model.

The customer fashion analysis system using artificial intelligence according to an embodiment of the present invention analyzes the fashion sensibility of customers visiting the store using the fashion sensibility cluster information output by the AI fashion sensibility analysis unit, and analyzes the sensibility of the products on display. It can further include AI MD that identifies the needs of customers by calculating the degree to which visiting customers' fashion sensibilities match the products sold in the store through comparative analysis.

According to the customer fashion analysis system using artificial intelligence according to the present invention, the time to classify the customer's emotion can be greatly shortened and the cost can be reduced by classifying the customer's fashion using artificial intelligence.

And according to the present invention, by quickly identifying consumer sensibilities, changes in consumers can be easily identified and targets for fashion products to be planned and developed can be efficiently set. This prevents fashion products that do not properly reflect consumer sensibilities from being stocked up in stores and prevents the waste of raw materials used in fashion products.

Additionally, according to the present invention, it is possible to remotely diagnose fashion strategies in a non-face-to-face situation by providing a system in which users who run a fashion store can be interviewed remotely on a principle similar to how a doctor diagnoses a patient.

In addition, the customer fashion analysis system using artificial intelligence according to the present invention is a type of expert system and inputs the knowledge of fashion strategy experts, for example, by inputting 20 years of data to build a knowledge base system, Users can receive a diagnosis of the store's fashion strategy through a question-and-answer format similar to a patient interviewing a doctor, and can receive information from remote stores in real time and use it for analysis.

In addition, the customer fashion analysis system using artificial intelligence according to the present invention enables more scientific and accurate customer fashion analysis than before by utilizing CCTV equipped with artificial intelligence, and enables comprehensive consulting using the web, creating a new concept commonly used in the Untact era. It can be a business model. By developing AI using 20 years of data, it is possible to expand the number of stores that can be covered per basic expert. In addition, analysis by AI can reduce analysis bias based on expert tendencies, improving the quality of fashion analysis, and reducing technology implementation risks by utilizing already proven AI technology.

Figure 1 is a block diagram showing the configuration of a customer fashion analysis system using artificial intelligence according to an embodiment of the present invention.
Figure 2 shows in more detail the configuration of a customer fashion analysis system using artificial intelligence according to an embodiment of the present invention.
Figure 3 shows in more detail the AI fashion sensibility classification unit of the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention.
Figure 4 is a block diagram showing the configuration of artificial intelligence for AI fashion emotion analysis and cluster classification in the AI fashion emotion classification unit of the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention.
Figure 5 shows an example of convolution operation on an MxNx3 image matrix with a 3x3x3 kernel used in the AI fashion sensibility classification unit of the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention.
Figure 6 shows a basic fashion sensibility diagram that represents fashion images into 8 emotional elements with 4 image axes to cluster consumers' sensibilities.
FIGS. 7A to 7H show an example of each of the eight emotional elements when a fashion image is represented by eight emotional elements along four image axes.
Figure 8 shows the basic fashion sensibility diagram and the fashion sensibility diagnosis results as a fashion sensibility diagram.
Figure 9 shows clustering that combines the customer's fashion image with classic sensibility, elegant sensibility, romantic sensibility, casual sensibility, modern sensibility, etc. and age group to analyze the customer's fashion taste in the customer fashion analysis system using artificial intelligence according to the present invention. This shows an example.
Figure 10a shows an emotional image diagram for the modern fashion-oriented (Colorful casual) cluster, and Figure 10b shows an emotional image diagram for the Classic cluster.
Figure 11 shows an example of the results of photographing and analyzing a customer's fashion in the customer fashion analysis system using artificial intelligence according to the present invention and storing the resulting data as a directory and file.
Figure 12 shows an example of clustering the customer's fashion sensibility into a practical group that emphasizes classical beauty, a fashion group that emphasizes feminine beauty, an artistic fashion group, and an innovative practical group in order to analyze the customer's fashion sensibility in the customer fashion analysis system using artificial intelligence according to the present invention. will be.
Figure 13 shows the development of the AI Fashion Lachivium System Integration demonstration project to which the customer fashion analysis system using artificial intelligence according to the present invention is applied.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. The embodiments described in this specification and the configurations shown in the drawings are only preferred embodiments of the present invention and do not represent the entire technical idea of the present invention, so various equivalents can be substituted for them at the time of filing the present application. It should be understood that there may be variations.

