KR102261336B1 - Service systems for advertisement contents and revenue sharing that can match advertisement contents by facial recognition based on artificial intelligence technologies - Google Patents

Abstract

The present invention relates to an advertisement content sharing service system capable of performing artificial intelligence advertisement matching by face recognition and distributing advertisement revenue. The advertisement content sharing service system capable of performing artificial intelligence-based face recognition advertisement matching and distributing advertisement revenue comprises: a camera (2) disposed in an affiliated store and configured to obtain an image by photographing a customer's face; a face recognition unit (4) which recognizes the face by analyzing the face image obtained by the camera (2) based on artificial intelligence, wherein the face recognition unit recognizes and analyzes only the gender, age, emotional state, and GPS-based region without specifying the customer's identity in detail and without creating a database; an advertisement management unit (3) which retrieves a suitable advertisement by matching advertisement content with the recognized customer's face based on artificial intelligence; a display unit (7) which displays the advertisement transmitted from the advertisement management unit (3); a unit price calculation unit (9) which calculates an advertisement unit price by analyzing customer information for each affiliate store; a revenue settlement unit (11) which calculates and settles an increase and a decrease in sales according to the advertisement content; and a big data DB unit (13) in which the advertisement content collected from advertisers and customer data of each affiliate are stored. Accordingly, the present invention may search a specific face by detecting a partial area of a face.

Description

BACKGROUND ART Service systems for advertisement contents and revenue sharing that can match advertisement contents by facial recognition based on artificial intelligence technologies}

The present invention relates to a system capable of providing a face recognition advertisement matching service based on artificial intelligence technology. More specifically, a big data DB unit is built by labeling advertisement contents provided from an advertiser, and a camera installed in each affiliate store. A customer who analyzes the customer's face obtained by means of AI, but does not specifically specify the customer's identity and does not create a DB, but only identifies the customer's gender, age, emotion, and GPS-based region and uses AI to visit the affiliated store By matching the optimized advertisement content, the advertisement effect is enhanced, and the advertisement revenue settlement is made easy by the intelligent advertisement revenue settlement algorithm according to the analysis of the types and characteristics of advertisements exposed to the number of customers by artificial intelligence face recognition. It relates to services and technologies capable of performing intelligent video security crime prevention functions at the same time.

In recent years, the importance of advertising is increasing. A large amount of advertisements are being conducted through broadcast media such as TV, newspapers, flyers, and the like.

In particular, many large discount marts are opening up, and sales competition between marts and corners is also getting fiercer. Accordingly, various exhibition and promotion methods are being mobilized.

Also, in the past, the main method was to produce leaflets one by one in order to announce discount event products or new products.

In this way, the production of advertisement flyers has a significant cost burden because it mainly uses high-quality paper and color printing, and it is difficult to measure the effect, and it also causes various problems such as waste discharge and environmental destruction.

Alternatively, a method of sending spam information data such as e-mail or text message in bulk is also widely used, but recently there is a problem in that many customers delete or spam the information before reading it due to too much spam information.

Moreover, it can be said that the above-mentioned conventional public relations methods are very low in efficiency compared to the cost required because they are one-sided information delivery rather than a voluntary approach from the point of view of a customer who needs information.

Therefore, the conventional advertising method has the following problems.

First, advertising content is randomly exposed to a large number of general consumers, and unilateral advertising content is provided without considering the consumer's gender, age, emotion, etc., so advertising efficiency is lowered.

Second, there is a problem in that it is difficult to calculate the advertising fee because there is a lack of objective data on the correlation between the provided advertising content and the generation of sales.

Third, since advertisement contents are unilaterally provided without considering local circumstances or customer circumstances, it is an inefficient structure without comparison of advertisement effects or visualized data on effectiveness.

Patent Publication No. 10-2009-0083643 (Method, system and apparatus for providing publicity service using private broadcasting)

Accordingly, the present invention has been devised to solve such a problem, and an object of the present invention is to label advertisement content provided by an advertiser in various categories to build a big data DB unit, and to obtain data obtained by a camera installed in each affiliate store. We extract customers' faces from real-time images, but do not specify the customer's identity in detail and do not create a database, but recognize and analyze only gender, age, emotion, and regional characteristics based on GPS, and use AI to provide information to customers who visit the affiliated store. By matching suitable advertising content, advertising effectiveness is enhanced, and advertising revenue can be easily settled by the artificial intelligence advertising revenue settlement algorithm, while at the same time providing services and technologies that can perform security and crime prevention functions based on artificial intelligence video analysis. will do

Another object of the present invention is to provide a system capable of recognizing a face by mounting an artificial intelligence camera or artificial intelligence set-top box in a smartphone app, a personal computer web service, a face recognition edge terminal, and an advertisement display device. Alternatively, it is to provide a service and technology capable of providing recommended advertisement content based on individual tendency analysis.

Another object of the present invention is when it is necessary to quickly perform face recognition, AI-based face recognition advertisement matching and advertisement content that can search for a specific face by detecting only a partial area of the face by switching the normal mode to the quick mode To provide a shared service system.

In order to achieve the above object, an embodiment of the present invention,

a camera (2) disposed in the affiliated store to obtain an image by photographing a customer's face;

The face image obtained from the camera (2) is analyzed based on artificial intelligence to recognize the face, but without specifying the customer’s identity or creating a database, only the gender, age, emotional state, and GPS-based region are recognized. a face recognition unit 4 to analyze;

an advertisement management unit (3) for retrieving suitable advertisements by matching advertisement contents with the recognized customer's face based on artificial intelligence;

a display unit 7 for displaying advertisements transmitted from the advertisement management unit 3;

a unit price calculation unit 9 for calculating an advertisement unit price by analyzing customer information big data for each affiliate store;

a revenue settlement unit 11 for calculating revenue by calculating an increase/decrease in advertisement revenue according to the number of exposures to advertisement content and an advertisement exposure time and an increase/decrease in sales according to advertisement exposure; And

An advertisement content sharing service system (1) capable of distributing advertisement revenue and artificial intelligence-based face recognition advertisement matching including a big data DB unit (13) in which advertisement contents from advertisers and data about customers of each affiliate are stored to provide.

As described above, the advertisement content sharing service system capable of distributing advertisement revenue and face recognition advertisement matching based on artificial intelligence according to an embodiment of the present invention has the following advantages.

First, a big data DB unit is built based on the advertisement content provided by the advertiser, and the customer's face obtained by the camera installed at each affiliate is recognized, but the customer's identity is not specifically specified and the DB is not created. By classifying into regions based on gender, age, emotional state, and GPS, and matching advertisement contents suitable for the affiliated store by AI, the advertisement effect can be enhanced, and advertisement revenue can be easily settled, while at the same time, security by artificial intelligence video analysis and a crime prevention function.

