Intelligent advertisement generation method based on multimedia advertisement promotion

Abstract

The present invention discloses an intelligent advertisement generation method based on multimedia advertisement promotion, which belongs to the field of advertisement delivery technology, and specifically includes collecting facial and full-body images of persons in a scene, judging the age, gender and viewing direction of the persons according to the facial images; if a person looks at a display screen for more than a certain time, marking the person as a person interested in the current advertisement content; building a target person database, storing facial images, interest advertisements, historical riding time and advertisement playing records within the historical riding time of all target persons; if the facial image of the person in the current scene matches that in the information database, playing the interest advertisement; if not, obtaining the interest advertisements corresponding to all target persons of the same age group and gender as the non-target person, generating an advertisement collection, sorting the interest advertisements according to the frequency of the interest advertisements, and performing advertisement rotation; adjusting the advertisement delivery strategy according to different scenes and needs, and realizing accurate advertisement delivery.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement delivery, and in particular to an intelligent advertisement generation method based on multimedia advertisement promotion. Background Art

[0002] In this era of diversified information, advertising has gradually become an inseparable form of media in people's lives, affecting people's cognition of the objective world. As the number of floors in buildings increases, elevators have become an indispensable means of transportation for people to enter and exit high-rise buildings. Among them, since people must take elevators to enter and exit high-rise buildings, and elevators are used frequently, elevator advertisements are usually placed in elevators.

[0003] As one of the most important types of modern advertising, elevator advertising is affecting a rapidly increasing number of people. However, the current delivery methods of elevator advertising are relatively simple, generally only using paper advertising or video carousel. At present, the delivery of multimedia advertising is still in a "one size fits all" state. In the same period of time, any passenger entering the elevator will see the same advertisement, and it is impossible to accurately deliver advertising according to the specific situation of the audience, which greatly reduces the efficiency of advertising delivery. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent advertisement generation method based on multimedia advertisement promotion to solve the following technical problems:

[0005] At present, the delivery of multimedia advertisements is still in a "same-face" state. In the same period of time, any passenger entering the elevator will see the same advertisement. It is impossible to accurately deliver advertisements based on the specific circumstances of the audience, which greatly reduces the efficiency of advertising delivery.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] The intelligent advertisement generation method based on multimedia advertisement promotion includes the following steps:

[0008] S1, pre-stores a collection of advertisements corresponding to each age group and gender;

[0009] S2, obtaining a facial image of a person in the scene and performing facial image recognition to obtain the age and gender of the person, rotating a number of advertisement contents in an advertisement collection corresponding to the age and gender, and judging the front viewing direction of the person in real time based on the facial image. If the front viewing direction coincides with the display screen direction, recording the currently playing advertisement and the coincidence time. If the coincidence time is greater than or equal to a preset threshold, marking the currently playing advertisement content as an advertisement of interest to the person, and marking the person as a target person.

[0010] S3, build a target person database to store facial images, interest advertisements, historical ride times and advertisement playback records of all target persons during the historical ride times;

[0011] S4, constructing a face recognition model based on the TGansfoGmeG network to perform face recognition on the people in the scene. If the recognition result is a target person, obtaining the historical ride time and interest advertisement collection of the target person; estimating the current scene ride time t' according to the historical ride time, and preferentially selecting interest advertisements with advertisement duration less than or equal to the average ride time t' from the interest advertisement collection for advertisement rotation; if the recognition result is a non-target person, performing facial image recognition to obtain the age and gender of the person, obtaining all target persons of the same age and gender as the non-target person from the target person database, obtaining the interest advertisements corresponding to all target persons to generate an advertisement collection, and sorting the advertisements according to the frequency of the interest advertisements, and rotating the advertisements according to the arrangement order.

[0012] As a further solution of the present invention: in S2, the process of determining the front viewing angle is:

[0013] The HaaG feature classifier based on the OpenCV library detects the positions of facial features in facial images and determines the viewing direction of the person based on the positions of the facial features.

