Advertising matching method and digital signage advertising system utilizing the same

Abstract

An ad matching method and a digital signage advertising system utilizing the same are provided. Among the ad matching methods, the method for matching an ad to be displayed on a digital signage comprises the steps of: a server collecting an image captured by a camera; the server generating motion vectors for each object included in the image and determining an object to be analyzed among the objects whose motion vector direction falls within a predetermined range; the server determining the gender and age group of the object to be analyzed using a deep learning model; the server analyzing external environment data including weather, current location, and time of shooting; the server selecting a target ad that corresponds to the result of analyzing the external environment data from an ad list that satisfies both the condition of targeting the most gender and the condition of targeting the most age group among the objects to be analyzed; and the server displaying the target ad through the digital signage after a certain amount of time has elapsed since the time the image was captured.

Description

Technology Field

[0001] The present invention relates to an advertising matching method and a digital signage advertising system utilizing the same. Background Technology

[0002] Personalized advertising is a technology that provides advertisements optimized for individual consumers by analyzing users' interests, behavioral patterns, and demographic information. While traditional advertising methods delivered the same message to the masses, personalized advertising can expect high advertising effectiveness by tailoring ad content to the characteristics of individual users. Driven by advancements in internet and mobile technologies, as well as the introduction of big data and artificial intelligence (AI), personalized advertising is rapidly evolving and has now established itself as a core element of digital marketing.

[0003] In the past, advertisements were delivered uniformly through mass media such as newspapers, TV, and radio, making it difficult to quantitatively analyze their effectiveness. However, with the widespread adoption of the Internet and smartphones, data on users' online activities has accumulated, leading to advancements in technologies to optimize advertising based on this data. In particular, methods that analyze data collected from search engines, social media, and e-commerce platforms to predict individual consumers' interests and purchasing intentions have become commonplace.

[0004] However, in the digital signage advertising environment, there are limitations to applying personalized advertising due to the difficulty of analyzing ad viewers. While online advertising allows for the analysis of personal information based on cookies and user accounts, outdoor advertising and digital signage in public spaces target an unspecified audience, making it difficult to identify the characteristics of individual viewers in real time. The problem to be solved

[0005] The problem that the present invention aims to solve is to provide a method for effectively exposing advertisements to an unspecified number of people in an advertising environment utilizing digital signage, and an advertising system utilizing the same.

[0006] The problems of the present disclosure are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by a person skilled in the art from the description below. means of solving the problem

[0007] An ad matching method according to an embodiment of the present invention for solving the above problem comprises, in a method for matching an ad to be displayed on a digital signage, the steps of: a server collecting an image captured from a camera; the server generating motion vectors for each object included in the image and determining an object to be analyzed among the objects whose direction of the motion vectors falls within a predetermined range; the server determining the gender and age group of the object to be analyzed using a deep learning model; the server analyzing external environment data including weather, current location, and time of shooting; the server selecting a target ad corresponding to the result of analyzing the external environment data from an ad list that satisfies both the condition of targeting the most gender and the condition of targeting the most age group among the objects to be analyzed; and the server displaying the target ad through the digital signage after a certain amount of time has elapsed since the time the image was captured.

[0008] The object to be analyzed includes a pedestrian, and the image includes a first frame image and a second frame image. In the step of determining the object to be analyzed, the server extracts a first feature point of the objects from the first frame image and extracts a second feature point of the objects from the second frame image, and can determine the object to be analyzed by comparing the first feature point and the second feature point.

[0009] In the step of determining the object to be analyzed, the server can generate a motion vector for each object by comparing the first feature point and the second feature point.

[0010] In the step of determining the gender and age group, the server determines the gender and age group by analyzing the facial features of the object to be analyzed, and if it is difficult to directly recognize the facial features, determines the gender and age group by analyzing physical features including body proportions, body shape, and style, and if it is difficult to directly recognize the physical features, predicts the gender and age group by utilizing movement patterns.

[0011] In the step of displaying the above, the specified time may be calculated based on the reference object among the objects to be analyzed, the reference object whose motion vector direction is most similar to the direction of a predetermined reference vector.