Second, by recognizing the face of a person who enters or passes a specific location in a closed space such as a building or an open space such as outdoors, based on artificial intelligence technology, the person's gender, age, and emotional state such as joys and sorrows are detected. It has the advantage of being able to analyze and display customized advertisements according to the situation as images on the billboard.

Third, when providing advertising content, there is an advantage in that it can include recommended advertisements according to preferences in specific products or surrounding commercial areas by gender and age.

Fourth, the provision of advertisements considering the public interest also has the advantage of providing customized public notification advertisements by identifying the floating population of a specific area.

1 is a diagram schematically showing the structure of an advertisement content sharing service system capable of distributing advertisement revenue and matching face recognition advertisements based on artificial intelligence according to an embodiment of the present invention.
FIG. 2 is a block diagram schematically showing the structure of an advertisement content sharing service system capable of distributing advertisement matching and advertisement revenue shown in FIG. 1 .
FIG. 3 is a diagram showing a platform architecture structure of an advertisement content sharing service system capable of distributing advertisement matching and advertisement revenue shown in FIG. 1 .
4 is a view showing an application example of the advertisement content sharing service system capable of distributing the face recognition advertisement matching and advertisement revenue shown in FIG. 1 .
FIG. 5 is a block diagram schematically showing the structure of the face recognition unit shown in FIG. 1 .
6 is a block diagram schematically showing the structure of the extraction unit shown in FIG. 5 .
FIG. 7 is a view showing a state in which face data is first divided into regions in order to recognize a face by the advertisement content sharing service system capable of distributing advertisement revenue and matching face recognition advertisements based on artificial intelligence shown in FIG. 1 .
FIG. 8 is a view showing a state in which the face data divided into regions shown in FIG. 7 is secondarily calculated and analyzed according to the brightness of each pixel.
Fig. 9(a) is a diagram showing a state in which a position, size, and ratio are recognized by distributing a large number of reference points on the contour in face data, and Fig. 9(b) is a diagram showing a state in which a plurality of reference points are distributed inside the face.
FIG. 10 is a diagram illustrating a CPU structure of the face recognition unit shown in FIG. 1 .
11 is a block diagram schematically showing the structure of the advertisement management unit shown in FIG.
12 is a flowchart illustrating a method for providing advertisements based on face recognition according to another embodiment of the present invention.
13 is a flowchart illustrating the procedure of the face recognition-based advertisement providing method of FIG. 12 .
FIG. 14 is a flowchart illustrating a process of recognizing a face in a fast mode among the face recognition-based advertisement providing methods shown in FIG.

Hereinafter, an artificial intelligence-based face recognition advertisement matching service providing system according to the present invention will be described in detail with reference to the accompanying drawings.

1 to 11, the artificial intelligence-based face recognition advertisement matching service providing system 1 proposed by the present invention includes a camera 2 that is disposed in a franchise and obtains an image by photographing a customer's face; ; Recognizes a face by analyzing the face image obtained from the camera 2 based on artificial intelligence, but recognizes only the gender, age, emotional state, and GPS-based area without specifying the customer's identity in detail and without DB. and a face recognition unit 4 to analyze; an advertisement management unit (3) for retrieving suitable advertisements by matching advertisement contents with the recognized customer's face based on artificial intelligence; a display unit 7 for displaying advertisements transmitted from the advertisement management unit 3; a unit price calculation unit 9 for calculating an advertisement unit price by analyzing customer information for each affiliate store; a revenue settlement unit 11 for calculating and calculating sales increase/decrease according to advertisement content; and a big data DB unit 13 in which advertisement contents collected from advertisers and data on customers of each affiliated store are stored.

In this artificial intelligence-based face recognition advertisement matching service providing system,

The camera 2 refers to various types of cameras 2 , for example, including an IP camera, an IR camera, an HD camera, a depth camera, and the like, and may be a camera mounted on a smartphone.

The camera 2 transmits and receives data through a wired/wireless network such as a local area network (LAN), a wide area network (WAN), a virtual private network (VPN), a mobile communication network, or a satellite communication network.

The face recognition unit 4 analyzes and recognizes facial images of customers in the affiliated store or pedestrians outside the affiliated store obtained by the camera 2 based on artificial intelligence and classifies them by gender and age.

The face recognition unit 4 may recognize a face by analyzing the obtained video, extracting one or more face data, vectorizing the face data, performing function processing, and detecting facial feature points. In this case, in recognizing a face, the customer's identity is not specifically specified, but only gender, age, emotional state, and GPS-based region are recognized and analyzed. Also, the analyzed face data is not converted into DB.

In addition, the face recognition unit 4 includes an AI core such as machine learning and deep learning, an AI framework such as Tensorflow, SSD, and YOLO, an Android or Linux-based OS, and a processor such as CPU, GPU, and TPU. It supports at least 4 channels, and interfaces include USB, HDMI, Ethernet, 2.4G/5G dual-band, and Bluetooth.

In more detail,

As shown in FIG. 5 , the face recognition unit 4 analyzes an image inputted in real time by the camera 2 , simultaneously extracts one or more face data, and converts the extracted one or more face data into a vector to perform a function an extraction unit 15 for processing; a preprocessor 17 that performs post-processing by removing reflected light, angle of view, dust, and noise from the face data processed by the extraction unit 15; a feature detection unit 19 that simultaneously analyzes and detects one or more facial feature points according to the feature point parameters defined by determining the degree of similarity between the facial data processed by the preprocessor 17; A face discrimination unit 21 that registers one or more face images detected by the feature detection unit 19 in the learned database 13 or compares and analyzes them with previously registered facial feature points to recognize the gender, age, and emotional state of the face. Wow; And it includes a control unit 23 that performs face recognition by controlling the components.

When the face recognition unit 4 is described in more detail,

The extraction unit 15 extracts face data from the input video data. In this case, the face data is extracted by using various webs, for example, a Data Processing Manager (DPM), a Vision Analytics System (VAS), or the like.

As shown in FIG. 6, the extraction unit 15 includes a face dividing module 20 for dividing a face into a plurality of regions; a local feature analysis module (22) for recognizing the shape of a pixel; a vector feature analysis module (24) for calculating a vector for the brightness of a pixel; Includes a module (Surface texture analysis; 26) for analyzing the texture of the surface of the face.

When explaining the extraction process, as shown in FIG. 7 , a face region is extracted in a rectangular or oval shape from an image input in real time.

Then, after dividing the extracted face region into a plurality of regions by the face segmentation module 20, a P point (Peak Point) is extracted from the face image, and the eyes, nose, mouth, and ears are separated from the P point as the center. become aware

That is, when tracking or recognizing an object in an image, a key-point is extracted and matched from the image. In this case, it is important to extract a corner point. There are various methods for extracting corner points, such as the FAST method.