Fashion brand product planning is accomplished through the following steps: brand diagnosis, target setting, trend prediction, concept setting, product composition, material selection, design, VMD, and fashion advisory.

The brand diagnosis stage includes the process of researching and collecting trends such as popular items, selecting and classifying data suitable for brand planning, analyzing customer style analysis, and checking clusters. In addition, in the brand diagnosis stage, the brand is diagnosed through look analysis such as silhouette, material, color, design details, functional sewing, image, product line, store display, directing technology, and sales promotion plan. To this end, the brand is comprehensively diagnosed by collecting fashion data and classifying consumer clusters.

The target setting step is to set a target by classifying consumer clusters using a CNN algorithm of consumer emotion classification. The style of customers entering the store is captured and the fashion style is analyzed. Trend prediction involves diagnosing fashion sensibility and predicting trends based on 8 axes of fashion sensibility using AI image analysis and RNN-based time series prediction algorithms. The concept setting stage includes setting main and sub-season theme concepts through AI image analysis and using AI-analyzed image data.

The product composition stage includes item design, concept theme composition, and the process of predicting trends that meet the needs of consumer clusters. The material selection stage includes the process of selecting materials by fashion sensibility using classification & regression algorithms, stylizing them, checking them with AI, producing samples, wearing them, and then checking them again with AI. The design stage includes the process of analyzing and checking designs by fashion sensibility using classification & regression algorithms, analyzing them with AI, and then re-checking them with AI after design. The VMD stage is an integrated management system that includes customer management using facial recognition technology. The fashion advisory stage includes product recommendation through customer/store visit records, white/black list management including expected age, gender, visit and purchase records, and customer management and fashion coordination advice using facial recognition technology. do.

The customer fashion analysis system using artificial intelligence according to the present invention covers the entire fashion industry, including brand diagnosis, target setting, trend prediction, concept setting, product composition, material selection, brand image, product line, in-store display, directing technology, and sales promotion. It is an artificial intelligence system that classifies consumers' emotions into clusters through artificial intelligence and sets targets for products in order to diagnose the work process.

The fashion analysis system using artificial intelligence according to the present invention is a system that transmits videos of customers entering and leaving the store to a server in real time to analyze the customer's emotions. It utilizes artificial intelligence to determine the location of the customer in the store from the customer's video, By analyzing tops/bottoms, shoes, bags, etc., you can analyze customer emotions by dividing them into multiple categories, for example, 15 types of categories. It also provides a function to prevent personal information leakage by blocking and anonymizing the customer's face in the video.

The customer fashion analysis system using artificial intelligence according to the present invention can be installed in stores not only domestically but also overseas, and can operate artificial intelligence in real time on site to disperse complex data processing processes and prevent leakage of sensitive personal information. can do.

The fashion analysis system using artificial intelligence according to the present invention has two main functions centered on two core technologies. The first is a system created by applying CNN (Convolution Neural Network) and YOLO (You Only Look Once) models to analyze fashion Target consumers and consumer clusters can be diagnosed by analyzing the fashion sensibilities and fashion lifestyles of customers entering the store. Second, by utilizing the expert system (expert system), which is actively applied in the application field of artificial intelligence technology, it is possible to provide a solution that presents a strategic development direction by diagnosing the work of each fashion company's product planning process. .

Figure 1 is a block diagram showing the configuration of a customer fashion analysis system using artificial intelligence according to an embodiment of the present invention. Referring to Figure 1, the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention includes a customer fashion information collection unit 110 and an AI fashion emotion analysis unit 120, and an AI MD 130. More may be included.

The customer fashion information collection unit 110 collects the customer's fashion information by photographing the customer's fashion. The customer fashion information collection unit 110 is installed in the store and includes a CCTV that photographs the customer's fashion, and the video captured by the CCTV is transmitted to the AI fashion sensibility analysis unit 120 through a network.