[0012] The direction of the above reference vector may be the normal direction toward the digital signage.

[0013] The above-mentioned fixed time may be the time predicted for the reference object to arrive at a reference area set within a certain range from the location of the digital signage.

[0014] The above camera captures a first area where a pedestrian moves and a second area where a vehicle moves, and in the step of determining the object to be analyzed, the server can determine the object to be analyzed with respect to the pedestrian in the first area and determine the object to be analyzed with respect to the vehicle occupant in the second area.

[0015] A digital signage advertising system according to an embodiment of the present invention for solving the above problem comprises a camera that captures an area where pedestrians or vehicles pass, a server that analyzes objects included in an image captured by the camera and determines to display a target advertisement at a specific time, and a digital signage on which the target advertisement is displayed, wherein the server performs preprocessing to remove noise from the image, generates motion vectors for each object included in the noise-removed image, determines an object to be analyzed among the objects whose direction of the motion vector falls within a predetermined range, determines the gender and age group of the object to be analyzed using a deep learning model, and displays a target advertisement determined from an advertisement list that satisfies both the condition of targeting the most gender and the condition of targeting the most age group among the objects to be analyzed, through the digital signage after a certain amount of time has elapsed since the time the image was captured.

[0016] The above preprocessing includes noise removal, resolution adjustment, brightness and contrast adjustment, correction for object detection, background removal, and object feature point extraction, and the motion vector can be predicted by comparing the object feature points extracted from a plurality of images. Effects of the invention

[0017] According to an embodiment of the present invention, in an advertising environment utilizing digital signage, advertisements can be effectively exposed to an unspecified number of people.

[0018] In addition, through efficient data processing such as utilizing deep learning models, advertisements can be displayed on digital signage to an unspecified number of people at a specific time.

[0019] The effects according to the embodiments are not limited to those exemplified above, and a wider variety of effects are included in this specification. Brief explanation of the drawing

[0020] FIG. 1 is a block diagram schematically showing the overall configuration of a digital signage advertising system according to one embodiment of the present invention. FIG. 2 is a flowchart illustrating a method for matching target advertisements to be displayed using digital signage by a digital signage advertising system according to one embodiment of the present invention. Figure 3 is a diagram illustrating some steps of Figure 2. Figure 4 is a flowchart illustrating some steps of Figure 2. FIG. 5 is a block diagram showing the schematic configuration of a server according to one embodiment of the present invention. Specific details for implementing the invention

[0021] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims.

[0022] Although terms such as "first," "second," etc., are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are used merely to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may be the second component within the technical scope of the present invention. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0023] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. Identical or similar reference numerals are used for identical components in the drawings.

[0024] FIG. 1 is a block diagram schematically showing the overall configuration of a digital signage advertising system according to one embodiment of the present invention.

[0025] A digital signage advertising system (1) (hereinafter referred to as the 'system') utilizes digital signage (30) and includes a server (10), a camera (20), and digital signage (30).

[0026] In one embodiment, the server (10) is connected to the camera (20) and the digital signage (30) to transmit and receive data. The server (10) can analyze images collected from the camera (20) and match advertisements to be displayed through the digital signage (30).

[0027] The camera (20) can photograph the location where pedestrians or vehicles are moving at regular intervals. To accurately recognize pedestrians and vehicles, the camera (20) may be a high-resolution (HD, 4K or higher) camera (20). Additionally, the camera (20) may be equipped with a wide-angle lens to photograph wide roads and pedestrian walkways at once, and may be equipped with a variable focal lens with adjustable focal length to clearly capture distant objects.

[0028] For example, the camera (20) may be equipped with infrared (IR) technology and HDR (High Dynamic Range) technology to enable accurate shooting even at night or in bad weather. The camera (20) can detect pedestrians and vehicles in dark environments through infrared technology, and can obtain clear images even in daylight with strong light or at tunnel entrances through HDR technology. Additionally, the camera (20) may be equipped with a thermal imaging sensor, which can be utilized to detect objects even in foggy weather.