Therefore, it is necessary to determine whether an arbitrary P point is a corner point, and it is determined by the following process.

As shown in FIG. 8 , an arbitrary pixel is first selected by the local shape analysis module 22 from the extracted face data. In this case, the brightness of the selected pixel P is defined as Ip, and the threshold value is defined as t.

Then, an imaginary circle is formed around the point P, and the radius of the circle is assumed to be about 3 pixels, and 16 pixel values tangent to the circle are extracted. Of course, the size of the radius can be changed appropriately.

And, by the vector operation module 24, n or more pixels that are brighter than Ip+t by a certain value (>p+t) are consecutive in a circle made of 16 pixels, or a certain value or more than Ip-t If n or more dark (<pt) pixels are continuous, p is determined as a corner point.

For example, it is first calculated whether pixels 1 and 9 are brighter than P, and then pixels 5 and 13 are calculated. If P is a corner, then at least 3 of the computed 4 pixels, i.e. pixels 1, 5, 9, 13, are brighter than IP+t or darker than IP-t.

In this case, the number of consecutive pixels centered on P may be varied and applied, such as not only 3, but also 4, 9, 12, etc. FIG.

Further, in the above-described recognition process, when the number of n is small, for example, less than 12, there is a problem in that the number of pixel candidates for selecting P is small, so that many cannot be removed. In addition, when detecting a feature of a face image, the pixel selection may not be optimal because it depends on the distribution of corner shapes and the like.

Also, pixels determined not to be corners are excluded from the analysis. In addition, multiple features may be detected around a corner, making corner determination difficult in some cases.

Therefore, in order to compensate for these shortcomings, the following process is performed.

That is, corner points are detected by machine learning or deep learning. At this time, in order to execute deep learning, VAS (Vision Analytics System), SIM (Sense Inference Manager), etc. are used.

In more detail, the same process for detecting corner points from the face image is performed as described above.

Then, for all detected pixels, each pixel (x) of 16 pixels around the pixel is stored as a vector.

At this time, when the brightness value of the pixel is expressed as a vector (P) by the vector operation module 24 as shown in the following equation, it is much brighter than P (Pb), much darker than P (Pd), similar to P ( Ps) is classified into three values. And, using this, the brightness distribution of pixels on the circumference is expressed as a 16-dimensional ternary vector.

--식1

-- Expression 1

A new Boolean variable in Equation 1 is defined as Kp, where Kp is true if P is a corner, and false otherwise.

Therefore, by applying the Kp variable to the above formula, processing is performed based on a decision tree.

The above equations Pd, Ps, and Pb are repeatedly applied until the entropy of Kp becomes 0.

Through this process, a decision tree capable of detecting each corner is generated.

In this case, when a certain point P is recognized as a vertex, P and neighboring points adjacent to P are often also detected as vertices.

Therefore, in order to solve this problem, a preprocessing step is performed using a non-maximal suppression method.

The non-maximum suppression method is a method in which, when several adjacent points are detected as vertices, only the point with the maximum peak is left and the rest is removed.

And, since the method of determining P by the decision tree determines whether a vertex is an on/off method, it is difficult to quantify the degree of depth. (V) is applied. (Equation 2)

------식(2)

------Equation (2)

{V: score function, x: pixel value, P: corner point}

When calculating the P point by the ratio maximum suppression method, V is calculated by inputting each of the corner points detected through the decision tree into Equation (2).

That is, V means the sum of all absolute values (|x-p|) of the difference between p and 16 pixel values (x) surrounding it. Then, points with maximum cornerity are identified by calculating V for two adjacent p values, and discarding the smaller V value.

In this way, the shape and position of the eyes, nose tip, eyebrows, both ends of the mouth, ears, etc. are determined by the reference points identified as corners. These points are useful for distinguishing each individual because the shape of each individual is very different and the surrounding shape is also complex.

Other reference points are located on the cheeks or the middle of the forehead, etc. These points are not characteristic of each individual, but are necessary to grasp the overall shape of the face.

As described above, one or more face data extracted by the facial feature detection unit 19 is processed by the pre-processing unit 17, and the pre-processing unit 17 removes dust and noise such as reflected light, angle of view, etc. from the face data.

After identifying the P points in this way, as shown in FIGS. 9A and 9B , the P points are used as detailed reference points to find the position, size , and proportion of the face.

That is, each reference point is selected in random order by the facial feature detection unit 19 to be vectorized and applied to the similarity function, but the position is slightly changed to increase the similarity of the facial image image. In this process, the vector size is gradually increased from focus 1 to 5.

As shown in Fig. 9a, many reference points are distributed in the shape of the face, and the number of reference points is relatively small . As shown in FIG. 9B , a plurality of reference points are distributed at similar intervals on the inside of the face .

In the case of focus 1 in the above process, if the previously set values are used as they are, 8 pixels will be the maximum size of the position difference vector. However, according to the training set or desired performance level, the direction and frequency of the wavelet can be set arbitrarily by the user, so the pixel units in steps 1 and 2 are also adjusted accordingly and the face is determined by the pixel spacing between each defined P point. to detect the feature points of

And, if necessary, the texture on the surface of the face is also analyzed by the texture analysis module 26 .

In this way, the facial feature detection unit 19 converts features such as shape, brightness, texture, etc. of multiple pixels of the face image into vectors, processes and extracts them by a function, and then combines them with each other to determine one or more facial feature points according to the feature point parameters. Analyze and detect at the same time.

On the other hand, in order to shorten the recognition time for the face image processed by the preprocessor 17, after detecting only a part of the face image, if a specific face does not match, the remaining undetected face image is additionally detected to proceed with the recognition process. may be

That is, when the general mode is changed to the quick mode by the control unit 23, the face is divided by the face division module 20 by a predetermined region, and then the facial features of the corresponding region are detected by the facial feature detection unit 19. After detection, matching face data is retrieved from the database 13 by the face discrimination unit 21 and compared.

In more detail, horizontal and vertical lines are formed around the nose of the face and divided into a plurality of regions, for example, 4 regions, which are even-numbered regions that can be symmetrical, and the first, second, and second regions in the clockwise direction from the upper left region. 3 and 4 are set.

Then, the feature of the left eye region, which is the first region, is first detected. Then, it is calculated whether there is a part matching the corresponding area in the database 13 .

If there is no matching part, a matching face image is searched for by detecting the feature of the fourth region, that is, the region immediately below the left eye region.

That is, since the feature of the second region that is symmetrical with the first region can be processed by a mirror technique, it is preferable to detect the feature of the fourth region that is not symmetrical.

Then, when detection is not performed even in the fourth region, the entire face region is detected.