The AI fashion sensibility analysis unit 120 receives the customer's fashion video data collected through the customer fashion information collection unit 110, classifies it into fashion sensibility image clusters according to the customer's pre-learned fashion sensibility, and outputs it. The AI fashion sensibility analysis unit 120 uses the labeled customer's fashion images as learning data based on a basic fashion sensibility diagram consisting of a plurality of fashion sensibility image axes to classify a large amount of fashion images according to fashion sensibility, and creates a mapped model. After clustering modeling through machine learning (supervised learning), the input fashion images are analyzed to classify which cluster the input fashion images belong to based on the learned model.

The plurality of fashion emotional image axes are Classic - Modern fashion emotional image axis, Romantic - Masculine fashion emotional image axis, Elegance - Active fashion emotional image axis, Sophisticated - Includes a country sensibility image axis.

In addition, the AI fashion emotion analysis unit 120 is composed of a CNN (Convolutional Netural Network) structure for supervised machine learning, and when receiving an input fashion image, a convolution (convolution) to find features from the input fashion image, the conball Through the pooling process, which downsamples the data that has gone through the solution, and the flattening process, which converts data in a data matrix format into a one-dimensional array, it is connected to a fully connected layer to form a CNN and learn. . At this time, the convolution divides the input image into three channels according to RGB (Red, Green, Blue) colors, finds features in each channel, and goes through a filtering process to classify them by feature. One feature map can be formed by adding three channel values.

Additionally, the AI fashion emotion analysis unit 120 includes an AI customer fashion image analysis unit 122 and an AI fashion emotion classification unit 124. The AI customer fashion image analysis unit 122 receives the customer fashion video data collected through the customer fashion information collection unit 110, anonymizes the customer's face information, and analyzes the store customer's fashion to determine top and bottom fashion. , shoe fashion, bag fashion, etc. are classified by customer fashion image. The AI fashion sensibility classification unit 124 uses the customer's fashion images labeled as learning data based on a basic fashion sensibility diagram consisting of a plurality of fashion sensibility image axes to classify a large amount of fashion images according to fashion sensibility, and creates a mapped model. After clustering modeling through machine learning (supervised learning), the input fashion images are analyzed to classify which cluster the input fashion images belong to based on the learned model.

Meanwhile, the AI MD 130 analyzes the fashion sensibility of customers who visit the store using the fashion sensibility cluster information output by the AI fashion sensibility analysis unit 120 and compares and analyzes it with the sensibility of the products on display to determine the customer's fashion sensibility. Identify customer needs by calculating the degree to which fashion sensibility matches the products sold in stores. In addition, it can receive expert knowledge strategy information, analyze consumer clusters based on consumer emotions using AI, diagnose the product planning process, and provide a solution that presents the direction of strategic development as a result of fashion strategy diagnosis.

Figure 2 shows the configuration of a customer fashion analysis system using artificial intelligence according to an embodiment of the present invention. In the customer fashion analysis system using artificial intelligence according to the present invention, the customer fashion information collection unit 110 is equipped with CCTV-type products as a photography means, and collects customer fashion information by filming the fashion of customers visiting the store through CCTV. Various artificial intelligence models are applied to the AI fashion emotion analysis unit (120), and a video analysis artificial intelligence algorithm is run in real time on general CCTV hardware (H/W).

And in the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention, the AI customer fashion image analysis unit 122 uses the YOLO (You Only Look Once) model to find people in the video, and once again Apply and use YOLO to separate various parts of fashion such as face, hat, top, bottom, shoes, and bag. In addition, the AI fashion emotional classification unit 124 uses a Faster CNN (Convolutional Neural Network) model to classify the fashion of hats, tops, bottoms, shoes, and bags, and uses a DNN (Deep Neural Network) to determine which combination of each. Classify whether it is a fashion category or not. In the embodiment of the present invention, it is classified into a total of 15 types, utilizes more than 30,000 pieces of data accumulated over 20 years, and uses CCTV equipped with artificial intelligence. The types of fashion classification are not limited to 15 types and are expandable.