[0029] The camera (20) can be applied in various forms, such as a fixed CCTV, a mobile camera, or a camera with built-in digital signage, to photograph pedestrians or vehicles on the street. For example, if the camera (20) is a fixed CCTV camera, it can be installed at traffic lights, streetlights, bus stops, etc. on the street to continuously collect pedestrian and vehicle flow. Through this, the flow of people by time of day can be analyzed, and advertisements can be optimized to match the time of day when specific age groups and genders gather in large numbers.

[0030] If the camera (20) is a mobile camera, it can be mounted on a drone or vehicle to photograph a specific area. Here, when applied to a drone, it is effective for analyzing pedestrian flow in event venues, festivals, and outdoor spaces, and when applied to a vehicle, it can be installed on a bus or taxi to analyze the flow of people in the city center in real time and use to adjust advertising strategies.

[0031] If the camera (20) is built into the digital signage, it can be used to analyze the age and gender of pedestrians or vehicle users and then adjust advertising strategies that match those characteristics.

[0032] Digital signage (30) refers to a smart advertising display device that provides advertising, information, and entertainment content. Digital signage (30) can provide customized advertisements to users through real-time data analysis. In particular, digital signage (30) is equipped with a large display of a certain size or larger, which can significantly increase the probability that pedestrians and vehicle users will receive advertisements on the street. While existing small billboards or mobile advertisements require users to actively pay attention, digital signage (30) equipped with a large display provides a wide viewing angle and high visibility, allowing pedestrians and vehicle users to naturally recognize advertisements.

[0033] Digital signage (30) can be manufactured in various sizes depending on the installation location and purpose. It is mainly used in outdoor and indoor environments and can be customized for various spaces such as large billboards, transportation facilities, commercial spaces, and public places. In particular, digital signage (30) can provide large advertising effects outdoors and functions for providing information and enhancing customer experience indoors.

[0034] FIG. 2 is a flowchart illustrating a method for matching a target advertisement to be displayed using digital signage by a digital signage advertising system according to an embodiment of the present invention. FIG. 3 is a diagram for explaining some steps of FIG. 2. FIG. 4 is a flowchart for explaining some steps of FIG. 2.

[0035] Referring to FIG. 2, the target ad matching method includes an image collection step (S110), a preprocessing step (S120), a target object determination step (S130), a gender and age group analysis step (S140), an external environment analysis step (S150), a target ad selection step (S160), and an ad display step (S170).

[0036] The image collection step (S110) is a step in which the server (10) collects an image captured from the camera (20).

[0037] Referring together with FIG. 3, the camera (20) can photograph at least one area adjacent to the camera (20) and the digital signage (30). In one embodiment, the camera (20) can photograph a first area (R1) where a pedestrian moves and a second area (R2) where a vehicle moves. According to an embodiment, the system (1) may include a plurality of cameras.

[0038] In one embodiment, the camera (20) can photograph the first area (R1) and the second area (R2) at regular intervals. The camera (20) can provide the captured images to the server (10) at regular intervals. The server (10) can collect images from the camera (20).

[0039] The preprocessing step (S120) is a step in which the server (10) preprocesses images collected from the camera (20) for analysis.

[0040] To provide customized advertisements through digital signage (30), images of pedestrians and vehicles moving on the street must be effectively analyzed. However, the original images may contain noise, and the image quality may vary depending on various lighting conditions and shooting angles. Therefore, an image preprocessing process is essential to improve the accuracy of AI-based object recognition and analysis.

[0041] In one embodiment, the preprocessing step (S120) may include detailed steps such as noise removal, resolution adjustment, brightness and contrast adjustment, correction for object detection, background removal, and object feature point extraction. The detailed steps constituting the preprocessing step (S120) may be performed using computer vision, deep learning models, and image processing algorithms.

[0042] For example, the server (10) can remove noise by applying a Gaussian Blur, using a Median Filter, or applying an Adaptive Filter. The server (10) can remove noise that occurs depending on the weather, lighting conditions, camera lens condition, etc.