In this way, the face is divided into a plurality of regions, and features are sequentially detected and matched from a single region, but when a specific face is not matched, the features of other regions except for the symmetrically arranged region are detected and matched. By comparing data, face recognition can be performed quickly.

In addition, the recognition time can be quickly shortened by first performing the feature on the specific shape of the face before performing the feature point detection on the first to fourth regions. For example, if a certain part of the face has features such as large dots, scars, facial asymmetry, etc., and can be easily detected by the face discrimination unit 21, the recognition time can be shortened by preferentially processing these feature points. .

Meanwhile, the face determination unit 21 determines one or more face images detected by the preprocessor 17 .

The face determination unit 21 may be processed by a central processing unit (CPU) on which a face recognition analysis engine is mounted.

For reference, as shown in FIG. 10, the central processing unit is configured by a face analysis engine, a GPU, an ISP (Image Signal Processor), an ALSA (Advanced Linux Sound Architecture), an RTSP (Real Time Streaming Protocol), and a communication module. signals are sent and received.

In the face determination unit 21, various algorithms can be used to analyze the shape of the face, for example, by comparing the shape of the face by a deep learning method and determining the similarity, it is a method of analyzing the shape. .

In other words, deep learning is a machine learning method built on the basis of artificial neural networks (ANNs) to enable computers to learn on their own like humans using multiple data.

Since classification and clustering of face data is possible using an artificial neural network, similarity determination can be performed by placing various layers on the data desired for classification or clustering.

That is, the degree of similarity can be determined by extracting shape features of face data with an artificial neural network and classifying or clustering them by shape using the features as input values for other machine learning algorithms.

The artificial neural network includes a deep neural network, and the deep neural network refers to a neural network composed of several layers among neural network algorithms.

That is, the artificial neural network is composed of multiple layers, each layer is composed of several nodes, and an operation to classify the shape actually occurs at each node, and this operation process simulates the process occurring in the neurons constituting the human neural network. designed to do

When a node receives a stimulus of a certain size or more, it responds, and the magnitude of the response is roughly proportional to the value multiplied by the input value and the node's coefficient (or weights).

In general, a node receives multiple inputs and has as many coefficients as the number of inputs. Therefore, by adjusting this coefficient, different weights can be given to different inputs.

Finally, the multiplied values are all added and the sum is fed into the activation function. The result of the activation function corresponds to the output of the node, and this output is ultimately used for classification or regression analysis.

Each layer consists of several nodes, and whether each node is activated/deactivated is determined according to the input value. At this time, the input data becomes the input of the first layer, and after that, the output of each layer becomes the input of the next layer again.

All coefficients change little by little during the shape learning process of face data, and as a result, reflect which input each node considers important. And training of the neural network is the process of updating this coefficient.

When learning the shape of face data, in this deep neural network, different layer features are learned for each layer.

That is, simple and specific features are learned for low-level features, {eg, a shape composing the shape of face data}, and more complex and abstract features are learned for high-level features. {Example: nose height, eyebrow shape, lip shape, eyebrows}

Through this abstraction learning process, the deep neural network understands high-dimensional data, and hundreds of millions or billions of coefficients are involved in this process. (A non-linear function is used in this process.)

In addition, deep neural networks can use the data to figure out the latent structures of the data. In other words, potential structures such as contour, nose height, eyes, lip shape, eyebrow shape, and forehead shape of the face data can be identified. Through this, similarity between data can be effectively identified even if the data is not labeled, and as a result, deep neural networks are effective for clustering face data.

For example, a large amount of face data can be input using a neural network, and similar face data can be collected and classified.

The difference between this deep neural network and general machine learning is that feature extraction is performed automatically.

In the past, in order to extract effective features, experts in the relevant field had to devise and apply formulas or methods for extracting features directly for a long time. This method has a problem that it takes a lot of time to develop, evaluate, and supplement.

A deep neural network is an algorithm designed to make a computer take over this process, and it trains it to perform much faster and more effectively than humans.

Even when learning unlabeled data, the neural network can automatically extract features from facial data. There are several methods for this automatic extraction, and in general, this process learns to make the output equal to the input when passed through the neural network.

No matter what kind of label is (use the input as is/use a separate label), the neural network finds the correlation between the input and the output. In some cases, after training the neural network to some extent with labeled data, it is possible to continue learning by adding unlabeled data. Using this method, the performance of the neural network can be maximized.

The last layer of a deep neural network is the output layer. In most cases, the activation function of the output layer is logistic or softmax, and the probability of a specific label can be finally obtained from the output layer. For example, when facial data is input, whether the contour of the face is long vertically or horizontally, the height of the nose, the sagging of the corners of the mouth, the sagging of the corners of the eyes, the size and length of wrinkles, etc. can be obtained with respective probabilities.

First, all coefficients of the neural net are initialized before training starts. Then, face data is repeatedly input to learn. If the learning proceeds smoothly, the coefficients will be updated to appropriate values, and various classifications and predictions are possible with this artificial neural network.

In the learning process, the updating process of these coefficients occurs repeatedly.

The principle of coefficient update is to estimate the coefficient first, measure the error that occurs when the coefficient is used, and then slightly update the coefficient based on the error.

At this time, the multiple coefficients of the neural network are collectively called a model, and the model may be in an initialized state or in a state in which learning is completed.

The initialized model does not perform any meaningful work, but as learning progresses, the model outputs results similar to the actual value rather than a random value.

This is because the artificial neural network does not know anything before data is input, and the reason for initializing coefficients to arbitrary values is also the same. And as the data is read, the coefficients are updated little by little in the correct direction.

Through this update process, the artificial neural network can group similar face data by classifying the input face data.

In this case, the face data may be clustered by type, for example, may be clustered according to gender, age, and emotional state.

And, emotions may be classified into joy, sadness, anger, depression, and the like.

The face data clustered in this way is registered in the database 13 .

In this case, the data registered in the database 13 means data that can recognize only the gender, age, emotion, and GPS-based region, not specific data that can identify the customer.

On the other hand, as shown in FIG. 11 , the advertisement management unit 3 collects advertisement contents and retrieves and displays suitable advertisements according to the clustered faces of customers.

The advertisement management unit 3 includes: a collection module 30 for collecting and storing advertisement contents; a classification module 32 for classifying the collected advertisement contents; a search module 34 for searching a path for selecting advertisement content corresponding to the detected facial feature; a recommendation module 36 for recommending searched advertisement contents; a display module 38 for displaying the recommended advertisement content through the display unit 7; and a management module 40 for managing user information.

In more detail,

The collection module 30 collects and stores advertisement contents provided online or offline from advertisers. In this case, by acquiring information on an advertisement exposure target from the advertiser in a complex way, the advertisement contents can be searched by the advertisement contents matching algorithm.

