Advertisement promotion system and method for determining consumption index based on dressing of crowd in elevator

Abstract

The invention relates to an advertisement promotion system and method for determining a consumption index based on dressing of people in an elevator. Comprising an image acquisition module, an image restoration module, an image recognition module, a data integration module, a consumption index calculation module and a data security and privacy protection module. Panoramic images of crowds in an elevator are captured through a plurality of cameras, denoising, enhancement and background segmentation are carried out on the images through a deep learning algorithm, dressing grades are recognized through a hybrid neural network, integration is carried out in combination with member data, payment data and social media data, and finally consumption indexes are generated and dynamically adjusted. The advertisement content is accurately pushed according to the consumption index, and multi-level privacy protection measures such as an encryption algorithm and a protocol are adopted. According to the method, the advertisement accuracy and the user participation degree are remarkably improved, the advertisement content can be adjusted in real time, different scene requirements are met, meanwhile, user privacy and data safety are guaranteed, and the method is suitable for advertisement popularization in various complex elevator environments.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and data analysis, and in particular to an advertising promotion system and method for determining a consumption index based on the clothing of people in an elevator. Background Art

[0002] With the intelligent transformation of the advertising industry, advertising push systems based on user online behavior data have gradually become mainstream. However, this approach that relies on online data has the following shortcomings:

[0003] Data uniformity: The existing system mainly relies on online behavioral data, ignoring users' offline behavior and failing to fully reflect users' actual consumption capacity and preferences.

[0004] Lack of real-time performance: The content of advertising push is not updated in a timely manner, making it difficult to accurately deliver ads based on the user's real-time status in a specific scenario.

[0005] Poor adaptability to offline scenarios: In complex offline environments such as elevators, existing technologies cannot guarantee the accuracy of image acquisition and recognition, thus affecting the effectiveness of advertising.

[0006] Insufficient privacy protection: The existing system lacks effective privacy protection mechanisms in data collection and use, which may lead to the leakage or abuse of user data.

[0007] In the existing technology, although there are some advertising systems based on computer vision, they mainly focus on simple crowd classification, have difficulty in identifying the user's clothing level, and lack the ability to integrate multi-source data. These shortcomings seriously affect the accuracy and effectiveness of advertising push. Summary of the invention

[0008] In order to solve the problems of limited recognition accuracy, insufficient data integration capability and lack of dynamic adjustment capability, the present invention proposes an advertising promotion system and method for determining the consumption index based on the clothing of people in the elevator. The system captures user images through multiple cameras, combines image processing and recognition technology, accurately identifies the user's clothing level, and integrates it with multi-source data to generate a dynamically adjusted consumption index, thereby achieving accurate advertising push. In addition, the system adopts strict privacy protection measures to ensure the security of user data.

[0009] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0010] An advertising promotion system for determining a consumption index based on the clothing of people in an elevator, the system comprising:

[0011] Image acquisition module: including multiple high-resolution cameras installed on the top, side and bottom of the elevator to capture panoramic images inside the elevator and adapt to different lighting conditions and crowd density through autofocus and auto-exposure control algorithms;

[0012] Image restoration module: including a denoising unit, a restoration and enhancement unit, and a background segmentation unit, each of which dynamically adjusts processing parameters through an adaptive learning algorithm;

[0013] Image recognition module: used to identify the clothing grades in the image and classify them into different brands and grades;

[0014] Data integration module: used to clean, match and integrate the clothing grade data generated by the image recognition module with the membership data, payment data and social media data;

[0015] Consumption index calculation module: used to generate and dynamically adjust the consumption index based on user behavior analysis and biometric analysis;

[0016] Data security and privacy protection module: used to encrypt and store all collected and processed data and ensure the security of data transmission.

[0017] Preferably, the automatic focus and automatic exposure control algorithm is implemented by the following steps:

[0018] Set the initial exposure time and initial gain, and output the initial exposure setting;

[0019] Set the target brightness and control coefficient, and output the brightness target value;

[0020] Capture the current image and calculate the average brightness of the image, and output the current brightness value;

[0021] Calculate the error between the current brightness and the target brightness, adjust the exposure time and gain according to the error, and output the adjusted parameters;

[0022] Apply the new exposure time and gain to the camera, continue to adjust until the ideal image brightness is achieved, and output the optimized image brightness parameters.