[0014] As a further solution of the present invention: in S2, if there are multiple advertising contents within the overlapping time T, the played advertisements are sorted according to the playback order and marked as G1, G2, ..., Gn, and the proportion of the duration of any played advertisement to the overlapping time is obtained and marked as T1, T2, ..., Tn. If Tn is greater than or equal to a preset threshold, all currently played advertising contents are marked as advertisements of interest to the person. If Tn is less than the preset threshold, the advertising content before (n-1) the current playback is marked as advertisements of interest to the person.

[0015] As a further solution of the present invention: in S4, the process of face recognition is:

[0016] A face recognition model is constructed based on the TGansfoGmeG network, and the facial image of the person in the current scene is sent to the face recognition model to extract facial features; the facial image of the target person in the target person database is sent to the face recognition model to extract facial features; the facial features are matched, and the target person whose similarity is greater than a preset threshold is marked as a match.

[0017] As a further solution of the present invention: in S4, the calculation process of the riding time t' is:

[0018] A rectangular coordinate system is established, with the number of rides as the horizontal coordinate and the ride time as the vertical coordinate, and a line graph of the ride time versus the number of rides is generated. The slope of the line between each two adjacent points in the line graph is calculated and marked as the segment slope k. A valid interval of [-0.5, 0.5] is set, and the segment slopes that are not in the valid interval are filtered out. The mean ride time in the valid interval is calculated and calibrated as the ride time t'.

[0019] As a further solution of the present invention: S4 further includes obtaining the advertisement playing record of the target user during the last ride and marking it as G1, G2, ..., Gm, obtaining the original playing time H of the played advertisement Gm and calculating If A is greater than a preset threshold, the current carousel advertisement content is modified; if A is less than the preset threshold, the current carousel advertisement content is maintained.

[0020] As a further solution of the present invention: the specific process of correction is:

[0021] Get the playing time H' of the played advertisement Gm in the playing record, divide the played advertisement Gm according to the ratio A, obtain the playing time Hm of any sub-played advertisement, and calculate If B is equal to A, the current carousel advertisement content is maintained; if B is less than A, the advertisement content [0, H] of the advertisement Gm is corrected to [Hm (A-B+1) , H], where Hm (A-B+1) According to the time sequence number A-B+1, the start time of the sub-advertisement is obtained, and the current carousel advertisement content is obtained and marked as G1, G2, ..., G b , where b is the number of advertisements of interest to the target user, the advertisement Gm' is corrected to G1 and the advertisements in the rotation are reordered and marked as G1, G2, ..., G b+1 .

[0022] As a further solution of the present invention: S4 also includes real-time acquisition of facial images of non-target members and real-time determination of the person's frontal viewing direction. If the frontal viewing direction coincides with the display screen direction, the coincidence time T is recorded. If the coincidence time T is greater than or equal to a preset threshold, the person is marked as a target person and stored in the target person database.

[0023] Beneficial effects of the present invention:

[0024] The present invention monitors the user's viewing direction and body language in real time, analyzes the user's behavior and characteristics in real time, and responds to changes in user needs in a timely manner. It can more accurately identify the user's interest points, thereby pushing advertising content that is more in line with the user's interests, increasing the conversion rate of advertisements, and estimating the current scene ride time t' based on the historical ride time, and preferentially selecting interest advertisements with advertising durations less than or equal to the ride time average t' from the interest advertisement collection for advertisement rotation; it can be understood that when determining advertisements that meet the interests of the target object, since the advertising content of each interest advertisement is different, the actual playback duration is also different. Prioritizing the selection of advertisements that match the user's ride time can improve the relevance of the advertisement to the user's current scene, thereby increasing the user's interest in and interaction with the advertisement, thereby adjusting the advertising delivery strategy according to different scenarios and needs, and achieving precise advertising delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below in conjunction with the accompanying drawings.