The classification module 32 classifies the collected advertisement contents according to characteristics. For example, it is classified into health-related advertisements, sports-related, health-related, and financial-related advertisements.

The search module 34 searches a path for selecting advertisement content corresponding to the facial feature detected by the facial feature detection unit 19 .

The recommendation module 36 fetches and recommends the advertisement contents for which the route is searched from the database according to the classified face types. For example, a bar suitable for retrieving advertisement contents according to age is a method of retrieving advertisement contents for children in the case of a younger group, and fetching health-related contents in the case of an elderly group.

In addition, in the case of young people, health-related advertising content is withdrawn, and in the case of the elderly, financial-related content is withdrawn.

In addition, advertisement contents can be retrieved according to gender. In the case of men, sports-related contents are retrieved, and in the case of women, cosmetic-related contents are retrieved.

In addition, advertisement contents can be retrieved according to emotions. In the case of joy, travel contents are retrieved, in case of sadness, alcohol-related contents are retrieved, and in the case of anger, sound sources or games that can control anger. , travel-related advertisement content is fetched, and in a depressed state, bright content such as travel or sports is fetched.

In this way, the recommendation module 36 selects and retrieves appropriate advertisement content according to the recognized face data.

And, when it is difficult for the recommendation module 36 to recommend advertisement content according to the classified face type, for example, when the classified ratios of men and women are similar, or when the ratios of young people and the elderly are similar, the advertisement corresponding to either side It is difficult to retrieve the content.

In this case, the advertisement management unit 3 additionally includes the determination module 43 to calculate the ratio of the classified face types, and if the ratio is within a predetermined range, all the corresponding advertisement contents are displayed alternately.

That is, the determination module 43 calculates the classification ratio of the face data in real time, and if the gender, age, and region ratios are within a certain range, for example, 5:5, the determination module 43 determines the corresponding ratio to either side. Instead of displaying only the corresponding advertisement contents, both advertisement contents are displayed alternately.

In this way, the advertisement effect can be enhanced by displaying both sides of the advertisement content instead of displaying only one side of the advertisement content.

In this case, the advertisement time may be adjusted according to the ratio of both sides. For example, in the case of 4:6, the advertisement time of both sides may be displayed proportionally according to this ratio.

In addition, when displaying advertisement content, the recommendation module 36 may recommend it according to priority.

That is, the first step is to classify the customer’s face data by gender, emotion, and region as described above and display the corresponding advertisement content, the second step is to recommend the advertisement content with good sales effect, and the third step is to link the time data. It is a method of recommending advertisement content suitable for the time by dividing it into morning and evening or by dividing it into four seasons.

That is, in the first step, as described above, face data is classified by gender, emotion, and region by the recommendation module 36, and the corresponding advertisement content is retrieved and displayed.

In the second step, the types and sales of products sold by terminals such as POS machines, card payment machines, and electronic safes installed in the affiliated stores can be grasped, and the advertisement exposure time is identified by the advertisement management unit 3 and compared with each other. Specifically, it is possible to select a product with a good sales effect by grasping the increase or decrease in sales in real time due to the advertisement.

In the third step, the advertisement management unit 3 links the time data with the face data to classify it into morning and evening or four seasons, and the recommendation module 36 recommends advertisement content suitable for each time or season. . In this case, it is preferable to recommend advertisement content suitable for a time or season among the items selected as having a good sales effect in the second step.

For example, when exposing advertisements for clothing, if the gender is female, first expose advertisements for women's clothing, but expose advertisements for women's coats with high sales, and in this case, if the season is summer, A way to display advertisements. By recommending the advertisement content in this way, the efficiency of the advertisement can be improved.

The display module 38 may display the recommended advertisement content on the advertisement display window to proceed with the advertisement.

In this case, the advertisement display window 32 includes various types of advertisement display windows 32 , and for example, a TV installed in an affiliated store, an LED on an electric sign board, and the like is possible.

Meanwhile, as shown in FIG. 4 , the unit price calculation unit 9 calculates an advertisement unit price by analyzing customer information for each affiliate store.

The unit price calculation unit 9 classifies customers who have visited the affiliated store by gender, age, emotion, and region, and calculates an advertisement unit price based on the number of visitors.

In this case, the number of visitors may be analyzed by day, week, and month to calculate the advertisement unit price.

For example, if the customers who visit the cosmetics store are women, young people, and 100 people per day, a predetermined amount is calculated as the advertising unit price. In this case, even if the number of visits is the same, if the customer is a male and an elderly person, it is determined that the advertisement effect is relatively high, so it is possible to lower the advertisement unit price.

Then, the revenue settlement unit 11 calculates the increase/decrease of the advertisement cost according to the number of exposures to the advertisement content and the advertisement exposure time to settle the revenue.

At this time, there are various methods of calculating the revenue, for example, a method of calculating the number of advertisement exposures and advertisement exposure time according to the number of visitors by gender and age of each affiliated store or total affiliated stores in each region or the increase or decrease in sales caused by advertisement exposure.

That is, in the case of the method of settlement based on the increase or decrease in sales, the increase or decrease in sales generated during the period in which advertisement contents are provided to the corresponding affiliated store is calculated, and the revenue is settled by applying the advertisement unit price.

In this case, the revenue settlement unit 11 may link the advertisement management unit 3 and the terminal installed in the affiliate store in order to more objectively calculate the advertisement revenue.

That is, the type and sales of products sold by terminals such as POS machines, card payment machines, and electronic safes installed in affiliated stores can be identified, and the advertisement exposure time is identified by the advertisement management unit 3 and compared with each other to specifically correspond The increase or decrease in sales due to advertising can be grasped in real time.

In the case of the method of calculating the number of visitors, the revenue is settled by classifying customers who have visited the affiliated store by gender, age, and emotion, and applying an advertisement unit price according to the number of times of advertisements and advertisement exposure time.

In this way, it is possible to calculate a unit price for advertisement content for each affiliate store and to settle advertisement revenue.

Hereinafter, the operation process of the artificial intelligence-based face recognition advertisement matching service providing system according to a preferred embodiment of the present invention will be described in more detail with reference to the accompanying drawings.

As shown in FIGS. 12 and 13 , the method of recognizing a face by the artificial intelligence-based face recognition advertisement matching service providing system 1 uses the face recognition unit 4 to detect customers in the affiliated store or pedestrians outside the affiliated store. Faces are recognized from the captured video, advertisement contents are collected from advertisers, and advertisement contents are retrieved and displayed by matching suitable advertisements according to the faces of customers recognized by the camera and system, and advertisement revenue is settled.

At this time, when the face recognition unit 4 recognizes the customer's face, the customer's identity is not specifically specified, but only gender, age, emotional state, and GPS-based region are recognized and analyzed, and the analyzed face data is not converted to DB.