[0023] Preferably, the image restoration module comprises:

[0024] De-noising unit: The deep learning denoising network DnCNN is used to denoise the image. The specific steps are as follows:

[0025] Collect a large number of noisy training images and corresponding clean images;

[0026] Construct a DnCNN network, including multiple convolutional layers, each followed by a ReLU activation function and a normalization operation after each convolutional layer;

[0027] Using mean square error as the loss function L(θ) is:

[0028]

[0029] Where N is the number of samples in the training set; y i The i-th noisy input image; f(y i ; θ) is the DnCNN model for the input image y i The processed output; x i is the denoised target image, used to calculate the loss; ||.|| 2 is the squared Euclidean norm, which is used to calculate the sum of squares of pixel-level differences between two images;

[0030] The stochastic gradient descent algorithm is used for training;

[0031] Input the noisy image y to the trained DnCNN model and output the denoised image

[0032]

[0033] Where y is the input image;

[0034] Restoration enhancement unit: Use multi-pixel feedback enhancement MSRCR technology to perform image enhancement processing. The specific steps are as follows:

[0035] Set the scale parameter F of the Gaussian filter i and color restoration parameters α, β;

[0036] For each color channel I of the image i Perform multi-scale Retinex processing:

[0037] R i (x,y)=logI i (x,y)-log[F i (x,y)*I i (x,y)]

[0038] In the formula, I i (x, y) is the input pixel value of the i-th color channel of the image, located at position (x, y); F i (x, y) is the Gaussian filter function for image enhancement processing, corresponding to the i-th color channel; R i (x, y) is the result of Retinex processing, which is used to remove the influence of illumination and obtain the reflection component of the image; log is a logarithmic function, which is used to compress the dynamic range of pixel values;

[0039] Apply the color restoration formula to calculate C i (x,y):

[0040] C i (x,y)=β[R i (x,y)+αlog(I i (x,y))]

[0041] In the formula, C i (x, y) is the image value after color restoration, which combines the Retinex processing result and the logarithmic value of the original image; α, β are weight parameters in the color restoration process, which control the degree of color restoration;

[0042] The Retinex outputs of different scales are divided into weights w s Fusion, get the enhanced image MSR i (x,y):

[0043]

[0044] Where: MSR i (x, y) is the image value after multi-scale Retinex processing, which combines the Retinex results at different scales; w s is the weight of the Retinex result at each scale, which controls the influence of the results of different scales in the final fused image; s is the weight of the Retinex result at each scale, which controls the influence of the results of different scales in the final fused image;

[0045] The background segmentation unit uses the deep learning segmentation network Mask R-CNN to achieve accurate separation of the image background and the target object. The specific steps are as follows:

[0046] Collect and annotate training images, including bounding boxes and segmentation masks of objects;

[0047] Construct the Mask R-CNN network and train it using the multi-task loss function L, which includes the classification loss L cls , regression loss L reg and the mask loss L mask , the formula is as follows:

[0048] L=L cls +L reg +L mask

[0049] Input the image to the trained Mask R-CNN model and output the bounding box and segmentation mask of the object;

[0050] Each unit dynamically adjusts processing parameters through an adaptive learning algorithm, and the steps are as follows:

[0051] The system analyzes the characteristics of the input image in real time and automatically adjusts the learning rate and convolution kernel of the denoising network, the filter scale and color restoration parameters of the image enhancement, and the loss function weight of the background segmentation model, thereby ensuring that each unit can achieve the best processing effect under different conditions.

[0052] Preferably, the EffiViT-Net model used by the image recognition module is constructed by the following steps:

[0053] Normalize the captured crowd images in the elevator and scale them to a standard size suitable for EfficientNet model input;

[0054] The preprocessed image is input into the pre-trained EfficientNet model to extract low-level and mid-level features, and a fixed-length feature vector is generated through global pooling;

[0055] Perform a linear transformation on the feature vector and embed it into the Transformer input space to adapt to the sequence modeling of VisionTransformers;

[0056] The embedded feature vector is input into the encoder module of Vision Transformers, and the multi-head self-attention mechanism is used to perform global feature modeling and generate global feature representation;

[0057] A multi-layer perceptron (MLP) classification head is added to the global feature representation to classify the clothing grade and brand in the image, and the model is trained end-to-end using the cross entropy loss function.

[0058] During the training process, the model performance is evaluated through the validation set, and the hyperparameters are dynamically adjusted, ultimately obtaining the EffiViT-Net model that performs best in the recognition task.

[0059] Preferably, the data integration module uses a graph neural network (GNN) method to map different types of data into the same embedding space, including the following data:

[0060] Member data: age, gender, purchase history;

[0061] Payment data: payment amount, payment method;

[0062] Social media data: likes, comments;

[0063] Fault data in elevators: floor dwell time and time period data.

[0064] Preferably, the consumption index calculation module generates a cumulative consumption index through behavioral analysis and biometric analysis, wherein the behavioral analysis includes shopping bag brand, movement trajectory and behavior residence time, and the biometric analysis uses image recognition and deep learning algorithms to infer age, gender, skin color based on captured facial features, skin color information and behavior patterns, and generates analysis results in combination with shopping frequency and preferences.

[0065] Preferably, the consumption index calculation module dynamically adjusts the consumption index by analyzing historical consumption data, current behavior data and real-time environmental data.