[0026] Figure 1 It is a flow chart of the intelligent advertisement generation method based on multimedia advertisement promotion of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] See also Figure 1 As shown, the present invention is an intelligent advertisement generation method based on multimedia advertisement promotion, S1, pre-stores advertisement collections corresponding to various age groups and genders;

[0029] S2, obtaining a facial image of a person in the scene and performing facial image recognition to obtain the age and gender of the person, rotating a number of advertisement contents in an advertisement collection corresponding to the age and gender, and judging the front viewing direction of the person in real time based on the facial image. If the front viewing direction coincides with the display screen direction, recording the currently playing advertisement and the coincidence time. If the coincidence time is greater than or equal to a preset threshold, marking the currently playing advertisement content as an advertisement of interest to the person, and marking the person as a target person.

[0030] S3, build a target person database to store facial images, interest advertisements, historical ride times and advertisement playback records of all target persons during the historical ride times;

[0031] S4, constructing a face recognition model based on the TGansfoGmeG network to perform face recognition on the person in the scene. If the recognition result is a target person, obtaining the historical ride time and interest advertisement collection of the target person; calculating the ride time t' according to the historical ride time, and preferentially selecting interest advertisements with advertisement duration less than or equal to the average ride time t' from the interest advertisement collection for advertisement rotation; if the recognition result is a non-target person, performing facial image recognition to obtain the age and gender of the person, obtaining all target persons of the same age and gender as the non-target person from the target person database, obtaining the interest advertisements corresponding to all target persons to generate an advertisement collection, and sorting the advertisements according to the frequency of the interest advertisements, and rotating the advertisements according to the arrangement order.

[0032] The present invention monitors the user's viewing direction and body language in real time, analyzes the user's behavior and characteristics in real time, and responds to changes in user needs in a timely manner. It can more accurately identify the user's interest points, thereby pushing advertising content that is more in line with the user's interests, increasing the conversion rate of advertisements, and estimating the current scene ride time t' based on the historical ride time, and preferentially selecting interest advertisements with advertising durations less than or equal to the ride time average t' from the interest advertisement collection for advertisement rotation; it can be understood that when determining advertisements that meet the interests of the target object, since the advertising content of each interest advertisement is different, the actual playback duration is also different. Prioritizing the selection of advertisements that match the user's ride time can improve the relevance of the advertisement to the user's current scene, thereby increasing the user's interest in and interaction with the advertisement, thereby adjusting the advertising delivery strategy according to different scenarios and needs, and achieving precise advertising delivery.

[0033] It is worth noting that in a timing scenario, there may be multiple people entering the scene, so the threshold of the number of passengers can be preset according to the actual scenario, and the number of people in the scene can be monitored in real time. If the number of people is less than the preset threshold, the interest advertisement of the first identified target person is played; if the number of people is greater than or equal to the preset threshold, the age and gender of all people in the scene are obtained through facial image recognition to obtain the group with the largest number of people, and then several advertisements in the advertisement collection corresponding to the group with the largest number of people are played in rotation; personalized advertising promotion is provided when there are fewer people to enhance user experience and increase user attention to advertisements; when there are more people, the group with the largest number of people is located by identifying age and gender, and advertisements that are more in line with the interests of most viewers are played to improve the overall attractiveness of the advertisement, thereby dynamically adjusting the advertising delivery strategy according to different scenarios and needs and achieving accurate advertising delivery;

[0034] It is worth noting that by monitoring the user's viewing direction and body language, analyzing the user's behavior and characteristics in real time, responding to changes in user needs in a timely manner, identifying the user's interest points, and sorting the target user's interest advertisements according to the user's riding time, and accurately delivering them to the target user according to the arrangement order, it will result in advertisements that are too long for the target user. Usually, each time the advertisement is played incompletely during the user's riding time, and the content is repeated each time, which reduces the memorability and conversion rate of the advertisement. Therefore, the present invention records the advertisement playback records in the scene during the target user's historical riding time, obtains the playback time and original playback time of the last advertisement played during the last riding time, and continues the unfinished advertisement content of the previous advertisement on the basis of the last playback, thereby maintaining the continuity of the advertisement and the coherence of the information, while avoiding repeated recommendations of content that has been seen, thereby improving the user experience. At the same time, users are exposed to different parts of the same advertisement at different time points, and the advertisement content that can be played in segments can deepen the user's memory and improve the promotion efficiency of the advertisement, thereby achieving accurate delivery of advertisements according to real-time scene information while improving the conversion efficiency of the advertisement.