In more detail, the present invention performs consumer clustering of a customer's fashion using the customer's fashion video as input data. Consumer clustering uses existing labeled data to perform clustering modeling through supervised learning using a library (e.g. Keras), and then analyzes newly given images to create a corresponding model based on the learned model. This is the process of analyzing which cluster an image belongs to. Here, the process of labeling the data may correspond to preprocessing, and modeling may correspond to the training process, and accuracy can be confirmed by testing after training.

Figure 3 shows in more detail the AI fashion sensibility classification unit 120 of the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention. Referring to FIG. 3, the customer fashion image collected by the customer fashion information collection unit 110 is received as an input image 310, and the input image 310 is input to the learned AI fashion sensibility classification unit 120. Then, the fashion sensibility is classified for each image through the learned artificial intelligence, and a fashion sensibility cluster 330 is output for each image.

Figure 4 is a block diagram showing the configuration of artificial intelligence for AI fashion emotion analysis and cluster classification in the AI fashion emotion classification unit 120 of the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention. Referring to Figure 4, Supervised Learning used in the customer fashion analysis system using artificial intelligence according to the present invention is a CNN (Convolutional Neural Network) structure, and when a photo is provided through a video, image data 410 of the photo is generated. A network is formed and learned through the processes of convolution, pooling, and flattening.

Convolution is the task of finding features from a given image. When the first image is given to the corresponding layer, it is divided into three channels according to RGB (Red, Green, Blue) color, and then features are found in each channel. Examples of the characteristics of each channel include fashion silhouette, fashion shape, and fashion material (hard, soft). After going through a filtering process to classify by feature, the three channel values are added to form one feature map. The Feature Map may be a result output each time it passes through each layer. In the case of Convolution, the first layer, there are RGB values included in each pixel, and these are divided into three channels to find patterns. In this process, to solve the problem that the size of the image becomes smaller and the information in the edge pixels disappears, zero-padding can be applied to give the border 0.

Referring to Figure 4, Batch Normalization (420, 430) does not learn (train) all at once when given 1,000 input images, but performs the calculation by grouping them into a certain number, for example, 32 images in batches. Training time can be shortened by normalizing the field.

Pooling plays the role of down sampling the data received from the convolutional layer. For example, if the size of the data received from the convolutional layer is 32x32x3, it is reduced to 16x16x3 to reduce the size of the data and emphasize specific data. do. The pooling method used in the customer fashion analysis system using artificial intelligence according to the present invention is max pooling, which takes the largest value from each matrix input value. Repeat the process of reducing the size of the feature map through max pooling.

Referring to FIG. 4, Dropout 440 uses Bernoulli distribution to randomly deactivate neurons during the training process to prevent biased learning. If it belongs to one category, there will be repeated patterns, so rather than focusing on a specific one, you can learn what is repeated frequently and shorten the learning (training) time.

In this way, the process of collecting features in the photo and reducing the size of the image through pooling is repeated several times to form a CNN structure.

Flattening and dense are layers that connect according to the number of classes at the last part of CNN construction. By converting data in matrix format into a primary array, they are connected to the last fully-connected layer and help form a CNN. Reference number 450 indicates output clustering for the input image 410.

As convolution and max pooling are repeated, they are learned to be classified into a preset number of categories (clusters). In the case of 2x2 pooling, Max pooling outputs the largest value among the four matrix values. For example, if the four values of the feature map that underwent convolution are 12, 20, 8, and 6, the three values are discarded and 200,000 are taken. Through this process, an artificial neural network is constructed, connected, and finally mapped and classified into a preset number of categories.

Figure 5 shows an example of a convolution operation on an MxNx3 image matrix with a 3x3x3 kernel used in the AI fashion sensibility classification unit 120 of the customer fashion analysis system using artificial intelligence according to an embodiment of the present invention. will be. Referring to Figure 5, red, green, and blue are divided and expressed as a matrix, then multiplied by 3x3 and added. Through this process, the outline of the image can be found. Likewise, in the case of the artificial intelligence model used in the present invention, YOLO is used for segmentation and object detection, and Faster CNN is used for classification. These models can be said to be proven AI models that have been verified for many years and are used in various fields.