[0066] Preferably, the privacy protection module not only uses the AES-256 encryption algorithm for data encryption, but also combines multi-factor authentication and a regular key update strategy.

[0067] Preferably, the system further comprises an advertisement push module, and the advertisement push module pushes personalized advertisement content to a display device in the elevator in real time based on the consumption index generated by the consumption index calculation module.

[0068] An advertising promotion method for determining a consumption index based on the clothing of people in an elevator, the method comprising the following steps:

[0069] S1: Captures crowd images in elevators using multiple high-resolution cameras;

[0070] S2: Image restoration, denoising, enhancement and background segmentation are performed through the image restoration module;

[0071] S3: Use EffiViT-Net to identify the clothing grade of the image and generate corresponding clothing grade data;

[0072] S4: Integrate the clothing grade data into membership data, payment data and social media data, use the GNN method to map and integrate the data, and generate the input data set for consumption index calculation;

[0073] S5: Based on the data integration result of step S4, the consumption index is generated and dynamically adjusted through the consumption index calculation module in combination with behavior analysis and biometric analysis;

[0074] S6: Ensure compliance with data security and privacy protection requirements throughout the data processing process, including the use of AES-256 encryption algorithm and SSL / TLS protocol, combined with multi-factor authentication and regular key updates.

[0075] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention realizes accurate identification of the dress level of people in the elevator through image recognition and data integration technology, and combines member data, payment data and social media data to generate a dynamically adjusted consumption index. This method greatly improves the accuracy of advertising delivery, makes the advertising content closer to the actual needs of users, and thus improves the click-through rate and user participation of advertisements. By adopting efficient image processing and recognition algorithms, the analysis and data integration of user images can be completed in a short time, ensuring that the content of advertising push can timely reflect the current status and consumption tendency of users. By introducing graph neural network (GNN) technology, the present invention can effectively integrate and process different types of data (such as clothing data, member data, payment data, etc.) to generate a comprehensive user portrait. The present invention adopts advanced privacy protection measures in the data processing process, including AES-256 encryption algorithm, SSL / TLS protocol, multi-factor authentication and regular key update strategy, to ensure that user data is strictly protected during collection, storage and transmission, prevent data leakage and abuse, and improve the security of the system and user trust. The advertising promotion system can work effectively in various complex elevator environments, including different light conditions and crowd density. Through the adaptive learning algorithm, the system can dynamically adjust the processing parameters to ensure the quality of image acquisition and processing, thereby achieving efficient advertising push in various environments. Experimental results show that the system of the present invention is significantly superior to the traditional system in terms of advertising click-through rate, user dwell time and advertising content relevance. Through accurate user analysis and personalized advertising push, the present invention significantly improves the actual effect of advertising and brings higher return on investment to advertisers. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0077] Figure 1 It is a structural schematic diagram of an embodiment of the present invention;

[0078] Figure 2 is a detailed architecture diagram of an embodiment of the present invention;

[0079] Figure 3 A schematic diagram of a processing flow of an image restoration module according to an embodiment of the present invention;

[0080] Figure 4 A schematic diagram of the working principle of an image recognition module according to an embodiment of the present invention;

[0081] Figure 5 A schematic diagram of the data integration and consumption index calculation process of an embodiment of the present invention;

[0082] Figure 6 A schematic diagram of a data security and privacy protection module according to an embodiment of the present invention;

[0083] Figure 7 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0085] Example 1

[0086] like Figure 1 , Figure 2 FIG. 1 is an embodiment of the present invention, which provides an advertising promotion system for determining a consumption index based on the clothing of people in an elevator, including:

[0087] Image acquisition module: including multiple high-resolution cameras installed on the top, side and bottom of the elevator to capture panoramic images inside the elevator and adapt to different lighting conditions and crowd density through autofocus and auto-exposure control algorithms;

[0088] Furthermore, the automatic focus and automatic exposure control algorithm is implemented by the following steps:

[0089] Set the initial exposure time and initial gain, and output the initial exposure setting;

[0090] Set the target brightness and control coefficient, and output the brightness target value;

[0091] Capture the current image and calculate the average brightness of the image, and output the current brightness value;

[0092] Calculate the error between the current brightness and the target brightness, adjust the exposure time and gain according to the error, and output the adjusted parameters;

[0093] Apply the new exposure time and gain to the camera, continue to adjust until the ideal image brightness is achieved, and output the optimized image brightness parameters.

[0094] Specifically, camera configuration: use multiple high-resolution cameras, installed on the top, side and bottom of the elevator to capture images from different angles inside the elevator.

[0095] Autofocus and autoexposure: The camera is equipped with autofocus and autoexposure control algorithms, which can dynamically adjust the camera parameters according to the light changes and crowd density in the elevator, ensuring clear and high-quality images in different environments.

[0096] Adaptability: Whether it is daytime, nighttime or during peak traffic hours, the system can adapt to changes in light and space to ensure image stability and clarity.

