Vehicle side window advertisement display management system and method

Abstract

The invention discloses a traffic tool side window advertisement display management system and method, particularly relates to the field of intelligent bus advertisement systems, is used for solving the problems of accurate matching and seamless display of advertisements in a dynamic driving situation, and is used for accurately extracting vehicle driving state data through a multi-source data fusion and Kalman filtering technology. And carrying out feature analysis by using a deep learning model, and generating a driving situation label in real time. Based on the labels, the system intelligently screens out advertisements highly matched with the current situation, analyzes duration and robustness in combination with historical driving data, and optimizes an advertisement selection strategy by using a reinforcement learning algorithm. And finally, through a pre-caching mechanism and a seamless switching algorithm, it is ensured that the optimal advertisement content can be dynamically loaded and stably displayed on a bus side window display, and continuity and fluency of advertisement playing are achieved. The correlation and efficiency of advertisement putting are remarkably improved, and the riding experience of passengers and the advertisement communication effect are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent public transportation advertising systems, and more specifically, to a vehicle side window advertising display management system and method. Background Art

[0002] With the popularization of public transportation and the development of information technology, buses, subways, high-speed railways, etc. have gradually become important tools for passengers to travel. In order to improve the riding experience, the bus side window advertising system came into being and became a new way of information dissemination. Side window advertising not only brings commercial value to buses, but also provides convenient information services for passengers. Through the on-board display screen, buses can display dynamic information such as station information, traffic conditions, weather warnings, etc. to passengers. At the same time, the development of intelligent technology enables the driving status of vehicles and changes in the in-vehicle environment to be perceived in real time, providing new opportunities for the precise push of advertising content. However, most of the existing bus side window advertising systems rely on a fixed advertising library, the advertising content is preset and does not change with the driving status, and there is a lack of real-time linkage, resulting in a low degree of fit between advertising and driving situations, and a decrease in advertising effect and passenger attention. Therefore, intelligently adjusting the side window advertising content based on the real-time driving status of the bus has become a key technical requirement for improving advertising efficiency and enhancing passenger experience.

[0003] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a transportation side window advertising display management system and method, which accurately captures vehicle driving status data by integrating multi-source data fusion and Kalman filtering technology, and extracts features in combination with a deep learning model to achieve real-time driving scenario recognition; based on the generated scenario tags, intelligently screens candidate advertisements with highly semantic matches from the labeled advertising content library, and analyzes the duration and robustness of historical driving data at the same time, optimizes the advertising selection strategy through a reinforcement learning algorithm to ensure that the selected advertisement is the most appropriate in the current scenario; the selected optimal advertisement is then dynamically loaded and stably displayed on the bus side window display through an advanced pre-caching mechanism and a seamless switching algorithm to ensure the continuity and smoothness of the advertising playback; accurate data processing, intelligent scenario perception and optimized advertising selection, as well as efficient dynamic loading and display are achieved, forming a highly coordinated and adaptive advertising delivery system; not only improves the relevance and display effect of the advertisement, but also ensures the stable presentation of the advertisement in a dynamic driving environment, significantly enhances the visual experience of passengers and the effectiveness of advertising dissemination, so as to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: S1, based on multi-source data fusion technology, extracts the comprehensive data stream of the vehicle and ensures the accuracy of the driving status data through Kalman filtering and denoising.

[0006] S2, extracts features from driving status data through a deep learning model, establishes a driving situation recognition model based on acceleration and steering features, and generates real-time situation labels for the vehicle.

[0007] S3, based on driving context labels, selects candidate advertisements with multi-dimensional semantic matching from the labeled advertising content library, combines the duration robustness of the driving state, optimizes the advertisement selection strategy through the reinforcement learning model, realizes context-driven advertisement selection, and obtains the optimal advertisement.

[0008] S4 uses a pre-caching mechanism and seamless switching algorithm to dynamically load and display the optimal advertising content on the bus side window display in real time.

[0009] In a preferred embodiment, step S1 includes the following contents: S1.1, first integrate the data from different sensors to form a comprehensive data stream that describes the vehicle's driving status; for each data source, first perform standardization processing to eliminate the impact between different dimensions and magnitudes.