Meanwhile, the fashion analysis system using artificial intelligence according to the present invention can classify and segment fashion sensibilities. For example, as shown in FIG. 5 described above, fashion sensibility can be segmented for convolution operation on the MxNx3 image matrix using a 3x3x3 kernel. In other words, the fashion sensibility of customers visiting the store is diagnosed by analyzing videos or input photos of customers visiting the store and deriving a fashion sensibility diagram diagnosis result based on the basic fashion sensibility diagram. Figure 8 shows a basic fashion sensibility diagram (a) for fashion sensibility diagnosis and a fashion sensibility diagram (b) showing the fashion sensibility diagnosis results.

Labeled data, which is learning data used in a customer fashion analysis system using artificial intelligence according to an embodiment of the present invention, will be described in more detail.

Figure 6 shows a basic fashion sensibility diagram that represents fashion images into 8 emotional elements with 4 image axes to cluster consumers' sensibilities. The four fashion emotional image axes are Classic - Modern, Romantic - Masculine, Elegance - Active, and Sophisticated - Country. I can hear it. The fashion emotional image axes are not limited to four and may be expanded. Like MBTI, this fashion emotional image axis can diagnose 16 types of fashion sensibility by diagnosing contrasting emotional axes.

The Classic-Modern fashion emotional image axis has the meaning of temporality connecting the past and the future and is based on time. Classic is an image of 'humanity, basic, exemplary, eminent, classic', and Figure 7a shows an example of classic fashion. Modern is a 'modern, futuristic, functional, simple' image, and Figure 7b shows an example of modern fashion.

The romantic-masculine (Mannish) fashion emotional image axis is based on masculinity and femininity, with sensibility close to femininity being romantic. A sensibility close to masculinity or androgyny is expressed as Mannish. Romantic is a feminine, gentle, and gentle image, and Figure 7c shows an example of romantic fashion. Mannish is an image of a woman with a cool style and masculine image, and Figure 7d shows an example of Mannish fashion.

The Elegance-Active fashion emotional image axis is static and dynamic. Something related to activity. Active if there is a lot of activity. If it is close to static, it is expressed as elegance. Elegance is an image representing elegant, classy, noble, and beautiful sensibility, and Figure 7e shows an example of elegance fashion. Active is an image that shows a casual sensibility similar to sportive with a sporty image in the background, and Figure 7f shows an example of active fashion.

The sophisticated-country fashion sensibility image axis is regional. classic. Based on urban sophistication, an urban image is expressed as Sophisticated, and the opposite is expressed as Country. Sophisticated is an image representing an intellectual, career woman, and urban sensibility, and Figure 7g shows an example of Sophisticated fashion. Country is an image representing country and primitive sensibility, and Figure 7h shows an example of country fashion.

Meanwhile, the present invention clusters customers' fashion in advance, learns a large amount of images belonging to the cluster, and then classifies the images of customers entering a specific place, for example, a store, into clusters using artificial intelligence to store the store. It can be used to understand the fashion sensibility of customers, plan, produce, and display fashion products to match the identified fashion sensibility, and change the interior. Here, the number of clusters can be added, deleted, or changed considering various factors such as trends, culture, and age.

In Figure 6, in order to cluster consumers' emotions, fashion images are represented by eight emotional elements in four image axes, and each of the four image axes is marked with a point indicating a score or percentage (%). If one point is 5 points or 5%, each point becomes 0 points (%), 5 points (%), 10 points (%), 15 points (%), 20 points as it progresses outward from the middle of the axis. You can score points (%), 25 points (%), or percentage (%).

The present invention classifies customers' fashion images into various preset clusterings by scoring them based on the eight emotional elements shown in FIG. 6, and each clustered cluster can be represented as an emotional image diagram with four axes in FIG. 6.

Figure 9 shows clustering that combines the customer's fashion image with classic sensibility, elegant sensibility, romantic sensibility, casual sensibility, modern sensibility, etc. and age group to analyze the customer's fashion taste in the customer fashion analysis system using artificial intelligence according to the present invention. This shows an example. In an embodiment of the present invention, 15 clusters of customer emotions can be set.