[0043] The server (10) can adjust the resolution using algorithms such as image resizing and aspect ratio preservation. Since deep learning-based object detection models such as YOLO and Faster R-CNN are optimized for the size of the input image, the image size must be adjusted to match this. The size can be changed while maintaining the aspect ratio so that the shape of the pedestrian or vehicle is not distorted, and information loss can be minimized by down-scaling using the INTER_AREA method, for example, using the cv2.resize() function of OpenCV.

[0044] The server (10) can adjust brightness and contrast by using algorithms such as applying histogram equalization and utilizing gamma correction. Since contrast increases during the day when there is strong light and decreases during the evening or at night, for example, by applying the CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm, a clear image can be obtained by adjusting the brightness difference between regions. In addition, the brightness of an image taken in a dark environment can be improved by adjusting the γ (gamma) value, for example, by setting the gamma value to 1.5 to 2.0, the subject can be made to appear clearer when shooting at night.

[0045] For example, the server (10) can separate a stationary background from a moving object (pedestrian, vehicle) by applying OpenCV Background Subtraction or by utilizing a MOG2, KNN-based background removal model.

[0046] The analysis target object determination step (S130) is a step in which the server (10) extracts an analysis target object among the objects included in the collected image.

[0047] The object subject to analysis refers to an object that can be exposed to an advertisement displayed on the digital signage (30) at a specific time depending on the direction of movement. For example, in the case of an object moving away from the digital signage (30), the probability of being exposed to an advertisement displayed on the digital signage (30) is very low, but in the case of an object moving toward the digital signage (30), the probability of being exposed to an advertisement displayed on the digital signage (30) is above a certain level.

[0048] In one embodiment, the server (10) can identify objects (pedestrians, vehicles) in images collected from the camera (20) by utilizing a deep learning model. For example, the server (10) can utilize object detection models such as YOLO (You Only Look Once) and Faster R-CNN for object identification.

[0049] In one embodiment, the server (10) can extract first feature points of objects from a first frame image collected from a camera (20) and extract second feature points of objects from a second frame image and compare the first feature points and the second feature points. According to the embodiment, the process of extracting each feature point may be performed in a preprocessing step (S120).

[0050] For example, algorithms such as the Harris Corner Detector, SIFT (Scale-Invariant Feature Transform), or ORB (Oriented FAST and Rotated BRIEF) can be used as methods to extract feature points. Here, the first frame image and the second frame image are images taken at different points in time. For example, the second frame image may be an image collected continuously with the first frame image on a frame-by-frame basis.

[0051] In one embodiment, the server (10) can extract objects having a second feature point that is the same as a first feature point from a second frame image and analyze them as objects for advertising matching. The server (10) can exclude objects having a second feature point different from a first feature point from the analysis target.

[0052] In one embodiment, the server (10) can predict the motion vectors of all objects included in the second frame image by comparing the first feature point and the second feature point to calculate the direction and distance of movement. The server (10) can determine that among the objects, objects whose motion vector directions all fall within a predetermined range are objects to be analyzed. Additionally, the server (10) can determine the number of objects to be analyzed.

[0053] For example, in FIG. 3, the server (10) may determine the first object (901), the second object (902), and the fourth object (904) as subjects for analysis, wherein the direction of each motion vector falls within a predetermined range among the first object (901) having a first motion vector (MV1) included in the first area (R1), the second object (902) having a second motion vector (MV2), the third object (903) having a third motion vector (MV3), the fourth object having a fourth motion vector (MV4), and the fifth object (905) having a fifth motion vector (MV5). Additionally, the server (10) may determine that the sixth object (906) having a sixth motion vector (MV6) included in the second area (R2) falls within a predetermined range for the direction of each motion vector, and determine the sixth object (906) as a subject for analysis. Furthermore, in the example of FIG. 3, the server (10) may determine the number of subjects for analysis to be four. The method by which the server (10) determines whether the direction of each object's motion vector falls within a predetermined range will be described later.