[0010] S1.2 uses a multi-dimensional adaptive fusion matrix to dynamically adjust the weight of each data source according to the changes in the real-time environment and vehicle status, including the following steps: First, the spatial positioning data, attitude angle data and acceleration data are standardized.

[0011] Then, set initial weights for each data source.

[0012] Next, the correlation between the data sources is calculated through the state correlation function, and the evaluation of the state correlation function is based on the mutual information in information theory.

[0013] Finally, the correlation between the data sources is quantified by calculating the mutual information, and the weight matrix is ​​updated according to the correlation between the data sources.

[0014] S1.3, using the updated weight matrix to perform weighted combination on the standardized data to obtain a comprehensive driving state vector reflecting the current driving state of the vehicle.

[0015] In a preferred embodiment, step S1 further includes the following contents: S1.4, using nonlinear Kalman filtering to optimize and filter the fused driving state vector, including the following two stages: Prediction stage: Based on the state at the previous moment, use the state transfer matrix Make predictions: ;in, is the predicted vehicle driving state, is the state transfer matrix, describing the state dynamics of the vehicle.

[0016] Update phase: Combine the current observation data , optimize the prediction results by updating the formula: ; ;in, is the Kalman gain, is the forecast error covariance, is the observation matrix, is the observation noise covariance, For observation data.

[0017] In a preferred embodiment, step S2 includes the following contents: S2.1, based on the driving state vector, a convolutional neural network is used to extract short-term feature fluctuations and capture local features in the vehicle driving state.

[0018] S2.2, each feature in the driving state vector has temporal correlation, and a long short-term memory network is used to mine the temporal features of the driving state.

[0019] S2.3, after obtaining the fused feature vector, the corresponding features are mapped to the high-dimensional situational space, and the feature relationship is captured through nonlinear mapping, and finally a real-time driving situation label is generated.

[0020] In a preferred embodiment, step S3 includes the following contents: S3.1, in the labeled advertising content library, retrieve all advertisements related to the current driving situation label to form a preliminary advertising set; use the deep semantic embedding model to map the driving situation label and advertising label vector to a high-dimensional semantic space to obtain the semantic embedding vector and ; Similarity calculation function: ;in, represents the vector inner product, represents the norm of a vector; Indicates The semantic similarity between each advertisement and the current context label is calculated; the candidate advertisements are sorted in descending order according to the similarity value, and the advertisements with similarity higher than the similarity threshold are selected to form a candidate advertisement set.

[0021] In a preferred embodiment, step S3 further includes the following contents: S3.2, buses travel on fixed routes, and the duration of driving status in different sections is regular; the duration of advertisements needs to match the estimated duration of the vehicle in the current driving situation to ensure that the advertisements can be played completely without affecting the passenger experience; collect historical driving data of buses on the same route, the same section, and the same driving situation label to obtain the duration sequence ; Use information entropy to evaluate the robustness of duration; If the information entropy is less than or equal to the corresponding threshold, the duration is considered to be sufficiently robust; Use kernel density estimation to obtain the probability density function of duration; Determine the expected duration The peak value of the probability density function corresponds to Value, i.e. duration of maximum likelihood estimation.

[0022] In a preferred embodiment, step S3 further includes the following contents: S3.3, for each advertisement in the candidate advertisement set, the corresponding playing time is ; Define the ad duration matching function: ; For duration series The time scale parameter is calculated based on the mutual information of the data; the function Defined as: ; Ensure that the duration of the advertisement does not exceed the expected duration; Sort the advertisements in descending order according to the advertisement duration matching degree, select the advertisements with matching degree higher than the corresponding threshold, and form the final candidate advertisement set.

[0023] In a preferred embodiment, step S3 further includes the following contents: S3.4, use reinforcement learning to drive advertising selection optimization; select the best advertisement from the final candidate advertisement set for delivery; through the reinforcement learning model, in each state, select the action with the maximum Q value according to the Q value table, that is, select the advertisement in the current candidate advertisement set that best matches the state and mark it as the best advertisement.

[0024] A vehicle side window advertising display management system comprises: a data fusion module, a situation recognition module, an advertising selection module and a dynamic loading module.

[0025] Data fusion module: Based on multi-source data fusion technology, it extracts the comprehensive data stream of the vehicle, and uses Kalman filtering to denoise the data to ensure the accuracy of the driving status data. The processed data is then passed to the situation recognition module.