For example, modern fashion-oriented (Colorful Casual), romantic and aesthetic type (Lovely Romantic), modern street casual (Modern Street Casual), young traditional (Young Traditional), traditional fashion-oriented (Soft Traditional), romantic. Romantic Elegance, Romantic Casual, Contemporary Fashion-oriented, Sensitive Modern, Practical-oriented (Career Casual), Chic Casual ), conservative and elegance-oriented type (Adult Elegance), intellectual and elegance-oriented type (Intelligence Elegance), ostentatious elegance-oriented type (Pret-a-courture), and traditional conservative type (Classic).

An example of clustering will be described using Colorful Casual clustering and Classic clustering among the various clusterings shown in FIGS. 10A and 10B. Referring to Figure 10a, the Colorful Casual cluster is an image positioning of Contemporary, Romantic, and Casual, and is characterized by a global sense, high cultural acceptance, and a sophisticated sense and self. I have a strong opinion and have a strong desire to stand out. In addition, they prefer casual yet fashionable clothes, are always up to date with the latest information, have diverse hobbies, are active, are good at shopping as a way to change their mood, and read fashion magazines often.

These characteristics can be expressed as the emotional image diagram in FIG. 6 as shown in FIG. 10A. In other words, the Colorful Casual cluster can be set as a cluster composed of 25% Romanatic and Sohisticated sensibilities, 20% Active sensibilities, 15% Modern sensibilities, 10% Elegance sensibilities, and 5% Mannish sensibilities.

The traditional conservative-oriented type (Classic) is characterized by its characteristics: it values intelligence and culture as well as the luxury, elegance, and tradition of the Conservative; it does not focus on trends; it focuses on quality and brand; it chooses what suits itself best; it values traditional things. In addition, they do not focus on fashion, but value traditional quality brands and respect etiquette. They mainly use department stores for shopping, value the atmosphere of the store, and emphasize traditional conservatism.

These characteristics can be expressed as the emotional image diagram in FIG. 6 as shown in FIG. 10b. In other words, the traditional conservative type (Classic) has 25% Classic sensibility, 5% Romanatic sensibility, 10% Elegance sensibility, 20% Sohisticated sensibility, 10% Modern sensibility, 15% Mannish sensibility, 10% Active sensibility, and Country sensibility. You can set it to a cluster consisting of 10%.

Likewise, other clusters can be clustered in a similar way to the Colorful Casual or Classic clusters and represented in the emotional image diagram in Figure 6. To briefly explain the characteristics of each cluster, Lovely Romantic is a race that has been loved in any era. ribbon. It is a fashion that embraces romantic details such as frills, and is generally a type that likes a romantic mood, enjoys a romantic and leisurely life, and pursues aesthetics. In addition, as a basic sensibility, they aim for cute things, prefer an innocent and clean image, tend to go shopping frequently to change their mood, and have a preference for department stores.

Modern Street Casual is sensitive to fashion and establishes a free-spirited style with a street sensibility. It has a cheerful, bright and healthy sensibility, and is strong in mix and match, with a modern unisex feel. It has the characteristics of pursuing.

The Young Traditional type generally has a fairly conservative and passive mindset, is conscious of what others think, values TPO when choosing clothes, and likes traditional styles. In addition, they mainly prefer and purchase brands, and do not like to be behind trends or want to stand out as cutting-edge. They have a high interest in fashion, but enjoy window shopping.

Traditional fashion-oriented type (Soft Traditional) clearly asserts one's thoughts and lifestyle, focuses on work with a sense of purpose, enjoys traditional suits with an urban feel, and has both behavioral and classy aspects. Regardless of trends, they have the characteristic of preferring a polite and neat style.

The Romantic Elegance type is a woman who is always conscious of her feminine and cautious stance and always acts elegantly. She is good at using accessories and perfume, likes soft and silky type items, and wears dresses elegantly. , it has the characteristic that dignity and elegant femininity are the top priority until the end.

The Romantic Casual type is a woman who is always aware of her feminine and cautious stance at work, loves her family, is elegant and dignified, but also aims for cuteness, is feminine but has a behavioral side, and never loses her romantic sensibility. It has characteristics.