[0054] The server (10) can reduce the amount of computation while extracting the accurate direction of movement by utilizing a method that predicts motion vectors based on feature point matching.

[0055] The gender and age group analysis step (S140) is a step in which the server (10) determines the gender and age group of the object to be analyzed.

[0056] The server (10) can analyze the gender by utilizing the facial features, body features, and movement patterns of each object to be analyzed. In one embodiment, the server (10) can first determine the gender and age group by analyzing the facial features of the object to be analyzed (S141). Nevertheless, if it is difficult to directly recognize the facial features, the server (10) can determine the gender and age group by analyzing body features including body proportions, body shape, and style (S142). Nevertheless, if it is difficult to directly recognize the body features, the server (10) can predict the gender and age group by utilizing movement patterns (S143).

[0057] In one embodiment, the server (10) can classify gender based on face images by applying a gender classification model based on ResNet, MobileNet, or VGG-16. At this time, the server (10) can be pre-trained with large-scale face datasets such as CelebA and Adience. For example, the dataset may include features where men have relatively wide jaws, thick eyebrows, and low-volume cheekbones, and features where women have relatively soft facial contours, thin eyebrows, and high cheekbones.

[0058] In one embodiment, the server (10) can analyze the gender of an object wearing a mask. For example, even if the entire face of the object is not visible, the gender can be predicted by utilizing the eyes, forehead, and facial contours. The server (10) can utilize a deep learning-based mask-wearing image gender classification model.

[0059] In one embodiment, the server (10) can predict gender by analyzing body proportions and style when it is difficult to directly recognize the face. For example, the server (10) can analyze the shoulder-pelvis ratio and body shape using OpenPose and MediaPipe. The server (10) can analyze body contours based on male characteristics, such as relatively wide shoulders, narrow hips, and a taller average height, and female characteristics, such as relatively narrow shoulders, wide hips, and a tendency for shorter strides. Additionally, the server (10) can predict gender by utilizing a CNN-based clothing image classification model based on the tendency of men and women to prefer different fashion styles on average.

[0060] In one embodiment, the server (10) can predict gender by utilizing movement patterns when it is difficult to directly recognize a face or when it is difficult to analyze body proportions and style. For example, the server (10) can analyze the gait based on the characteristics that men have a large stride and fast walking speed and have little upper body movement, and women have a relatively small stride and relatively large pelvic movement. In this case, the server (10) can use a deep learning-based gait analysis model (e.g., GaitSet, GaitGAN) or analyze the gait cycle after joint extraction using OpenPose or MediaPipe. Additionally, the server (10) can analyze the movement patterns of the object's arms and hands based on the characteristics that men swing their arms less on average and have small hand movements, and women have more delicate hand movements and relatively large arm swings.

[0061] The server (10) can analyze the age group by utilizing the facial features, body features, and movement patterns of each object to be analyzed. For example, the server (10) can classify the objects into age ranges (e.g., 0-10, 11-20, 21-30, 31-40, 41-50, 51+) through analysis, and the aging patterns by age group are as follows.

[0062] - 0-10 years: Round face, chubby cheeks, and almost no wrinkles

[0063] - Ages 11–20: Skin becomes smooth and facial contours gradually become distinct

[0064] - Ages 21–30: Skin elasticity is maintained, but the facial bone structure is in a mature form.

[0065] - Ages 31–40: Fine wrinkles begin to appear around the eyes and forehead

[0066] - Ages 41-50: Skin elasticity decreases and the facial contour tends to sag

[0067] - Age 51+: Deep wrinkles increase around the eyes and mouth, and skin elasticity significantly decreases.

[0068] The server (10) can extract face features using a CNN-based deep learning model (e.g., ResNet, VGG, MobileNet) and can utilize a pre-trained age classification model (e.g., FaceNet, DeepFace, InsightFace). The server (10) can classify into a specific age range after training with an age dataset (e.g., Adience, IMDB-WIKI).