[0026] Context recognition module: Use the deep learning model to extract features from the fused driving status data, combine the acceleration and steering features to establish a driving context recognition model, generate the real-time context label of the vehicle, and pass the label to the advertising selection module.

[0027] Advertisement selection module: Based on real-time context labels, candidate advertisements with multi-dimensional semantic matching are screened out from the labeled advertisement content library. Combined with the duration robustness of the driving state, the advertisement selection strategy is optimized through the reinforcement learning model, and finally the optimal advertisement is obtained and passed to the dynamic loading module.

[0028] Dynamic loading module: adopts pre-caching mechanism and seamless switching algorithm to dynamically load and accurately display the optimal advertising content on the bus side window display in real time, ensuring the continuity and smoothness of advertising playback.

[0029] Technical effects and advantages of a vehicle side window advertising display management system and method of the present invention: The present invention integrates multi-source data fusion technology and deep learning models, extracts and processes vehicle driving status data, builds a real-time driving scenario recognition system, and generates dynamic scenario labels. Based on these scenario labels, it is possible to intelligently screen out advertisements that highly match the current driving status in the labeled advertising content library, and analyze the duration and robustness of the driving status in combination with historical driving data, and continuously optimize the advertising selection strategy through reinforcement learning algorithms to achieve scenario-driven precision advertising delivery. Finally, a pre-caching mechanism and a seamless switching algorithm are used to ensure that the optimal advertising content can be dynamically loaded in real time and stably displayed on the bus side window display, ensuring the continuity and smoothness of advertising playback. Not only does it improve the relevance and efficiency of advertising delivery, but it also achieves a high degree of matching between advertising content and driving scenarios through intelligent data processing and optimization algorithms, significantly enhancing the visual experience of passengers and the advertising communication effect, reflecting the technical innovation and practical value of the system in the field of intelligent advertising delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention is a flow chart of a method for managing vehicle side window advertising display.

[0031] Figure 2 The present invention is a schematic structural diagram of a vehicle side window advertising display management system. DETAILED DESCRIPTION

[0032] 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.

[0033] Embodiment 1: Figure 1 The present invention provides a method for managing the display of advertisements on the side windows of vehicles, comprising: S1, based on multi-source data fusion technology, extracts the comprehensive data stream of the vehicle and ensures the accuracy of the driving status data through Kalman filtering and denoising.

[0034] S2, extracts features from driving status data through a deep learning model, establishes a driving situation recognition model based on acceleration and steering features, and generates real-time situation labels for the vehicle.

[0035] S3, based on driving context labels, selects candidate advertisements with multi-dimensional semantic matching from the labeled advertising content library, combines the duration robustness of the driving state, optimizes the advertisement selection strategy through the reinforcement learning model, realizes context-driven advertisement selection, and obtains the optimal advertisement.

[0036] S4 uses a pre-caching mechanism and seamless switching algorithm to dynamically load and display the optimal advertising content on the bus side window display in real time.

[0037] During the operation of the bus, it is of great significance to accurately obtain and analyze the spatial positioning, posture angle and motion acceleration information of the vehicle, especially in the context of dynamic switching of advertising content. Step S1 ensures the accuracy and stability of the vehicle driving status data through data fusion and processing methods, providing a high-quality data foundation for subsequent situation recognition and advertising matching.

[0038] Step S1 includes the following contents: S1.1, first integrate the data from different sensors to form a comprehensive data stream that fully describes the vehicle's driving status. Specifically, the fused data includes but is not limited to: Spatial positioning information: reflects the vehicle’s geographical location and trajectory.

[0039] Attitude angle information: describes the vehicle's pitch angle, roll angle, etc., reflecting the vehicle's tilt and steering status.

[0040] Motion acceleration information: reflects the vehicle's linear acceleration and angular acceleration, indicating the vehicle's motion change rate.

[0041] For each data source, standardization is first performed to eliminate the impact of different dimensions and magnitudes, ensuring that all data sources are effectively integrated under the same dimension.