[0035] In another preferred embodiment of the present invention, in S2, the process of determining the front viewing angle is:

[0036] The HaaG feature classifier based on the OpenCV library detects the positions of facial features in facial images and determines the viewing direction of the person based on the positions of the facial features.

[0037] The HaaG feature classifier is a widely verified image processing technology that can accurately detect key feature points in facial images, such as eyes, nose, mouth, etc., which helps to more accurately determine the viewing direction of the target person based on the special position of the face.

[0038] In another preferred embodiment of the present invention, in S2, if there are multiple advertising contents within the overlapping time T, the played advertisements are sorted according to the playback order and marked as G1, G2, ..., Gn, and the proportion of the duration of any played advertisement to the overlapping time is obtained and marked as T1, T2, ..., Tn. If Tn is greater than or equal to a preset threshold, all currently played advertising contents are marked as advertisements of interest to the person. If Tn is less than the preset threshold, the advertising content before (n-1) the current playback is marked as advertisements of interest to the person.

[0039] If there are multiple advertisements within the overlapping time T, it can be understood that the user has continuously paid attention to one or more advertisements for a period of time, which indicates that these advertisements have aroused the user's interest or attention. Therefore, the content of the first advertisement played must have attracted the user's interest. At the same time, the user may be interested in the last advertisement played, but the ride time is over and the user leaves the scene area, or the user is interested in the content of the previous advertisement played but not in the last advertisement watched. Therefore, the user's interest in the last advertisement cannot be accurately determined. Therefore, if Tn is less than the preset threshold, the advertisement content before the current advertisement (n-1) is marked as the person's interest advertisement.

[0040] In another preferred embodiment of the present invention, in S4, the process of face recognition is:

[0041] A face recognition model is constructed based on the TGansfoGmeG network, and the facial image of the person in the current scene is sent to the face recognition model to extract facial features; the facial image of the target person in the target person database is sent to the face recognition model to extract facial features; the facial features are matched, and the target person whose similarity is greater than a preset threshold is marked as a match.

[0042] The face recognition model built based on the TGansfoGmeG network is used for feature extraction and comparison, which improves the accuracy and efficiency of facial image matching and realizes the accurate recognition and judgment of the target person.

[0043] In another preferred embodiment of the present invention, in S4, the calculation process of the riding time t' is:

[0044] A rectangular coordinate system is established, with the number of rides as the horizontal coordinate and the ride time as the vertical coordinate, and a line graph of the ride time versus the number of rides is generated. The slope of the line between each two adjacent points in the line graph is calculated and marked as the segment slope k. A valid interval of [-0.5, 0.5] is set, and the segment slopes that are not in the valid interval are filtered out. The mean ride time in the valid interval is calculated and calibrated as the ride time t'.

[0045] Since the scenario may be affected by some external factors, such as holidays, abnormal operation of some elevators, or passenger flow that leads to the frequency of elevator use, which may cause significant changes in the ride time, when the segment slope k>0.5 or k<-0.5, the segment slope k is not included in the calculation of the average slope K to ensure the accuracy of the final ride time evaluation.

[0046] In another preferred embodiment of the present invention, the step S4 further includes obtaining the advertisement playing record of the target user during the last ride and marking it as G1, G2, ..., Gm, obtaining the original playing time H of the played advertisement Gm and calculating If A is greater than a preset threshold, the current carousel advertisement content is modified; if A is less than the preset threshold, the current carousel advertisement content is maintained.