[0097] Further, such as Figure 3 As shown, the image restoration module includes a denoising unit, a restoration enhancement unit, and a background segmentation unit, and each unit dynamically adjusts processing parameters through an adaptive learning algorithm;

[0098] Wherein, the image restoration module comprises:

[0099] De-noising unit: The deep learning denoising network DnCNN is used to denoise the image. The specific steps are as follows:

[0100] Collect a large number of noisy training images and corresponding clean images;

[0101] Construct a DnCNN network, including multiple convolutional layers, each followed by a ReLU activation function and a normalization operation after each convolutional layer;

[0102] Using mean square error as the loss function L(θ) is:

[0103]

[0104] Where N is the number of samples in the training set; y i The i-th noisy input image; f(y i ; θ) is the DnCNN model for the input image y i The processed output; x i is the denoised target image, used to calculate the loss; ||.|| 2 is the squared Euclidean norm, which is used to calculate the sum of squares of pixel-level differences between two images;

[0105] The stochastic gradient descent algorithm is used for training;

[0106] Input the noisy image y to the trained DnCNN model and output the denoised image

[0107]

[0108] Where y is the input image;

[0109] Restoration enhancement unit: Use multi-pixel feedback enhancement MSRCR technology to perform image enhancement processing. The specific steps are as follows:

[0110] Set the scale parameter F of the Gaussian filter i and color restoration parameters α, β;

[0111] Scale parameter F i , represents the size of Gaussian filters of different scales in the multi-scale Retinex algorithm, which is usually set to {15, 80, 250} to cope with different levels of image detail. The color restoration parameters α and β control the amplitude and strength of color restoration respectively, and the typical value is {125, 46}.

[0112] For each color channel I of the image i Perform multi-scale Retinex processing:

[0113] R i (x,y)=logI i (x,y)-log[F i (x,y)*I i (x,y)]

[0114] In the formula, I i (x, y) is the input pixel value of the i-th color channel of the image, located at position (x, y); F i (x, y) is the Gaussian filter function for image enhancement processing, corresponding to the i-th color channel; R i (x, y) is the result of Retinex processing, which is used to remove the influence of illumination and obtain the reflection component of the image; log is a logarithmic function, which is used to compress the dynamic range of pixel values;

[0115] Apply the color restoration formula to calculate C i (x,y):

[0116] C i (x,y)=β[R i (x,y)+αlog(I i (x,y))]

[0117] In the formula, C i (x, y) is the image value after color restoration, which combines the Retinex processing result and the logarithmic value of the original image; α, β are weight parameters in the color restoration process, which control the degree of color restoration;

[0118] The Retinex outputs of different scales are divided into weights w s Fusion, get the enhanced image MSR i (x,y):

[0119]

[0120] Where: MSR i (x, y) is the image value after multi-scale Retinex processing, which combines the Retinex results at different scales; w s is the weight of the Retinex result at each scale, which controls the influence of the results of different scales in the final fused image; s is the weight of the Retinex result at each scale, which controls the influence of the results of different scales in the final fused image;

[0121] The background segmentation unit uses the deep learning segmentation network Mask R-CNN to achieve accurate separation of the image background and the target object. The specific steps are as follows:

[0122] Collect and annotate training images, including bounding boxes and segmentation masks of objects;

[0123] Construct the Mask R-CNN network and train it using the multi-task loss function L, which includes the classification loss L cls , regression loss L reg and the mask loss L mask , the formula is as follows:

[0124] L=L cls +L reg +L mask

[0125] Input the image to the trained Mask R-CNN model and output the bounding box and segmentation mask of the object;

[0126] Each unit dynamically adjusts processing parameters through an adaptive learning algorithm, and the steps are as follows:

[0127] The system analyzes the characteristics of the input image in real time and automatically adjusts the learning rate and convolution kernel of the denoising network, the filter scale and color restoration parameters of the image enhancement, and the loss function weight of the background segmentation model, thereby ensuring that each unit can achieve the best processing effect under different conditions.

[0128] Specifically, the denoising unit uses a deep learning denoising network DnCNN to process noisy images. The trained convolutional neural network removes noise points in the image and improves the clarity of the image.

[0129] Restoration and enhancement unit: Uses multi-pixel feedback enhancement (MSRCR) technology to enhance the image. This technology uses a multi-scale Retinex algorithm to remove the effects of lighting, restore the true color and contrast of the image, and ensure the consistency of image quality under different lighting conditions.

[0130] Background segmentation unit: Use the Mask R-CNN network to perform background segmentation on the image and accurately separate the person and background in the image. This process can help the system better identify the target object (such as an individual in a crowd) and ignore irrelevant background information.

[0131] Specifically, an adaptive learning algorithm is an algorithm that can dynamically adjust system parameters according to the characteristics of input data, aiming to optimize the processing effect so that the system can maintain optimal performance in various environments.