[0042] S1.2, in order to improve the accuracy of fused data, a multi-dimensional adaptive fusion matrix is ​​used. This algorithm dynamically adjusts the weight of each data source according to the changes in the real-time environment and vehicle status, thereby ensuring that data sources with higher relevance occupy a larger proportion in the fusion process. This process includes the following steps: Initial data standardization: First, the spatial positioning data , attitude angle data and acceleration data Perform standardization.

[0043] Initial weight setting: Set the initial weight for each data source ,This weight represents the importance of spatial positioning, attitude angle and motion acceleration data in the initial fusion stage.

[0044] State correlation evaluation: The correlation between data sources is calculated through the state correlation function. The evaluation of the state correlation function is based on the mutual information in information theory. ; ,in and Represents different data sources at time By calculating the mutual information, the correlation between the data sources can be quantified.

[0045] Weight update: Dynamically update the weight matrix based on the relevance of each data source The specific weight update formula is: ;in is the correlation vector at the current moment, is the initial weight, is the number of data sources, An index or number that represents a different data source.

[0046] S1.3, use the updated weight matrix to perform weighted combination on the standardized data to obtain the comprehensive driving state vector , which contains the vehicle's spatial positioning, attitude angle, and motion acceleration information, and can accurately reflect the vehicle's current driving state. The specific calculation formula is: ;in, It is the integrated vehicle driving state vector, which contains the weighted results of multiple data sources.

[0047] S1.4, in order to further improve the stability and accuracy of the data, nonlinear Kalman filtering is used to optimize and filter the fused driving state vector. This process includes the following two stages: Prediction stage: Based on the state at the previous moment, use the state transfer matrix Make predictions: ;in, is the predicted vehicle driving state, is the state transfer matrix, describing the state dynamics of the vehicle.

[0048] Update phase: Combine the current observation data , optimize the prediction results by updating the formula: ; ;in, is the Kalman gain, is the forecast error covariance, is the observation matrix, is the observation noise covariance, For observation data.

[0049] Through this process, Kalman filtering can effectively minimize noise and errors, further improving the accuracy of driving status data.

[0050] Step S1 constructs a multi-source data fusion framework, adopts a multi-dimensional adaptive fusion matrix (MDAFM) algorithm and a nonlinear Kalman filter optimization method, and realizes high-precision real-time extraction of vehicle driving status, providing reliable data support for the bus advertising system. The multi-dimensional adaptive fusion matrix dynamically adjusts the fusion weight according to the real-time state correlation of each data source. Through correlation evaluation and adaptive weight adjustment, the positioning, attitude angle and motion acceleration data can be optimally fused in a changing environment to form an accurate driving state vector. Combined with nonlinear Kalman filtering, the fusion results are predicted and corrected to effectively eliminate sensor noise and errors, thereby further improving the stability and robustness of driving state data. The combination of multi-source fusion and dynamic filtering ensures the dynamic and accurate perception of the vehicle's driving status, which not only enables the advertising content to be accurately and intelligently adjusted based on the real-time status of the vehicle, but also lays a solid technical foundation for subsequent driving scenario recognition and intelligent matching of advertising content.

[0051] Step S2 includes the following contents: S2.1, based on the driving state vector, a convolutional neural network is used to extract short-term feature fluctuations and capture local features in the vehicle driving state.

[0052] Convolution operation: Use a one-dimensional convolution kernel to perform sliding calculations on the driving state vector to extract local features in each time period, such as sudden acceleration changes or rapid changes in angular velocity in a short period of time.

[0053] Convolutional layer output: The feature matrix generated by the convolutional layer can capture the instantaneous feature changes of the driving state vector. For example, when the vehicle makes a sharp turn, the lateral acceleration will increase significantly, and the feature matrix output by the convolutional layer will highlight this change.

[0054] S2.2, each feature in the driving state vector has temporal correlation, such as acceleration, deceleration, sharp turn and other state changes have certain continuity and sequence. Therefore, the long short-term memory network is used to further mine the temporal characteristics of the driving state.

[0055] LSTM input: The feature matrix output by the convolutional layer is input into the LSTM, which extracts dependencies in the time series based on the state changes before and after. For example, the acceleration changes of a vehicle when going uphill and turning have obvious dependencies, and the LSTM can capture this change pattern.