Contemporary fashion-oriented (Contemporary Casual) is a zoning centered around the career zone, and has a sophisticated sense of establishing one's own style. In addition, although she is sensitive to fashion, she experiences a highly sensitive fashion zone and digests trends to create her own style. As an active woman, she has a strong characteristic of being fashionable in the city and wanting freedom.

Sensitive Modern is a high-income female manager in a career that requires creative ideas in the Career-Adult zone, and has organizational skills that are compatible with a free-spirited, action-oriented and intelligent way of thinking. In addition, they sympathize with the sensibility and creativity of innovative designers, and their products are influenced by their imagination and life style, and have the characteristic of interpreting and expressing unique items in their own way. In addition, they pursue high-sensitivity fashion and establish their own style, and as an urban type with a modern sense, they are the type that insists on living actively and freely, and pursues urban and sophisticated things. Additionally, their basic sensibility is that they prefer an intelligent and sophisticated image, they mainly shop at famous manufacturers' stores, and are also interested in advertising.

The Career Casual type pursues stability and comfort, is family-oriented and sincere, values the inner self, and prefers a practical and neat style and an active and casual image. In addition, it includes a wide age group in their 30s and 40s and is characterized by both conservative and behavioral aspects.

Chic Casual is an adult-centered, future-oriented, fashion-driven group that has both a behavioral and sociable side. They also have a tendency to be classy but active, and wear clean and pretty colors. Like. In particular, it has the characteristic of using a lot of knitwear.

The conservative adult elegance type is an elegant woman with an old-fashioned image, a woman with elegance embedded in her body, has noble hobbies, and is a romantic type who wants to be loved even as she gets older. Additionally, they value courtesy, are conscious of others, often go shopping for a change, and enjoy reading fashion magazines.

The Intellectual Elegance type is an intellectual and elegance-oriented type with a somewhat conservative consciousness. They like sophistication and are fashion-oriented, but emphasize femininity. In addition, they aim for intellectual elegance, have a high interest in fashion, can accept this and express it as their own in a feminine way, always dress up in response to a given situation, and have a delicate characteristic of elegance.

Pret-a-courture types are top-class women who are conservative, classy, and intelligent. A woman with cultural skills and basic sensibilities of elegance. I want to live as a classy woman with elegance, and do not focus on trends, but focus on quality and brands, choose what suits me best, and like things with a sense of haute couture with unique elements. Also gorgeous. Elegance. They like luxurious and high-class items, and have a strong desire to always have class.

As described above, each cluster can be represented as an emotional image diagram such as FIGS. 10A and 10B based on FIG. 6 to correspond to the above-described characteristics.

Figure 11 shows an example of the results of photographing and analyzing a customer's fashion in the customer fashion analysis system using artificial intelligence according to the present invention and storing the result data as a directory and file.

Figure 12 shows an example of clustering the customer's fashion sensibility for analysis of the customer's fashion sensibility in the customer fashion analysis system using artificial intelligence according to the present invention. Referring to Figure 12, clustering of the customer's fashion sensibility emphasizes classic beauty. It can be clustered into the practical school, the fashion school that emphasizes feminine beauty, the artistic fashion school, and the innovative practical school.

Figure 13 shows the development of the AI Fashion Lachivium System Integration demonstration project to which the customer fashion analysis system using artificial intelligence according to the present invention is applied. Referring to FIG. 13, for fashion registration, the fashion image capturing module 1310 photographs customers visiting the store using a CCTV monitoring system and anonymizes the customers' faces using an anonymization processing solution. The AI fashion sensibility diagnosis module 1320 diagnoses fashion sensibility using an AI pattern analysis algorithm and outputs a fashion sensibility cluster. The fashion sensibility diagnosis report module 1330 corresponds to a fashion sensibility report generation system and reports the fashion sensibility for the fashion sensibility cluster as an output for the fashion sensibility diagnosis. And the Show, Print & Transmission module (1340) provides a display UI for fashion trend recommendation and can be linked to the fashion archive system.