[0132] Implementation details of the adaptive learning algorithm

[0133] Input characteristic analysis: The system first extracts the key features of the input image, such as brightness, noise level, etc., and performs preliminary analysis to generate characteristic indicators.

[0134] Parameter adjustment: Based on the input characteristics, the system automatically initializes the processing parameters of each module (such as convolution kernel size, learning rate, etc.), and adjusts these parameters in real time according to feedback during the processing process to ensure the best output effect.

[0135] Feedback optimization: By evaluating the quality of the processing results (such as through indicators such as mean square error MSE), the system automatically adjusts parameters to optimize the processing effect. This feedback loop continues, allowing the system to continuously adapt to changes in different input data.

[0136] Algorithm logic and application scenarios

[0137] Image restoration module:

[0138] Denoising: According to the image noise level, the parameters of the denoising network are adaptively adjusted to ensure clarity.

[0139] Enhancement: Dynamically adjust image enhancement parameters to adapt to different lighting conditions.

[0140] Segmentation: According to the background complexity, adjust the parameters of the segmentation network to improve the segmentation accuracy.

[0141] Image recognition module: According to the complexity of the image, the hyperparameters of the EffiViT-Net model are adjusted to ensure recognition accuracy.

[0142] Data integration module: automatically adjusts the embedding dimension of graph neural networks to optimize the fusion effect of multi-source data.

[0143] Application Advantages

[0144] Efficient and flexible: Automatically adjust parameters, reduce human intervention, adapt to changing environments, and ensure optimal performance.

[0145] Continuous optimization: Through feedback loops, the system continuously learns and improves to ensure stability and efficiency.

[0146] Further, such as Figure 4 As shown, the image recognition module is used to identify the clothing grades in the image and classify them into different brands and grades;

[0147] The EffiViT-Net model used in the image recognition module is constructed by the following steps:

[0148] Normalize the captured crowd images in the elevator and scale them to a standard size suitable for EfficientNet model input;

[0149] The preprocessed image is input into the pre-trained EfficientNet model to extract low-level and mid-level features, and a fixed-length feature vector is generated through global pooling;

[0150] Global pooling uses the average pooling (GAP) method, taking the average value of each feature channel as the output feature after pooling, thereby generating a feature vector of fixed length, ensuring that the scale of the feature vector is independent of the input image size.

[0151] Perform a linear transformation on the feature vector and embed it into the Transformer input space to adapt to the sequence modeling of VisionTransformers;

[0152] The embedded feature vector is input into the encoder module of Vision Transformers, and the multi-head self-attention mechanism is used to perform global feature modeling and generate global feature representation;

[0153] A multi-layer perceptron (MLP) classification head is added to the global feature representation to classify the clothing grade and brand in the image, and the model is trained end-to-end using the cross entropy loss function.

[0154] During the training process, the model performance is evaluated through the validation set, and the hyperparameters are dynamically adjusted, ultimately obtaining the EffiViT-Net model that performs best in the recognition task.

[0155] Specifically, the EffiViT-Net model: This module uses the hybrid neural network model EffiViT-Net based on EfficientNet and VisionTransformers, which is specifically used to identify and classify people's clothing grades. The model can distinguish clothing of different brands and grades by globally modeling image features.

[0156] Classification and labeling: The system will classify the identified clothing grade data and label it into different brands and grades, providing basic data for subsequent data integration and consumption index calculation.

[0157] Further, such as Figure 5 As shown, the data integration module is used to clean, match and integrate the clothing grade data generated by the image recognition module with the member data, payment data and social media data;

[0158] The data integration module uses the graph neural network (GNN) method to map different types of data into the same embedding space, including the following data:

[0159] Member data: age, gender, purchase history;

[0160] Payment data: payment amount, payment method;

[0161] Social media data: likes, comments;

[0162] Fault data in elevators: floor dwell time and time period data.

[0163] Specifically, multi-source data integration: This module cleans, matches and integrates the dress grade data with membership data, payment data and social media data through graph neural network (GNN). GNN can map different types of data into a unified embedding space to ensure the accuracy and consistency of data integration.

[0164] Data cleaning and matching: The system pre-processes the data, removes invalid or duplicate data, and performs matching to ensure that the integrated data is of high quality and reliability.

[0165] Further, the consumption index calculation module is used to generate and dynamically adjust the consumption index based on user behavior analysis and biometric analysis;

[0166] The generation of the consumption index is achieved through the following steps: First, the system collects the user's behavioral data and biometric data, and then integrates the data through the graph neural network to generate the initial consumption index. After that, the system dynamically adjusts the consumption index based on the real-time data to reflect the user's current status.

[0167] For example, when a user stays in an elevator for a long time and the brand of the shopping bag is a high-end brand, the consumption index will increase accordingly and trigger the push of advertising content related to high-end consumption.