[0056] Feature fusion: The output hidden state vector of the long short-term memory network fuses multiple time series features into a high-dimensional feature representation, so that the state changes at different time points can be expressed in a continuous manner, providing a basis for subsequent situation recognition.

[0057] S2.3, after obtaining the fused feature vector, further map the corresponding features to the high-dimensional situational space, capture more complex feature relationships through nonlinear mapping, and finally generate real-time driving situation labels.

[0058] Nonlinear mapping: Based on the kernel function, the fused feature vector is mapped to a high-dimensional space to reveal the nonlinear relationship between the features. For example, when the acceleration and steering angular velocity increase significantly at the same time, the system may judge that the vehicle is in a "sharp turn" situation, and nonlinear mapping can capture these complex relationships.

[0059] Context label generation function: Based on the mapped high-dimensional feature space, a context label generation function is defined to classify the vehicle’s current state into specific context labels such as “sudden acceleration”, “sudden deceleration”, “slow turn”, “sharp turn”, “uphill and downhill”, etc.

[0060] Step S2 accurately constructs the real-time driving situation label of the vehicle through feature analysis based on the driving state vector and situation recognition based on the deep learning model, laying the foundation for subsequent intelligent advertising selection. In this process, the system comprehensively uses the convolutional neural network to extract short-term state change features and the recurrent neural network to capture time series features, deeply integrates multi-dimensional data such as vehicle acceleration, steering angular velocity, and tilt angle, and generates a high-dimensional situation feature vector. Through the situation label generation mechanism, the system realizes the transformation from dynamic and complex driving state data to clear situation labels, providing accurate input for advertising in different situations.

[0061] Step S3 includes the following contents: S3.1, in the labeled advertising content library, retrieve all advertisements related to the current driving situation label to form a preliminary advertising set.

[0062] Each advertisement in the advertising content library is assigned a multi-dimensional attribute label vector, which includes various information such as theme, audience, and applicable context.

[0063] Use a deep semantic embedding model (such as Transformer or BERT) to map the driving context label and ad label vectors into a high-dimensional semantic space to obtain a semantic embedding vector. and .

[0064] Similarity calculation function: ;in, represents the vector inner product, represents the norm of a vector; Indicates The semantic similarity between each advertisement and the current context label is calculated. The candidate advertisements are sorted in descending order according to the similarity value, and the advertisements with similarity higher than the similarity threshold are selected to form the candidate advertisement set.

[0065] S3.2, buses travel on fixed routes, and the duration of driving status on different sections has a certain regularity.

[0066] The duration of the ad needs to match the estimated duration of the vehicle in the current driving situation to ensure that the ad can be played in full and does not affect the passenger experience.

[0067] Collect historical driving data of buses on the same route, the same road section, and the same driving scenario label to obtain a duration sequence .

[0068] Information Entropy To assess the robustness of duration: ;in, For duration data in discrete intervals The probability within , is the number of data in this interval, is the total data volume.

[0069] Information entropy reflects the uncertainty of duration. The smaller the information entropy value, the more concentrated the duration distribution is and the higher the robustness is.

[0070] If the information entropy is less than or equal to the corresponding threshold, the duration is considered to be sufficiently robust.

[0071] The probability density function of duration is obtained using kernel density estimation , using the Gaussian kernel function : ; is the bandwidth parameter, which affects the smoothness of the estimate.

[0072] Determine estimated duration The peak value of the probability density function corresponds to Value, i.e. duration of maximum likelihood estimation.

[0073] S3.3, for each advertisement in the candidate advertisement set, the corresponding playing time is .

[0074] Define the ad duration matching function: ; For duration series The time scale parameter of is calculated from the mutual information of the data. Defined as: ; Make sure the ad does not exceed its estimated duration.

[0075] The ads are sorted in descending order according to the ad duration matching degree, and the ads with matching degrees higher than the corresponding threshold are selected to form the final candidate ad set.

[0076] S3.4, use reinforcement learning to drive advertising selection optimization; select the best advertisement from the final set of candidate advertisements for delivery to improve advertising effectiveness and passenger experience.

[0077] State space: Contains driving situation labels, estimated duration, and attribute features of candidate ads (such as playback duration).

[0078] Action space: Select an ad from the final set of candidate ads to deliver.