If we describe this face recognition advertisement matching method in detail,

acquiring a video by the camera 2 (S10);

An extraction step (S20) of analyzing the face image of the video input in real time through the camera (2) based on artificial intelligence by the face recognition unit (4), extracting one or more face data, converting it into a vector, and performing a function processing (S20);

A pre-processing step (S25) of removing dust and noise such as reflected light, angle of view, etc. by the pre-processing unit 17 from the extracted one or more face data;

a feature detection step (S30) of simultaneously analyzing and detecting one or more facial feature points according to a feature point parameter defined by determining the similarity of one or more extracted facial data when the controller 23 is set to the normal mode;

Classifying and outputting one or more face images detected by the preprocessor 17 according to gender, age, and emotional state by the face determination unit 21 according to an AI-based algorithm (S40);

a customized advertisement step (S50) of collecting advertisement contents by the advertisement management unit (3) and fetching and displaying suitable advertisements according to the faces of the clustered customers;

Computing the unit price and revenue for the advertising content; And

and outputting advertisement content (S60).

In this advertisement providing method,

In the video acquisition step ( S10 ), a face image is acquired in real time by shooting a video by the camera 2 .

Then, in the face data extraction step (S20), face data is extracted from the image input in real time.

That is, first, as shown in FIG. 7 , face data is extracted in a rectangular or oval shape.

Then, a P point (Peak Point) is extracted from the extracted face area, and eyes, nose, mouth, and ears are recognized around this P point.

That is, when tracking or recognizing an object in an image, a keypoint is extracted from the image and matched. In this case, it is important to extract a corner point.

Therefore, it is necessary to determine whether an arbitrary P point is a corner point, and it is determined according to the following definition.

As shown in FIG. 8 , an arbitrary pixel is first selected from the extracted face data. In this case, the brightness of the selected pixel P is defined as Ip, and the threshold value is defined as t.

Then, an imaginary circle is formed around the point P, and the radius of the circle is assumed to be about 3 pixels, and 16 pixel values tangent to the circle are extracted. Of course, the size of the radius can be changed appropriately.

In addition, in a circle consisting of 16 pixels, there are n or more consecutive (>p+t) pixels that are brighter than Ip+t by a certain value, or there are n pixels that are darker than Ip-t by a certain value or more (<pt). If more than one is continuous, p is judged as a corner point.

For example, it is first calculated whether pixels 1 and 9 are brighter than P, and then pixels 5 and 13 are calculated. If P is a corner, then at least 3 of the computed 4 pixels, i.e. pixels 1, 5, 9, 13, are brighter than IP+t or darker than IP-t.

In this case, the number of consecutive pixels centered on P may be varied and applied, such as not only 3, but also 4, 9, 12, etc. FIG.

Further, in the above-described recognition process, when the number of n is small, for example, less than 12, there is a problem in that the number of pixel candidates for selecting P is small, so that many cannot be removed. In addition, when detecting a feature of a face image, the pixel selection may not be optimal because it depends on the distribution of corner shapes and the like. Also, pixels determined not to be corners are excluded from the analysis. In addition, multiple features may be detected around a corner, making corner determination difficult in some cases.

Therefore, in order to compensate for these shortcomings, the following process is performed.

That is, corner points are detected by machine learning or deep learning.

In more detail, the same process for detecting corner points from the face image is performed as described above.

Then, for all detected pixels, 16 pixels around the pixel are stored as a vector.

At this time, the brightness value of the pixel is classified into three values: a case much brighter than P, a case much darker than P, and a case similar to P as shown in the following equation. And, using this, the brightness distribution of pixels on the circumference is expressed as a 16-dimensional ternary vector.

A new Boolean variable in the above formula is defined as Kp, where Kp is True if P is a corner, and False otherwise.

Therefore, by applying the Kp variable to the above formula, processing is performed based on a decision tree.

The above equations Pd, Ps, and Pb are repeatedly applied until the entropy of Kp becomes 0.

Through this process, a decision tree capable of detecting each corner is generated.

In this case, when a certain point P is recognized as a vertex, P and neighboring points adjacent to P are often also detected as vertices.

Therefore, to solve this problem, post-processing is performed using a non-maximal suppression method.

The non-maximum suppression method is a method in which, when several adjacent points are detected as vertices, only the point with the maximum peak is left and the rest is removed.

In addition, since the method of determining P by the decision tree determines whether the vertex is an on/off method, it is difficult to quantify the depth, so the following separate quantification function is applied do.

------식(2)

------Equation (2)

When calculating the P point by the ratio maximum suppression method, V is calculated by inputting each of the corner points detected through the decision tree into Equation (2).

That is, V means the sum of all absolute values of the difference between p and 16 pixel values surrounding it. Then, points with maximum cornerity are identified by calculating V for two adjacent p values, and discarding the smaller V value.

When defining the reference points in this way, the shape and position of the eyes, nose tip, eyebrows, both ends of the mouth, ears, etc. are basically determined. These points are useful for distinguishing each individual because the shape of each individual is very different and the surrounding shape is also complex.

Other reference points are located on the cheeks or the middle of the forehead, etc. These points are not characteristic of each individual, but are necessary to grasp the overall shape of the face.

In the pre-processing step (S25), the pre-processing unit 17 performs a pre-processing step of removing and correcting dust and noise such as reflected light, angle of view, etc. from the face data.

Then, in the face feature detection step (S25), after identifying the P points as described above, the P points are used as detailed reference points to find the position, size , and ratio of the face as shown in FIGS. 6A and 6B .

That is, each reference point is selected in a random order and the position is slightly changed to increase the similarity of the face image image. In this process, gradually increase the focus from 1 to 5.

In the case of FIG. 9A, many reference points are distributed in the shape of the face, and the number of reference points is relatively small. In the graphs of FIG. 9B , a plurality of reference points are distributed at similar intervals on the inside of the face .

In the case of focus 1 in the above process, if the previously set values are used as they are, 8 pixels will be the maximum size of the position difference vector. However, according to the training set or desired performance level, the direction and frequency of the wavelet can be set arbitrarily by the user, so the pixel units in steps 1 and 2 are also adjusted accordingly, and the face is adjusted according to the pixel interval between each defined peak. Detect feature points.

When the facial feature detection step S25 is completed in this way, the face identification step S40 proceeds.

In the face identification step (S40), one or more face images detected by the preprocessor 17 are specified by the face identification unit 21 based on artificial intelligence and classified according to gender, age, and emotional state.

That is, the face determination unit 21 extracts facial features by analyzing the shape of the face to distinguish it from other faces. The face determination unit 21 may be processed by a central processing unit (CPU) on which a face recognition analysis engine is mounted.