[0069] In one embodiment, the server (10) can predict the age group by analyzing body proportions and body shape when the face is not visible or the shooting environment of the camera (20) is limited. For example, the server (10) can use methods such as analyzing the head size and height ratio of an object (e.g., children have relatively large head sizes) and can analyze body contours using OpenPose and MediaPipe.

[0070] In one embodiment, the server (10) can predict gender by utilizing movement patterns when it is difficult to directly recognize a face or when it is difficult to analyze body proportions and body shape. Since the way of walking and the ability to maintain posture differ depending on the age group, the age can be estimated by analyzing walking patterns and body movements.

[0071] The server (10) can sequentially analyze the facial features, body features, and movement patterns of each object to be analyzed, thereby increasing the accuracy and speed of data processing.

[0072] The external environment analysis step (S150) is a step in which the server (10) analyzes external environment data such as weather, current location, and time of shooting. The server (10) can analyze external environment data to maximize the effectiveness of customized advertising through digital signage (30). For example, weather affects consumer psychology and purchasing patterns, and location information can be useful for providing goods and services that may be needed in real time around the user (or the location of the digital signage (30)).

[0073] For example, since consumers' product interests change depending on weather conditions, real-time weather data can be analyzed to display advertisements tailored to the weather. During hot weather (summer, heatwaves), advertisements for cold beverages (iced coffee, carbonated drinks, beer), summer fashion (short sleeves, sandals, sunglasses), and sunscreen can be effective. During cold weather (winter, cold waves), advertisements for warm beverages (hot chocolate, coffee, tea), winter gear (down jackets, scarves, gloves, heaters), and indoor activities (OTT programs, food delivery) can be effective. On rainy days, advertisements for umbrellas, waterproof shoes, raincoats, public transportation, and taxi-hailing services can be effective, while on snowy days, advertisements for winter gear, ski resorts, hot spring trips, and car snow chains can be effective.

[0074] In addition, by analyzing the current location, advertisements related to the area where the object is located can be provided, thereby maximizing the advertising effect. For example, digital signage (30) in shopping malls and department stores can provide information on brands and discount events, provide advertisements for seasonal new products and limited discounts near fashion stores, and provide advertisements for popular product recommendations and sample giveaway events near cosmetics stores, thereby maximizing the effect.

[0075] Furthermore, since consumer behavior patterns vary by time of day, advertising effectiveness can be maximized by providing advertisements tailored to specific time slots. For example, advertising effectiveness can be maximized by offering ads for coffee, bakeries, news, and financial apps in the morning (06:00–09:00, commuting time); ads for nearby restaurants and delivery services during lunch (12:00–14:00, mealtime); ads for food delivery, convenience store discounts, and OTT services during the evening (18:00–21:00, after work); and ads for sleep products, overseas direct purchases, and online shopping mall discounts during late night (22:00–01:00, relaxation and self-improvement time).

[0076] The target advertisement selection step (S160) is a step in which the server (10) determines the advertisement to be displayed through the digital signage (30).

[0077] A list of advertisements based on gender and age can be stored in the server (10). For example, a list of advertisements classified as shown in Table 1 below can be stored in the server (10).

[0078] By gender and age male female 0-10 years old List 1 List 2 11-20 years old List 3 List 4 21-30 years old List 5 List 6 31-40 years old List 7 List 8 41-50 years old List 9 List 10 51+ years old List 11 List 12

[0079] For example, in Table 1, advertisements targeting women are those belonging to List 2, List 4, List 6, List 8, List 10, and List 12. Additionally, advertisements targeting the 21-30 age group are those belonging to List 5 and List 6. Advertisements belonging to List 6 are advertisements that satisfy both the condition of targeting women and the condition of targeting the 21-30 age group. In one embodiment, the server (10) can analyze the objects to be analyzed and determine from the list that targets both the gender with the largest proportion and the age group with the largest proportion. For example, if among the objects to be analyzed, there are more women than men and the age group with the largest proportion is 21-30, the server (10) can select a target advertisement to be displayed at a specific time from among the advertisements corresponding to "List 6" in Table 1.