[0079] State transfer: As the driving situation of the bus changes, the system will receive a new driving situation label at each time step, resulting in a change in state.

[0080] Advertisement playback completion rate : It is the ratio of the actual playing time of the advertisement to the total playing time of the advertisement. When the advertisement is played completely, the playing completion rate is 1. If the advertisement is interrupted due to a change in the context, the playing completion rate is less than 1.

[0081] Ad switching stability : Set a time window (such as 30 minutes) and count the number of ad switches within the time window , the ad switching stability score is: ;in is a tuning parameter, which represents the expected average number of switching times.

[0082] Reward function: ;in, and is the weight coefficient, which can be automatically adjusted according to the system goal and can be set equal initially.

[0083] The policy optimization algorithm uses deep reinforcement learning algorithms such as Deep Q Network (DQN) or Proximal Policy Optimization (PPO).

[0084] Through the reinforcement learning model, in each state, the action with the maximum Q value is selected according to the policy network or Q value table, that is, the advertisement that best matches the state in the current candidate advertisement set is selected and marked as the optimal advertisement. The selection of this advertisement is not only based on the current driving situation label and the matching degree of the advertisement duration, but also comprehensively considers the historical feedback data such as the advertisement playback completeness rate and system stability to ensure that the advertising delivery effect is consistent with the global optimality of the system goal.

[0085] Step S3 takes driving context labels as the core and constructs a systematic decision-making process from advertising screening to final delivery. The high-dimensional semantic embedding model is used to achieve accurate matching of context labels with multi-dimensional attributes of advertisements, ensuring that the initially screened advertising content has high relevance; on this basis, combined with the duration analysis of historical driving data, the robustness is evaluated using information entropy and the estimated duration is determined through kernel density estimation to ensure that the advertising playback duration matches the driving context. Advertising screening further relies on the duration matching function, combining dynamic time scale parameters to strictly screen the set of advertisements suitable for playback. Within this framework, the selection strategy is optimized through reinforcement learning. Both the state space and the reward function are based on objective and quantifiable system indicators. The advertising playback completeness rate and switching stability are comprehensively considered to establish an intelligent advertising delivery decision-making mechanism that adapts to the dynamic changes of driving contexts.

[0086] After completing the optimal advertisement selection based on the driving situation label, the selected optimal advertisement content needs to be efficiently and stably delivered to the bus side window display. Since buses travel on fixed routes, the driving status and duration of different sections have certain regularity and dynamic changes, and timely loading and switching of advertisement content is particularly important. Step S4 aims to ensure the continuous and smooth efficient display of advertisements during the operation of the vehicle through advanced pre-caching mechanisms and seamless switching algorithms.

[0087] Step S4 includes the following contents: After obtaining the best advertisement, the selected advertisement content is then efficiently loaded into the side window display through the priority cache mechanism. First, the best advertisement is stored in the main cache area of ​​the cache according to the predetermined priority, and the relevant reserve advertisements are stored in the auxiliary cache area according to the situation matching degree. Using the memory management algorithm, the main cache area always stores the high priority segment of the currently playing advertisement, and the auxiliary cache area is dynamically adjusted to store other high matching advertisement segments. A dynamic buffer scheduling strategy based on timestamp is adopted to monitor the driving situation changes in real time, and the seamless switching of advertisement content is realized through preloading and asynchronous loading technology.

[0088] Specific processing includes segmented pre-buffering, splitting the advertising content into multiple playback segments and loading them in parallel to ensure rapid response when the advertisement is switched. Edge computing nodes are used to optimize data transmission rate and reduce latency, combined with adaptive bitrate streaming technology to dynamically adjust the quality of the advertising data stream according to real-time network conditions. In addition, the system integrates an intelligent frame alignment algorithm to ensure the smoothness of visual transitions by accurately synchronizing the end frame of the current advertisement with the start frame of the new advertisement.

[0089] The use of multi-channel data transmission and efficient compression coding technology significantly reduces the space occupied by advertising data in the cache and the transmission time. The combination of dynamic buffer scheduling and multi-channel transmission enables advertising content to be displayed stably and smoothly in various driving scenarios.

[0090] Ultimately, the best ads are efficiently and stably delivered to the side window display through optimized caching and loading processes, ensuring that the ads always maintain high-quality display effects in complex driving environments.