[0047] It is understandable that by analyzing the user's historical ride time to estimate the current ride time and giving priority to advertisements that match the user's ride time, the relevance of the advertisement to the user's current scenario can be improved, thereby increasing the user's interest in and interaction with the advertisement. However, in real life, it is inevitable that some advertisements are too long compared to the target users. If only advertisements that match the user's ride time are played, the advertising content may be monotonous and lack diversity. Users may feel bored because they see the same advertisement repeatedly. If the target users are prioritized by the playback time alone and the advertising content is rotated according to the arrangement order, it will result in advertisements that are too long compared to the target users, and each playback will be incomplete and the content will be repeated each time, reducing the memorability and conversion rate of the advertisement. Therefore, it is necessary to make corresponding corrections to the carousel advertising content according to the target user's ride time. By adjusting the advertising content to adapt to the user's ride time and adjusting the advertising content according to the ride time, the advertisement can be more in line with the user's immediate needs, thereby improving the relevance and attractiveness of the advertisement and ensuring the effective communication of the advertising information.

[0048] In another preferred embodiment of the present invention, the specific process of correction is:

[0049] Get the playing time H' of the played advertisement Gm in the playing record, divide the played advertisement Gm according to the ratio A, obtain the playing time Hm of any sub-played advertisement, and calculate If B is equal to A, the current carousel advertisement content is maintained; if B is less than A, the advertisement content of the advertisement Gm is corrected to [Hm (A-B+1) , H], get the played ad Gm', get the current carousel ad content and mark it as G1, G2, ..., G b , where b is the number of advertisements of interest to the target user, the advertisement Gm' is corrected to G1 and the advertisements in the rotation are reordered and marked as G1, G2, ..., G b+1 .

[0050] It is understandable that when the duration of an advertisement that a user is interested in is much longer than the target user's ride duration, simply rotating the advertisement content based on the user's interest advertisements will result in the advertisements with too long duration being played incompletely each time and the contents of each rotation being substantially repeated, which reduces the memorability and conversion rate of the advertisements. Therefore, the playback duration H' of the played advertisement Gm in the playback record is obtained, and the played advertisement Gm is divided according to the ratio A to obtain the playback duration Hm of any sub-played advertisement. It is understandable that the played advertisement Gm is the last advertisement of interest to the target user during the target user's last ride duration, and the calculation is performed. The ratio B is understood to be that the larger the ratio B is, the more complete the advertisement Gm is played. Therefore, if B is equal to A, the current carousel advertisement content is maintained. If B is less than A, the advertisement content of the advertisement Gm is corrected to [Hm (A-B+1) , H], where Hm (A-B+1) According to the time sequence number A-B+1, the start time of the sub-advertisement is obtained, and the current carousel advertisement content is obtained and marked as G1, G2, ..., G b , where b is the number of advertisements of interest to the target user, the advertisement Gm' is corrected to G1 and the advertisements in the rotation are reordered and marked as G1, G2, ..., G b+1 It can be understood that the advertisement content of advertisement Gm is divided into A different advertisement contents, the content played by advertisement Gm played last time is obtained and the content played by Gm not played is replaced by [Hm (A-B+1) ,H], to play, continuing to watch the unfinished advertising content based on the last playback can maintain the continuity of the advertisement and the coherence of the information, while avoiding repeated recommendations of content that has already been watched, improving the user experience. At the same time, users are exposed to different parts of the same advertisement at different time points. The advertising content that can be played in segments can deepen the user's memory and improve the promotion efficiency of the advertisement, thereby achieving accurate advertising delivery based on real-time scene information while improving the conversion efficiency of the advertisement.

[0051] In another preferred embodiment of the present invention, S4 also includes real-time acquisition of facial images of non-target members and real-time determination of the person's front viewing direction. If the front viewing direction coincides with the display screen direction, the coincidence time T is recorded. If the coincidence time T is greater than or equal to a preset threshold, the person is marked as a target person and stored in the target person database.