[0080] In one embodiment, the server (10) can set an advertisement that matches the surrounding environment analysis result among the advertisements belonging to the selected list as a target advertisement.

[0081] In one embodiment, for advertisements belonging to a selected list, advertisement A i Assuming that it has a Feature Vector with Weather, Location, and Time Slot as elements, it is scored (S(A) using a vector matching method based on Cosine Similarity to select the advertisement most suitable for the current environment (weather, location, time slot). i ))can do (Mathematical Formula 1).

[0082]

[0083] In mathematical formula 1, each advertisement A i Is It is defined as having a 3-dimensional vector such as. is advertisement A i is a suitable weather vector, is advertisement A iis a suitable position vector, and is advertisement A i is a suitable time zone vector. In addition, the current environment (weather, location, time zone) is also It is defined in a vector format such as. is the current weather vector (e.g., clear, rain, snow, heat, cold, etc.), and is the current location vector (e.g., downtown, shopping mall, park, beach, etc.), and is the current time zone vector (e.g., morning, noon, evening, late night, etc.). The server (10) ultimately has the highest S(A i You can select an ad with a value as a target ad.

[0084] For example, the current environment is

[0085] Current weather: Hot weather (W hot = 1)

[0086] Current Location: Downtown (L downtown = 1),

[0087] Current time: Time to leave work (T evening Assuming that = 1)," am.

[0088] In addition, assuming that the vectors of the advertisements belonging to selected list 6 are as shown in Table 2 below,

[0089] List 6 Ice cream advertisement 1.0 (Heat) 0.5 (city center) 0.3 (PM) Coffee advertisement 0.5 (heat) 0.9 (City Center) 0.8 (evening) Late-night snack advertisement 0.3 (heat) 0.6 (City Center) 1.0 (night)

[0090] The cosine similarity for each advertisement is calculated as "Ice cream advertisement: 0.90, coffee advertisement: 0.97, late-night snack advertisement: 0.91," and the server (10) can determine the coffee advertisement as the target advertisement. Utilizing cosine similarity allows for the reasonable selection of the advertisement most similar to the current environment, has a low computational load, and is suitable for real-time advertisement optimization. In addition, it allows for flexible response to changes in the environment (e.g., when winter comes, advertisements for cold-weather products are naturally selected).

[0091] The advertisement display step (S170) is a step of determining the time at which a target advertisement is displayed and displaying the target advertisement through digital signage (30) at the said time.

[0092] In one embodiment, the server (10) can determine the time at which a target advertisement is displayed by analyzing the motion vectors of the objects. In one embodiment, the server (10) can determine the time at which the advertisement is displayed as the predicted time at which the first object (901) (reference object), whose motion vector direction is most similar to the direction of the reference vector, arrives at a reference area based on the position of the digital signage (30) when the motion vector direction moves at the average magnitude (velocity) of each motion vector of the objects belonging to the group of candidate advertisement objects, in a direction predicted to move from the position on the second frame image (Equation 2 below).

[0093]

[0094] In mathematical formula 2,

[0095] : Initial position of the first object (901) at time (time when the second frame image was taken).

[0096] : Average speed of objects belonging to the ad candidate object group ( Defined as).

[0097] (In the example of Fig. 3, )

[0098] : Reference area, here .

[0099] : Estimated time of arrival at the reference area."

[0100] In one embodiment, the reference vector may be a normal direction vector toward the digital signage (30).

[0101] FIG. 5 is a block diagram showing the schematic configuration of a server according to one embodiment of the present invention.

[0102] Referring to FIG. 5, the server (10) may each include at least one processor (110) and a memory (120) that stores instructions instructing the at least one processor (110) to perform at least one operation. Here, the at least one processor (110) may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. In one embodiment, the memory (120) may additionally be configured with at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory (120) may additionally be configured with at least one of a read-only memory (ROM) and a random access memory (RAM).

[0103] Additionally, the server (10) may include a transceiver (130) that performs communication through a communication network. Additionally, the server (10) may further include an input interface device (140), an output interface device (150), a storage device (160), etc. Each component included in the server (10) may be connected by a bus (170) to communicate with one another. The storage device (160) may be an HDD (hard disk drive), an SSD (solid state drive), etc.