[0091] Step S4 uses a double-layer cache architecture to preferentially store currently playing ads and highly matched alternative ads, and uses segmented pre-buffering technology to split ads into multiple playback segments and load them in parallel to ensure the continuity of the playback process. The intelligent frame alignment algorithm achieves a smooth transition of advertising content to avoid visual interruptions. Edge computing optimizes data transmission rate and latency, and adaptive bitrate streaming technology dynamically adjusts advertising quality to ensure stable playback in different network environments. Multi-channel data transmission and efficient compression encoding further improve loading efficiency, ensuring that ads are presented on the side window display with high resolution and smoothness. The overall design achieves high efficiency and stability in dynamic loading and switching of ads, ensuring the continuity and visual effects of advertising display in dynamic driving scenarios.

[0092] Embodiment 2: Figure 2 The invention provides a vehicle side window advertisement display management system, comprising: a data fusion module, a context recognition module, an advertisement selection module and a dynamic loading module.

[0093] Data fusion module: Based on multi-source data fusion technology, it extracts the comprehensive data stream of the vehicle, and uses Kalman filtering to denoise the data to ensure the accuracy of the driving status data. The processed data is then passed to the situation recognition module.

[0094] Context recognition module: Use the deep learning model to extract features from the fused driving status data, combine the acceleration and steering features to establish a driving context recognition model, generate the real-time context label of the vehicle, and pass the label to the advertising selection module.

[0095] Advertisement selection module: Based on real-time context labels, candidate advertisements with multi-dimensional semantic matching are screened out from the labeled advertisement content library. Combined with the duration robustness of the driving state, the advertisement selection strategy is optimized through the reinforcement learning model, and finally the optimal advertisement is obtained and passed to the dynamic loading module.

[0096] Dynamic loading module: adopts pre-caching mechanism and seamless switching algorithm to dynamically load and accurately display the optimal advertising content on the bus side window display in real time, ensuring the continuity and smoothness of advertising playback.

[0097] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0098] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0099] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0100] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of 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. A method for managing the display of advertisements on the side windows of vehicles, characterized in that: Includes steps: S1, based on multi-source data fusion technology, extracts the comprehensive data stream of the vehicle and uses Kalman filter denoising to ensure the accuracy of driving status data; S2, extracts features from driving status data through a deep learning model, builds a driving situation recognition model based on acceleration and steering features, and generates real-time situation labels for vehicles; S3, based on driving context labels, selects candidate ads with multi-dimensional semantic matching from the labeled advertising content library, combines the duration robustness of driving status, optimizes the advertising selection strategy through a reinforcement learning model, realizes context-driven advertising selection, and obtains the optimal advertisement; S4 uses a pre-caching mechanism and seamless switching algorithm to dynamically load and display the optimal advertising content on the bus side window display in real time.

2. A method for managing vehicle side window advertising display according to claim 1, characterized in that: Step S1 includes the following contents: S1.1, firstly integrate the data from different sensors to form a comprehensive data stream describing the driving status of the vehicle; for each data source, first perform standardization processing to eliminate the influence between different dimensions and magnitudes; S1.2 uses a multi-dimensional adaptive fusion matrix to dynamically adjust the weight of each data source according to the changes in the real-time environment and vehicle status, including the following steps: First, the spatial positioning data, attitude angle data, and acceleration data are standardized; Then, set the initial weight for each data source; Next, the correlation between the data sources is calculated through the state correlation function. The evaluation of the state correlation function is based on the mutual information in information theory. Finally, the correlation between the data sources is quantified by calculating the mutual information; the weight matrix is ​​updated according to the correlation of the data sources; S1.3, using the updated weight matrix to perform weighted combination on the standardized data to obtain a comprehensive driving state vector reflecting the current driving state of the vehicle.

3. A method for managing vehicle side window advertising display according to claim 2, characterized in that: Step S1 also includes the following contents: S1.4, using nonlinear Kalman filtering to optimize and filter the fused driving state vector, including the following two stages: Prediction stage: Based on the state at the previous moment, use the state transfer matrix Make predictions: ;in, is the predicted vehicle driving state, is the state transfer matrix, describing the state dynamics of the vehicle; Update phase: Combine the current observation data , optimize the prediction results by updating the formula: ; ;in, is the Kalman gain, is the forecast error covariance, is the observation matrix, is the observation noise covariance, For observation data.