[0168] Among them, the consumption index calculation module generates a cumulative consumption index through behavioral analysis and biometric analysis. The behavioral analysis includes shopping bag brand, movement trajectory and behavioral residence time. The biometric analysis uses image recognition and deep learning algorithms to infer age, gender, skin color based on captured facial features, skin color information and behavioral patterns, and generates analysis results in combination with shopping frequency and preferences.

[0169] Furthermore, the consumption index calculation module dynamically adjusts the consumption index by analyzing historical consumption data, current behavior data and real-time environmental data.

[0170] Specifically, through the long short-term memory network (LSTM), the system analyzes the user's behavioral data (such as shopping bag brand, movement trajectory, residence time, etc.) to identify the user's consumption behavior pattern.

[0171] Biometric analysis: Use deep learning algorithms to analyze users’ biometric characteristics (such as age, gender, skin color, etc.) and combine them with behavioral data to generate a user consumption index.

[0172] Dynamic adjustment: The system dynamically adjusts the consumption index based on real-time data to ensure that the advertising content can promptly reflect the user's current status and needs.

[0173] Further, such as Figure 6 As shown, the data security and privacy protection module is used to encrypt and store all collected and processed data and ensure the security of data transmission.

[0174] Among them, the privacy protection module not only uses the AES-256 encryption algorithm for data encryption, but also combines multi-factor authentication and regular key update strategies.

[0175] Specifically, data encryption: the system uses the AES-256 encryption algorithm to encrypt and store all collected and processed data to prevent data leakage.

[0176] Transmission security: Use SSL / TLS protocol to encrypt data transmission to ensure that it is not stolen or tampered with during the data transmission process.

[0177] Privacy protection strategy: The system also introduces multi-factor authentication and regular key update strategies to further enhance data security and protect user privacy.

[0178] Furthermore, the system further includes an advertisement push module, which pushes personalized advertisement content to a display device in the elevator in real time based on the consumption index generated by the consumption index calculation module.

[0179] Example 2

[0180] like Figure 7 As shown in FIG. 1 , another embodiment of the present invention is provided. The embodiment provides an advertising promotion method for determining a consumption index based on the clothing of people in an elevator. The method comprises the following steps:

[0181] S1: Captures crowd images in elevators using multiple high-resolution cameras;

[0182] S2: Image restoration, denoising, enhancement and background segmentation are performed through the image restoration module;

[0183] S3: Use EffiViT-Net to identify the clothing grade of the image and generate corresponding clothing grade data;

[0184] S4: Integrate the clothing grade data into membership data, payment data and social media data, use the GNN method to map and integrate the data, and generate the input data set for consumption index calculation;

[0185] S5: Based on the data integration result of step S4, the consumption index is generated and dynamically adjusted through the consumption index calculation module in combination with behavior analysis and biometric analysis;

[0186] S6: Ensure compliance with data security and privacy protection requirements throughout the data processing process, including the use of AES-256 encryption algorithm and SSL / TLS protocol, combined with multi-factor authentication and regular key updates.

[0187] Multi-Factor Authentication (MFA) is an identity authentication method that strengthens information security. It improves security by requiring users to provide multiple independent authentication factors to verify their identity. Generally, multi-factor authentication includes the following factors:

[0188] Knowledge factor: Something the user knows, such as a password or PIN.

[0189] Ownership factor: Something the user owns, such as a smart card, mobile phone, or hardware token.

[0190] Biometric factors: A biological characteristic of the user, such as a fingerprint, facial recognition, or iris scan.

[0191] The advantage of multi-factor authentication is that even if one of the authentication factors is compromised or cracked, the attacker still needs to break through the other factors to gain access. This greatly enhances the overall security of the system and prevents unauthorized access.

[0192] Example 3

[0193] Another embodiment of the present invention demonstrates the superiority and innovation of the system of the present invention in practical applications.

[0194] The experiments verified that the advertising promotion system of the present invention for determining the consumption index based on the clothing of people in the elevator has improved effects in practical applications compared with the traditional advertising system, especially in terms of advertising click-through rate and user participation.

[0195] Experimental methods:

[0196] Three commercial buildings with high traffic volume were selected as experimental sites, and the system of the present invention was installed in the elevator of each building. The experiment was set up for comparison between two groups: the experimental group used the system of the present invention for advertising push, and the control group used the traditional advertising playback system. The experiment lasted for 30 days, and the following data were mainly counted: Ad click-through rate (CTR, Click-Through Rate)

[0197] User engagement (time spent in the elevator)

[0198] Experimental results of ad content relevance score (based on user feedback):

[0199] The experimental results show that the experimental group is significantly better than the control group in all indicators. The specific data are as follows:

[0200] project Experimental group (using the system of the present invention) Control group (traditional advertising system) Ad Click-Through Rate (CTR) 12.80％ 5.30％ User engagement (average dwell time / second) 35.6 28.2 Relevance score of the ad content (1-5 points) 4.6 3.1

[0201] Analysis and conclusion:

[0202] Advertisement click-through rate: The ad click-through rate of the experimental group reached 12.8%, much higher than the 5.3% of the control group. This shows that the system of the present invention can more accurately push advertisements that match the needs of users by analyzing the user's clothing level and consumption index, thereby significantly improving the attractiveness and click-through rate of advertisements.