[0104] Meanwhile, the server (10) may include functions such as a push server, a gateway server, and a name server.

[0105] In one embodiment, the server (10) may further include a deep learning network (180). The deep learning network (180) may include a training data set and a deep learning model trained by the training data set.

[0106] Although embodiments according to the technical concept of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without changing its technical concept or essential features. The embodiments described above should be understood as illustrative in all respects and not restrictive. Explanation of the symbols

[0107] 1: Digital Signage Advertising System 10: Server 20: Camera 30: Digital Signage

Claims

Claim 1 A method for matching advertisements to be displayed on digital signage, comprising: a step in which a server collects an image captured from a camera; a step in which the server generates motion vectors for each object included in the image and determines an object to be analyzed among the objects whose motion vector direction falls within a predetermined range; a step in which the server determines the gender and age group of the object to be analyzed using a deep learning model; a step in which the server analyzes external environment data including weather, current location, and time of shooting; a step in which the server selects a target advertisement corresponding to the result of analyzing the external environment data from an advertisement list that satisfies both the condition of targeting the most gender and the condition of targeting the most age group among the objects to be analyzed; and a step in which the server displays the target advertisement through the digital signage after a certain period of time has elapsed since the time the image was captured. Claim 2 An ad matching method according to claim 1, wherein the object to be analyzed includes a pedestrian, the image includes a first frame image and a second frame image, and in the step of determining the object to be analyzed, the server extracts a first feature point of objects from the first frame image and extracts a second feature point of objects from the second frame image, and determines the object to be analyzed by comparing the first feature point and the second feature point. Claim 3 An ad matching method according to claim 2, wherein in the step of determining the object to be analyzed, the server compares the first feature point and the second feature point to generate a motion vector for each object. Claim 4 An ad matching method according to claim 1, wherein in the step of determining gender and age group, the server determines the gender and age group by analyzing the facial features of the object to be analyzed, and if it is difficult to directly recognize the facial features, determines the gender and age group by analyzing body features including body proportions, body shape, and style, and if it is difficult to directly recognize the body features, predicts the gender and age group by utilizing movement patterns. Claim 5 An ad matching method according to claim 1, wherein in the display step, the predetermined time is calculated based on the reference object among the analysis target objects whose motion vector direction is most similar to the direction of a predetermined reference vector. Claim 6 In claim 5, the direction of the reference vector is a normal direction toward the digital signage, an advertising matching method. Claim 7 An advertising matching method according to claim 5, wherein the aforementioned fixed time is the time predicted for the reference object to arrive at a reference area set within a certain range from the location of the digital signage. Claim 8 An ad matching method according to claim 1, wherein the camera photographs a first area where a pedestrian moves and a second area where a vehicle moves, and in the step of determining the object to be analyzed, the server determines the object to be analyzed with respect to the pedestrian in the first area and determines the object to be analyzed with respect to the vehicle occupant in the second area. Claim 9 A digital signage advertising system comprising: a camera that captures an area where pedestrians or vehicles pass; a server that analyzes objects included in an image captured by the camera and determines to display a target advertisement at a specific time; and a digital signage on which the target advertisement is displayed, wherein the server performs preprocessing to remove noise from the image, generates motion vectors for each object included in the noise-removed image, determines an object among the objects whose motion vector direction falls within a predetermined range, determines the gender and age group of the object among the object among the object among the object among the object of analysis using a deep learning model, and displays a target advertisement determined from an advertisement list that satisfies both the condition of targeting the most gender and the condition of targeting the most age group among the object among the object among the object among the object of analysis through the digital signage after a certain time has elapsed from the time the image was captured. Claim 10 A digital signage advertising system according to claim 9, wherein the preprocessing includes noise removal, resolution adjustment, brightness and contrast adjustment, correction for object detection, background removal, and object feature point extraction, and the motion vector is predicted by comparing the object feature points extracted from a plurality of images.

Images