4. A method for managing vehicle side window advertising display according to claim 3, characterized in that: Step S2 includes the following contents: S2.1, based on the driving state vector, a convolutional neural network is used to extract short-term feature fluctuations and capture local features in the vehicle driving state; S2.2, each feature in the driving state vector has temporal correlation, and the long short-term memory network is used to mine the temporal features of the driving state; S2.3, after obtaining the fused feature vector, the corresponding features are mapped to the high-dimensional situational space, and the feature relationship is captured through nonlinear mapping, and finally a real-time driving situation label is generated.

5. A method for managing vehicle side window advertising display according to claim 4, characterized in that: Step S3 includes the following contents: S3.1, in the labeled advertising content library, retrieve all advertisements related to the current driving situation label to form a preliminary advertising set; use the deep semantic embedding model to map the driving situation label and advertising label vector to a high-dimensional semantic space to obtain the semantic embedding vector and ; Similarity calculation function: ;in, represents the vector inner product, represents the norm of a vector; Indicates The semantic similarity between each advertisement and the current context label is calculated; the candidate advertisements are sorted in descending order according to the similarity value, and the advertisements with similarity higher than the similarity threshold are selected to form a candidate advertisement set.

6. A method for managing vehicle side window advertising display according to claim 5, characterized in that: Step S3 also includes the following contents: S3.2, buses travel on fixed routes, and the duration of driving status in different sections is regular; the duration of advertisements needs to match the estimated duration of the vehicle in the current driving situation to ensure that the advertisements can be played completely without affecting the passenger experience; collect historical driving data of buses on the same route, the same section, and the same driving situation label to obtain the duration sequence ; Use information entropy to evaluate the robustness of duration; If the information entropy is less than or equal to the corresponding threshold, the duration is considered to be sufficiently robust; Use kernel density estimation to obtain the probability density function of duration; Determine the expected duration The peak value of the probability density function corresponds to Value, i.e. duration of maximum likelihood estimation.

7. A method for managing vehicle side window advertising display according to claim 6, characterized in that: Step S3 also includes the following contents: S3.3, for each advertisement in the candidate advertisement set, the corresponding playing time is ; Define the ad duration matching function: ; For duration series The time scale parameter is calculated based on the mutual information of the data; the function Defined as: ; Make sure the ad does not exceed its estimated duration; The ads are sorted in descending order according to the ad duration matching degree, and the ads with matching degrees higher than the corresponding threshold are selected to form the final candidate ad set.

8. A method for managing vehicle side window advertising display according to claim 7, characterized in that: Step S3 also Includes the following: S3.4, use reinforcement learning to drive advertising selection optimization; select the best advertisement from the final candidate advertisement set for delivery; through the reinforcement learning model, in each state, select the action with the maximum Q value according to the Q value table, that is, select the advertisement in the current candidate advertisement set that best matches the state and mark it as the best advertisement.

9. A vehicle side window advertising display management system, used to implement a vehicle side window advertising display management method as claimed in any one of claims 1 to 8, characterized in that: include: Data fusion module, context recognition module, advertisement selection module and dynamic loading module; Data fusion module: Based on multi-source data fusion technology, it extracts the comprehensive data stream of the vehicle and uses Kalman filter denoising to ensure the accuracy of the driving status data. The processed data is then passed to the situation recognition module; Context recognition module: Use the deep learning model to extract features from the fused driving status data, combine acceleration and steering features to establish a driving context recognition model, generate real-time context labels for the vehicle, and pass the labels to the advertisement selection module; Advertisement selection module: Based on real-time context labels, candidate advertisements with multi-dimensional semantic matching are screened out from the labeled advertisement content library. Combined with the duration robustness of the driving state, the advertisement selection strategy is optimized through the reinforcement learning model, and the optimal advertisement is finally obtained and passed to the dynamic loading module. Dynamic loading module: adopts pre-caching mechanism and seamless switching algorithm to dynamically load and accurately display the optimal advertising content on the bus side window display in real time, ensuring the continuity and smoothness of advertising playback.

Images