In the face determination unit 21, various algorithms can be used to analyze the shape of the face, for example, by comparing the shape of the face by a deep learning method and determining the similarity, it is a method of analyzing the shape. .

That is, when learning the shape of face data, the deep neural network learns features of different layers for each layer.

That is, simple and specific features are learned for low-level features, {eg, a shape composing the shape of face data}, and more complex and abstract features are learned for high-level features. {Example: nose height, eyebrow shape, eyes, corners of mouth}

Through this abstraction learning process, the deep neural network understands high-dimensional data, and hundreds of millions or billions of coefficients are involved in this process. (A non-linear function is used in this process.)

In addition, deep neural networks can use the data to figure out the latent structures of the data. In other words, potential structures such as contour, nose height, eyes, lip shape, eyebrow shape, and forehead shape of the face data can be identified. Through this, similarity between data can be effectively identified even if the data is not labeled, and as a result, deep neural networks are effective for clustering face data.

For example, a large amount of face data can be input using a neural network, and similar face data can be collected and classified.

The last layer of a deep neural network is the output layer. In most cases, the activation function of the output layer is logistic or softmax, and the probability of a specific label can be finally obtained from the output layer. For example, when face data is input, whether the contour of the face is long vertically or horizontally, the height of the nose, the sagging of the corners of the eyes, etc. can be obtained with respective probabilities.

In this way and by repeatedly inputting face data, learning proceeds. If the learning proceeds smoothly, the coefficients will be updated to appropriate values, and various classifications and predictions are possible with this artificial neural network.

In the learning process, the updating process of these coefficients occurs repeatedly.

Through this update process, the artificial neural network can group similar face data according to gender, age, and emotional state by classifying the input face data. Then, the clustered face data is registered in the database 13 .

In this case, the data registered in the database 13 means data that can recognize only the gender, age, emotion, and GPS-based region, not specific data that can identify the customer.

On the other hand, as another embodiment of the present invention, the rapid recognition step (S70) may proceed. That is, in the quick recognition step (S70), in order to shorten the recognition time for the face image extracted by the face recognition unit 4, only a part of the face image is detected, and if a specific face is not matched, the remaining undetected face image is detected. Further detection speeds up the recognition process.

That is, when the control unit 23 is set to the quick mode, a face is divided into a plurality of areas, and features are sequentially detected and matched from a single area, but when a specific face is not matched, the symmetrically arranged area is excluded. Face recognition can be quickly performed by detecting features of different regions and comparing matching face data.

The quick recognition step (S70) includes, as shown in FIG. 14, a first step (S100) of changing and setting the normal mode to the quick mode by the control unit 23; a second step (S110) of dividing the face into a plurality of even regions by the face dividing module 20; a third step (S120) of firstly detecting facial features sequentially from a single region by the facial feature detection unit 19; a fourth step (S130) of secondarily detecting features of regions that are not symmetrically arranged if they are not matched; A fifth step (S140) of tertiarily detecting and matching the features of the entire face is included in the case of not being matched in the fourth step (S130).

More specifically, in the first step ( S100 ), when the user needs to quickly perform face recognition, the mode of the control unit 23 is switched from the normal mode to the quick mode, so that the face recognition can be performed quickly.

In the second step ( S110 ), the face is divided into a plurality of regions by the face dividing module 20 . That is, horizontal and vertical lines are formed around the nose of the face to divide into a plurality of regions, for example, 4 regions that are symmetrical even-numbered regions, and clockwise from the upper left region to the first, second, third, and fourth regions. set to the area.

In the third step (S120), facial features are primarily detected sequentially from a single region by the facial feature detection unit 19 . That is, the feature of the left eye region, which is the first region, is first detected. Then, it is calculated whether there is a part matching the corresponding area in the database 13 .

If there is no matching part, in a fourth step ( S130 ), a matching face image is searched for by detecting the features of the fourth region, that is, the region immediately below the left eye region.

That is, since the feature of the second region that is symmetrical with the first region can be processed by a mirror technique, it is preferable to detect the feature of the fourth region that is not symmetrical.

And, if it is not detected even in the fourth region, the entire face region is detected as a fifth step (S140).

In this way, the registered customer's face is divided into a plurality of regions, and features are sequentially detected and matched from a single region. However, if a specific face is not matched, the face recognition algorithm already learned through the face dataset is applied. By extracting a face region from the image input by the user and analyzing the classification, age, and emotional state of men and women from the extracted face region, face recognition can be quickly performed.

And, in the third step (S120), before performing the feature point detection for the first to fourth regions, the recognition time can be quickly shortened by adding the step of first performing the feature on the specific shape of the face. . For example, if a certain part of the face has features such as large dots, scars, facial asymmetry, etc., and can be easily detected by the face discrimination unit 21, the recognition time can be shortened by preferentially processing these feature points. .

When the step of recognizing a face is completed in this way, the advertisement content is collected by the advertisement management unit 3, face data is extracted from the clustered customers, and the face is analyzed by gender, age, and emotion from the extracted face data. A customized advertisement step (S50) of fetching and displaying a suitable advertisement according to the above is performed.

In more detail,

First, the collection module 30 collects and stores the advertisement content provided online or offline from the advertiser. In this case, by acquiring information on an advertisement exposure target from the advertiser in a complex way, the advertisement contents can be searched by the advertisement contents matching algorithm.

When the advertisement contents are collected, the classification module 32 classifies the collected advertisement contents according to characteristics. For example, it is classified into health-related advertisements, sports-related, health-related, and financial-related advertisements.

Then, the search module 34 searches a path for selecting advertisement content corresponding to the facial feature detected by the facial feature detection unit 19 .

When the route is found, the recommendation module 36 fetches the advertising content from the database according to the classified face type and recommends the searched advertising content. For example, a bar suitable for retrieving advertisement contents according to age is a method of retrieving advertisement contents for children in the case of a younger group, and fetching health-related contents in the case of an elderly group.

In addition, in the case of young people, health-related advertising content is withdrawn, and in the case of the elderly, financial-related content is withdrawn.

In addition, advertisement contents can be retrieved according to gender. In the case of men, sports-related contents are retrieved, and in the case of women, cosmetic-related contents are retrieved.

In addition, advertisement contents can be retrieved according to emotions. In the case of joy, travel contents are retrieved, in case of sadness, alcohol-related contents are retrieved, and in the case of anger, sound sources or games that can control anger. , travel-related advertisement content is fetched, and in a depressed state, bright content such as travel or sports is fetched.

In this way, the recommendation module 36 selects and retrieves appropriate advertisement content according to the recognized face data.

Then, the display module 38 may proceed with the advertisement by displaying the recommended advertisement content on the advertisement display window.

At this time, the advertisement display window 32 includes various types of advertisement display windows 32 , and for example, a TV installed in an affiliated store, an LED display board, etc. are possible.