The customer fashion analysis system using artificial intelligence according to the present invention analyzes the fashion sensibility of customers visiting the store and compares and analyzes it with the sensibility of the products on display, so that users of the customer fashion analysis system according to the present invention can determine the fashion sensibility of visiting customers. You can check how well it matches the product being sold. For example, as a result of analyzing the fashion sensibility of store customers using CCTV equipped with artificial intelligence, major customers said, “Elegance 40%. If the diagnosis is “35% modern, 25% romantic,” the sensibility of the prepared product must also be “40% elegant, 35% modern, and 25% romantic,” which can be interpreted as good sales.

By diagnosing the emotions of customers who visit the store and the prepared products through the customer fashion analysis system using artificial intelligence according to the present invention, the emotions of major customers can be reflected in the products in the next season's product planning, and the emotions of the customers and the products can be reflected. It can serve as a fashion stethoscope to diagnose your sink.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely illustrative, and those skilled in the art will understand that various modifications and other equivalent embodiments are possible therefrom. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the attached registration claims.

110: Customer fashion information collection department
120: AI fashion sensibility analysis department
122: AI customer fashion image analysis department
124: AI fashion sensibility classification unit
130: AIMD
410: input image
420, 430: BatchNormalization
440 : Dropout
450: Clustered output

Claims (7)

A customer fashion information collection unit that includes a CCTV installed in the store and photographs the customer's fashion, and collects the customer's fashion information by photographing the customer's fashion using the CCTV; and
An AI fashion sensibility analysis unit that receives the customer's fashion video data collected through the customer fashion information collection unit, classifies it into fashion sensibility image clusters according to the customer's pre-learned fashion sensibility, and outputs them;
The video captured by the CCTV of the customer fashion information collection unit is transmitted to the AI fashion sensibility analysis unit through a network,
The AI fashion emotion analysis unit performs supervised machine learning ( A customer fashion analysis system using artificial intelligence, which is characterized by clustering modeling through supervised learning, analyzing input fashion images, and classifying which cluster the input fashion images belong to based on the learned model. delete delete The method of claim 1, wherein the plurality of fashion emotional image axes are
Classic - Modern fashion sensibility image axis, Romantic-Masculine fashion sensibility image axis, Elegance - Active fashion sensibility image axis, Sophisticated - A customer fashion analysis system using artificial intelligence, characterized by including a country fashion sensibility image axis. The method of claim 1, wherein the AI fashion emotion analysis unit
It consists of a CNN (Convolutional Netural Network) structure for supervised machine learning, and when an input fashion image is received, it performs convolution (convolution) to find features from the input fashion image, and pooling (to down-sample the data that has gone through the convolution). After going through a pooling and flattening process that converts data in a data matrix format into a one-dimensional array, it is connected to a fully connected layer and learned to form a CNN.
The convolution divides the input image into three channels according to RGB (Red, Green, Blue) colors, searches for features in each channel, goes through a filtering process to classify by feature, and then divides the input image into three channels according to RGB (Red, Green, Blue) colors. A customer fashion analysis system using artificial intelligence, characterized in that one feature map is formed by adding the three corresponding channel values. The method of claim 1, wherein the AI fashion emotion analysis unit
Customer fashion video data collected through the customer fashion information collection department is received, the customer's face information is anonymized, the store customer's fashion is analyzed, and customer fashion images are classified into top fashion, bottom fashion, shoe fashion, bag fashion, etc. AI customer fashion image analysis department; and
Clustering through supervised machine learning using labeled customer fashion images as learning data based on a basic fashion sensibility diagram consisting of multiple fashion sensibility image axes to classify a large amount of fashion images according to fashion sensibility. A customer fashion analysis system using artificial intelligence that includes an AI fashion sensibility classification unit that analyzes input fashion images after modeling and classifies which cluster the input fashion images belong to based on the learned model. According to paragraph 1,
Using the fashion sensibility cluster information output by the AI fashion sensibility analysis unit, the fashion sensibility of customers visiting the store is analyzed and compared with the sensibility of the products on display, and the visiting customer's fashion sensibility is compared to the products sold in the store. A customer fashion analysis system using artificial intelligence that further includes AI MD that calculates the degree of match and identifies customer needs.