[0203] User engagement: The average stay time of users in the experimental group was 35.6 seconds, which was significantly higher than the 28.2 seconds in the control group. This shows that the system of the present invention can push more eye-catching advertising content, prolong the user's attention time, and thus increase the exposure effect of the advertisement.

[0204] Relevance of advertising content: Based on the relevance score of user feedback, the average score of the advertising content in the experimental group was 4.6 points, which was much higher than the 3.1 points in the control group. This shows that the system of the present invention can better understand and match user needs and improve the fit between advertising content and user interests.

[0205] Summarize:

[0206] The experimental data clearly show that the system of the present invention is significantly superior to the traditional system in terms of advertising click-through rate, user engagement and relevance of advertising content. By combining the clothing data of people in the elevator with other user behavior data, the system of the present invention effectively improves the accuracy and effectiveness of advertising delivery, showing strong innovation and application prospects.

[0207] In summary, the present invention proposes for the first time to collect crowd images in an elevator, identify the user's dress grade using image recognition technology, and calculate the user's consumption index in combination with other behavioral and biometric data. This method provides a new way to build a user portrait, which can more accurately reflect the user's consumption ability and preference. By integrating image recognition data with multi-source data such as member information, payment records, and social media data, the present invention constructs a comprehensive user portrait. The use of graph neural network (GNN) technology enables these data to be efficiently analyzed in the same embedding space, enhancing the accuracy and personalization of advertising push. The present invention introduces an advertising push mechanism based on consumption index, which can dynamically adjust advertising content according to real-time data to ensure that advertising can timely reflect the user's current status and needs. This real-time innovation significantly improves the relevance of advertising and user participation. The present invention pays attention to user privacy protection in system design, and adopts multi-level security measures, such as AES-256 encryption algorithm, SSL / TLS protocol, multi-factor authentication, etc., to ensure the security of user data during collection, processing, transmission and storage, and improve the legitimacy of the system and user trust.

[0208] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0209] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0210] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An advertising promotion system for determining a consumption index based on the clothing of people in an elevator, characterized in that: The system comprises: Image acquisition module: including multiple high-resolution cameras installed on the top, side and bottom of the elevator to capture panoramic images inside the elevator and adapt to different lighting conditions and crowd density through autofocus and auto-exposure control algorithms; Image restoration module: including a denoising unit, a restoration and enhancement unit, and a background segmentation unit, each of which dynamically adjusts processing parameters through an adaptive learning algorithm; Image recognition module: used to identify the clothing grade in the image and classify it into different brands and grades; Data integration module: used to clean, match and integrate the clothing grade data generated by the image recognition module with member data, payment data and social media data; Consumption index calculation module: used to generate and dynamically adjust the consumption index based on user behavior analysis and biometric analysis; Data security and privacy protection module: used to encrypt and store all collected and processed data and ensure the security of data transmission.

2. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The auto-focus and auto-exposure control algorithms are implemented by the following steps: Set the initial exposure time and initial gain, and output the initial exposure setting; Set the target brightness and control coefficient, and output the brightness target value; Capture the current image and calculate the average brightness of the image, and output the current brightness value; Calculate the error between the current brightness and the target brightness, adjust the exposure time and gain according to the error, and output the adjusted parameters; Apply the new exposure time and gain to the camera, continue to adjust until the ideal image brightness is achieved, and output the optimized image brightness parameters.

3. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The image restoration module comprises: The denoising unit: uses a deep learning denoising network DnCNN to denoise the image, and the specific steps are as follows: Collect a large number of noisy training images and corresponding clean images; Construct a DnCNN network, including multiple convolutional layers, each followed by a ReLU activation function and a normalization operation after each convolutional layer; Using mean square error as the loss function L(θ) is: Where N is the number of samples in the training set; y i The i-th noisy input image; f(y i ; θ) is the DnCNN model for the input image y i The processed output; x i is the denoised target image, used to calculate the loss; ||.|| 2 is the squared Euclidean norm, which is used to calculate the sum of squares of pixel-level differences between two images; the stochastic gradient descent algorithm is used for training; Input the noisy image y to the trained DnCNN model and output the denoised image Where y is the input image; The restoration enhancement unit: uses multi-pixel feedback enhancement MSRCR technology to perform image enhancement processing, and the specific steps are as follows: Set the scale parameter F of the Gaussian filter i and color restoration parameters α, β; For each color channel I of the image i Perform multi-scale Retinex processing: R i (x,y)=logI i (x,y)-log[F i (x,y)*I i (x,y)] In the formula, I i (x, y) is the input pixel value of the i-th color channel of the image, located at position (x, y); F i (x, y) is the Gaussian filter function for image enhancement processing, corresponding to the i-th color channel; R i (x, y) is the result of Retinex processing, which is used to remove the influence of illumination and obtain the reflection component of the image; log is a logarithmic function, which is used to compress the dynamic range of pixel values; Apply the color restoration formula to calculate C i (x,y): C i (x,y)=β[R i (x,y)+αlog(I i (x,y))] In the formula, C i (x, y) is the image value after color restoration, which combines the Retinex processing result and the logarithmic value of the original image; α, β are weight parameters in the color restoration process, which control the degree of color restoration; The Retinex outputs of different scales are divided into weights w s Fusion, get the enhanced image MSR i (x,y): Where: MSR i (x, y) is the image value after multi-scale Retinex processing; w s is the weight of the Retinex result at each scale, which controls the influence of the results of different scales in the final fused image; s is the weight of the Retinex result at each scale, which controls the influence of the results of different scales in the final fused image; The background segmentation unit uses the deep learning segmentation network Mask R-CNN to achieve accurate separation of the image background and the target object. The specific steps are as follows: Collect and annotate training images, including bounding boxes and segmentation masks of objects; Construct the Mask R-CNN network and train it using the multi-task loss function L, which includes the classification loss L cls , regression loss L reg and the mask loss L mask , the formula is as follows: L=L cls +L reg +L mask Input the image to the trained Mask R-CNN model and output the bounding box and segmentation mask of the object; Each unit dynamically adjusts processing parameters through an adaptive learning algorithm. The system analyzes and automatically adjusts the learning rate and convolution kernel of the denoising network, the filter scale and color restoration parameters of the image enhancement, and the loss function weight of the background segmentation model in real time according to the characteristics of the input image.

4. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The EffiViT-Net model used in the image recognition module is constructed by the following steps: normalizing the captured images of people in the elevator and scaling them to a standard size suitable for EfficientNet model input; The preprocessed image is input into the pre-trained EfficientNet model to extract low-level and mid-level features, and a fixed-length feature vector is generated through global pooling; Perform a linear transformation on the feature vector and embed it into the Transformer input space to adapt to the sequence modeling of VisionTransformers; The embedded feature vector is input into the encoder module of Vision Transformers, and the multi-head self-attention mechanism is used to perform global feature modeling and generate global feature representation; A multi-layer perceptron (MLP) classification head is added to the global feature representation to classify the clothing grade and brand in the image, and the model is trained end-to-end using the cross entropy loss function. During the training process, the model performance is evaluated through the validation set, and the hyperparameters are dynamically adjusted, ultimately obtaining the EffiViT-Net model that performs best in the recognition task.

5. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The data integration module uses the graph neural network (GNN) method to map different types of data into the same embedding space, including the following data: Member data includes: age, gender, purchase history; Payment data includes: payment amount and payment method; Social media data includes: likes, comments; The fault data in the elevator includes: floor dwell time and time period data.

6. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The consumption index calculation module generates a cumulative consumption index through behavioral analysis and biometric analysis. The behavioral analysis includes shopping bag brand, movement trajectory and behavior residence time. The biometric analysis uses image recognition and deep learning algorithms to infer age, gender, skin color based on captured facial features, skin color information and behavior patterns, and generates analysis results in combination with shopping frequency and preferences.

7. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The consumption index calculation module dynamically adjusts the consumption index by analyzing historical consumption data, current behavior data and real-time environment data.

8. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The privacy protection module not only uses the AES-256 encryption algorithm for data encryption, but also combines multi-factor authentication and regular key update strategies.

9. The advertising promotion system for determining the consumption index based on the clothing of people in the elevator according to claim 1 is characterized in that: The system further includes an advertisement push module, which pushes personalized advertisement content to a display device in the elevator in real time based on the consumption index generated by the consumption index calculation module.

10. An advertising promotion method for determining a consumption index based on the clothing of people in an elevator according to any one of claims 1 to 9, characterized in that: The method comprises the following steps: S1: Captures crowd images in elevators using multiple high-resolution cameras; S2: Image restoration, denoising, enhancement and background segmentation are performed through the image restoration module; S3: Use EffiViT-Net to identify the clothing grade of the image and generate corresponding clothing grade data; S4: Integrate the clothing grade data into membership data, payment data and social media data, use the GNN method to map and integrate the data, and generate the input data set for consumption index calculation; S5: Based on the data integration result of step S4, the consumption index is generated and dynamically adjusted through the consumption index calculation module in combination with behavior analysis and biometric analysis; S6: Ensure compliance with data security and privacy protection requirements throughout the data processing process, including the use of AES-256 encryption algorithm and SSL / TLS protocol, combined with multi-factor authentication and regular key updates.

Images