[0052] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An intelligent advertisement generation method based on multimedia advertisement promotion, characterized in that: The following steps are involved: S1, pre-stores a collection of advertisements corresponding to each age group and gender; S2, obtaining a facial image of a person in the scene and performing facial image recognition to obtain the age and gender of the person, rotating a number of advertisement contents in an advertisement collection corresponding to the age and gender, and judging the front viewing direction of the person in real time based on the facial image. If the front viewing direction coincides with the display screen direction, recording the currently playing advertisement and the coincidence time. If the coincidence time is greater than or equal to a preset threshold, marking the currently playing advertisement content as an advertisement of interest to the person, and marking the person as a target person. S3, build a target person database to store facial images, interest advertisements, historical ride times and advertisement playback records of all target persons during the historical ride times; S4, based on the TGansfoGmeG network, a face recognition model is constructed to perform face recognition on the person in the scene. If the recognition result is a target person, the historical ride time and interest advertisement collection of the target person are obtained; the current scene ride time t' is estimated according to the historical ride time, and interest advertisements with an advertisement duration less than or equal to the ride time t' are preferentially selected from the interest advertisement collection for advertisement rotation; if the recognition result is a non-target person, facial image recognition is performed to obtain the age and gender of the person, and all target persons of the same age group and gender as the non-target person are obtained from the target person database, and interest advertisements corresponding to all target persons are obtained to generate an advertisement collection, and interest advertisements are sorted according to the frequency of interest advertisements and advertisements are rotated; The calculation process of riding time t' is: Establish a rectangular coordinate system, with the number of rides as the horizontal coordinate and the ride time as the vertical coordinate, generate a line graph of the ride time versus the number of rides, calculate the slope of the line between each two adjacent points in the line graph and mark it as the segment slope k, set a valid interval [-0.5, 0.5], filter the segment slopes that are not in the valid interval, calculate the average ride time in the valid interval and mark the average ride time as the ride time t'; Get the advertising playback record of the target person during the last ride and mark it as G1, G2, ..., Gm, get the original playback time H of the played advertisement Gm and calculate If A is greater than or equal to a preset threshold, the current carousel advertisement content is modified; if A is less than the preset threshold, the current carousel advertisement content is maintained; The specific process of correction is: Get the playing time H' of the played advertisement Gm in the playing record, divide the played advertisement Gm according to the ratio A, obtain the playing time Hm of any sub-played advertisement, and calculate If B is equal to A, the current carousel advertisement content is maintained; if B is less than A, the advertisement content [0, H] of the advertisement Gm is corrected to [Hm （A-B+1） , H], where Hm （A-B+1） According to the time sequence number A-B+1, the start time of the sub-advertisement is obtained, and the current carousel advertisement content is obtained and marked as G1, G2, ..., G b , where b is the number of advertisements of interest to the target person, the advertisement Gm' is corrected to G1 and the advertisements in the rotation are reordered and marked as G1, G2, ..., G b+1 .

2. The method for generating intelligent advertisements based on multimedia advertising promotion according to claim 1, characterized in that: In S2, the process of determining the frontal viewing angle is as follows: The HaaG feature classifier based on the OpenCV library detects the positions of facial features in facial images and determines the viewing direction of the person based on the positions of the facial features.

3. The method for generating intelligent advertisements based on multimedia advertisement promotion according to claim 1, characterized in that: In S2, if there are multiple advertising contents within the overlapping time T, the played advertisements are sorted according to the playback order and marked as G1, G2, ..., Gn, and the proportion of any played advertisement to the overlapping time is obtained and marked as T1, T2, ..., Tn. If Tn is greater than or equal to a preset threshold, all currently played advertising contents are marked as advertisements of interest to the person. If Tn is less than the preset threshold, the advertising content before (n-1) the current playback is marked as advertisements of interest to the person.

4. The method for generating intelligent advertisements based on multimedia advertising promotion according to claim 1, characterized in that: In S4, the process of face recognition is as follows: Construct a face recognition model based on the TGansfoGmeG network, send the facial image of the person in the current scene into the face recognition model to extract facial features; send the facial image of the target person in the target person database into the face recognition model to extract facial features; Facial features are matched and target persons whose similarity is greater than a preset threshold are marked as a match.

5. The method for generating intelligent advertisements based on multimedia advertisement promotion according to claim 1, characterized in that: The S4 also includes real-time acquisition of facial images of non-target members and real-time determination of the person's frontal viewing direction. If the frontal viewing direction coincides with the display screen direction, the coincidence time T is recorded. If the coincidence time T is greater than or equal to a preset threshold, the person is marked as a target person and stored in the target person database.

Images