On the other hand, as shown in FIG. 12, analyzing customer information by the unit price calculating unit 9 for each affiliate store to calculate an advertisement unit price (S70); The step (S80) of calculating the revenue by calculating the sales increase or decrease due to advertisement exposure and advertisement exposure according to the number and time of exposure to advertisement content by the revenue settlement unit 11 may be additionally performed.

In the unit price calculation step (S70), the unit price calculation unit 9 analyzes customer information accumulated for each affiliate store to calculate the advertisement unit price.

The unit price calculating unit 9 classifies customers who have visited the affiliated store by gender, age, and emotion, and calculates an advertisement unit price based on the number of visitors and the time they stay.

In this case, the number of visitors may be analyzed by day, week, and month to calculate the advertisement unit price.

When this unit price calculation step (S70) is completed, the revenue settlement step (S80) proceeds.

In the revenue settlement step (S80), the revenue settlement unit 11 calculates the number and exposure time of advertisement content exposure, and sales increase/decrease according to advertisement exposure to settle the revenue.

In this case, there are various methods of calculating the revenue, for example, a method of adjusting the number of advertisement exposures, advertisement exposure time, and sales increase or decrease according to advertisement exposure or the number of visitors is possible.

That is, in the case of the method of settlement by the increase or decrease in sales, the increase or decrease in sales generated during the period in which advertisement contents are provided to the corresponding affiliated store is calculated, and the revenue is settled by applying the advertisement unit price.

In the case of the method of calculating the number of visitors, revenue is settled by classifying customers who visited regional franchisees and total affiliates connected to the platform by gender, age, and emotion, and applying the advertising unit price according to big data analysis according to the advertisement exposure time and number of times. method.

In this way, it is possible to calculate a unit price for advertisement content for each affiliate store and to settle advertisement revenue.

Claims (11)

a camera (2) disposed in the affiliated store to obtain an image by photographing a customer's face;
The face image obtained from the camera (2) is analyzed based on artificial intelligence to recognize the face, but without specifying the customer’s identity or creating a database, only the gender, age, emotional state, and GPS-based region are recognized. a face recognition unit 4 to analyze;
an advertisement management unit 3 for retrieving advertisements by matching advertisement contents with a recognized customer's face based on artificial intelligence;
a display unit 7 for displaying advertisements transmitted from the advertisement management unit 3;
a unit price calculation unit 9 for calculating an advertisement unit price by analyzing the movement amount information of the customer by time, day, week, and month of the customer for each affiliate store;
a revenue settlement unit 11 for calculating and setting advertisement revenue by analysis of exposure time and exposure effect of advertisement content and analysis of changes in sales due to advertisements; And
and a big data DB unit 13 in which advertisement contents collected from advertisers and data about customers of each affiliate are stored,
The face recognition unit 4 includes: a face dividing module 20 for dividing a face into a plurality of regions; a local feature analysis module (22) for recognizing the shape of a pixel; a vector feature analysis module (24) for calculating a vector for the brightness of a pixel; Includes a module for analyzing the texture of the surface of the face (Surface texture analysis; 26),
The face recognition unit 4 divides the face region into a plurality of square or oval shapes in the image input in real time,
A P point (Peak Point) is extracted from the extracted face area, and eyes, nose, mouth, and ears are recognized around this P point,
When determining whether the P point is a corner point, an imaginary circle having a predetermined pixel radius is formed around the P point, and it is determined by the pixel value in contact with the circle,
If n or more pixels that are brighter (>P+t) by more than a certain value than P point are consecutive, or if there are n or more consecutive pixels that are dark (<Pt) more than a certain value, point P is determined as a corner point,
It is judged by a decision tree method to determine whether the point P is a vertex,
When the brightness value of a pixel is much brighter than P point, much darker than P point, and similar to P point, it is classified into three values, and the brightness distribution of pixels on the circumference is expressed as a 16-dimensional ternary vector. After that, input this vector into the decision tree to determine whether the P point is
Recognize vertices by processing using a non-maximal suppression method,
In the method of determining the P point by the decision tree, the vertex is determined by the On/Off method, so the function is applied by the following formula,
When calculating the P point by the non-maximum suppression method, each corner point detected through the decision tree is input into Equation (1) to calculate V, and then V value higher than the corresponding P point among adjacent corner points An advertisement content sharing service system (1) capable of distributing advertisement revenue and face recognition advertisement matching based on artificial intelligence that identifies points with maximum cornerity in a manner that sequentially removes corner points with

------Equation 1
{V: score function, x: pixel value, P: corner point} delete delete delete The method of claim 1,
The advertisement management unit 3 includes: a collection module 30 for collecting and storing advertisement contents; a classification module 32 for classifying the collected advertisement contents; a search module 34 for searching a path for selecting advertisement content corresponding to the detected facial feature; a recommendation module 36 for recommending searched advertisement contents by artificial intelligence; a display module 38 for displaying recommended advertisement contents; An artificial intelligence-based face recognition advertisement matching including a management module 40 that manages user information, affiliate information, device information, advertisement content information, and customer metadata information, and advertisement contents sharing service system capable of distributing advertisement revenue (One). 6. The method of claim 5,
The advertisement management unit 3 additionally includes a determination module 43 to calculate the ratio of the classified face types, and if the ratio is within a certain range, artificial intelligence-based face recognition advertisement matching and displaying all of the corresponding advertisement contents alternately Advertisement content sharing service system (1) capable of distributing advertising revenue. 6. The method of claim 5,
The recommendation module 36 of the advertisement management unit 3 firstly classifies the customer's face data by gender, emotion, and region to display the corresponding advertisement contents, and secondly recommends advertisement contents with good sales effect, In the third phase, an advertisement content sharing service system that can distribute advertisement revenue and artificial intelligence-based face recognition advertisement matching that recommends advertisement contents at the time by linking time data into morning and evening or by dividing into 4 seasons ( One). The method of claim 1,
The control unit 23 can perform face recognition by changing the normal mode to the quick mode and dividing the face into 1 to 4 regions by the face dividing module 20, and using the facial feature detecting unit 19 to perform face recognition. It first detects facial features sequentially from a single area, if they do not match, it detects the features of areas that are not symmetrically arranged, and if they do not match, tertiarily detects the features of the entire face. An advertisement content sharing service system (1) that can distribute artificial intelligence-based face recognition advertisement matching and advertisement revenue by analyzing and matching age and age. 9. The method of claim 8,
The control unit 23 performs face recognition advertisement matching and advertisement revenue based on artificial intelligence that can shorten the recognition time by first performing the characteristics of the specific shape of the face before performing the feature point detection for the first to fourth regions. Distributable advertising content sharing service system (1). delete delete

Images