AI-based methods for identifying and analyzing visitor flow in cultural and tourism towns
By collecting multimodal data and using AI technology to reconstruct the spatiotemporal movement trajectories of individual visitors, the problem of inaccurate visitor flow prediction for cultural and tourism projects has been solved, enabling more accurate visitor flow analysis and planning support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA CONSTR EIGHT ENG DIV CORP LTD
- Filing Date
- 2026-02-27
- Publication Date
- 2026-06-02
AI Technical Summary
In existing technologies, the prediction of visitor flow for cultural and tourism projects in construction engineering relies on references to similar projects, resulting in inaccurate prediction data and posing significant risks.
Using an AI-based approach, multimodal data (mobile terminal signaling data, video stream data, and transaction data) is collected. Pedestrian re-identification models and cross-modal identity association technologies are then used to reconstruct the spatiotemporal movement trajectories of individual passengers, extract multi-dimensional features, form passenger group profiles, and perform clustering and association analyses to generate spatial heat maps and passenger flow direction matrices.
It enables more accurate and comprehensive identification and analysis of visitor flow, providing a refined data foundation while protecting user privacy, and gaining a deeper understanding of visitor behavior characteristics and needs, thus providing data support for scenic area planning.
Abstract
Description
Technical Field
[0001] This invention relates to the field of data analysis technology, specifically to a method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology. Background Technology
[0002] In the planning and implementation of cultural tourism projects within the construction industry, it is necessary to predict and analyze the future visitor flow of the scenic area. This information is then used to further design and plan the overall layout, flow of visitors, and even the location and function of shops. Conventional methods for predicting visitor flow in scenic areas rely on references to similar cultural tourism projects that have been successfully operating in China. However, inaccurate data poses a significant risk to the success of the entire plan. Summary of the Invention
[0003] To overcome the shortcomings of existing technologies, this paper presents an AI-based method for identifying and analyzing visitor flow in cultural and tourism towns. This method addresses the problem that conventional methods for predicting visitor flow in cultural and tourism projects during construction projects can easily lead to inaccurate forecasts.
[0004] To achieve the above objectives, a method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology is provided, including the following steps: Multimodal data is collected within the cultural and tourism town, including mobile terminal signaling data, video stream data, and transaction data of individual visitors; The multimodal data is spatiotemporally aligned to generate a data set; Individual passenger flow data are identified from the video stream data in the dataset to construct a discrete spatiotemporal trajectory point sequence of the individual passenger flow data. The discrete spatiotemporal trajectory point sequence is associated with the anonymized device identifier in the mobile terminal signaling data, and the anonymized transaction account identifier in the transaction flow data corresponding to the anonymized device identifier is associated to generate the data record of the individual passenger flow. The discrete spatiotemporal trajectory point sequence is filled and optimized to reconstruct the complete spatiotemporal movement trajectory of each individual passenger flow; Multi-dimensional feature attributes of each passenger flow individual are extracted from the complete spatiotemporal movement trajectory and the data records, and then multiple passenger flow groups are formed through cluster analysis and corresponding group profile labels are configured. For the complete spatiotemporal movement trajectories of multiple visitor groups, the analysis results are calculated and obtained. The analysis results include the spatial heat map and visitor flow direction OD matrix of multiple visitor groups in the cultural and tourism town, the attractiveness of the regional characteristic elements of the cultural and tourism town to the visitor groups, and the classic tour routes of the cultural and tourism town.
[0005] Furthermore, the mobile terminal signaling data includes an anonymized device identifier, a timestamp sequence, and a corresponding base station location information sequence for each mobile terminal.
[0006] Furthermore, the video stream data is a sequence of consecutive image frames containing pedestrian targets.
[0007] Furthermore, the transaction log data includes anonymized transaction account identifiers, transaction time, transaction amount, and the geographical location information of the transaction merchant.
[0008] Furthermore, before generating a dataset by aligning the multimodal data in time and space, the mobile terminal signaling data is subjected to drift correction and redundancy removal, the video stream data is subjected to noise filtering and illumination normalization, and the transaction flow data is subjected to outlier detection and format standardization.
[0009] Furthermore, the step of generating a data set by spatiotemporal alignment of the multimodal data includes synchronizing and calibrating the timestamp sequence of the mobile terminal signaling data, the image frame timestamp of the video stream data, and the transaction time of the transaction flow data based on a preset global reference clock, and using a geographic information system coordinate system to uniformly map the base station location information sequence, the geographical location of the video surveillance equipment, and the geographical location information of the transaction merchants to the same geographic space coordinate system to generate the data set.
[0010] Furthermore, the step of identifying individual passengers from the video stream data in the dataset to construct a discrete spatiotemporal trajectory point sequence for the individual passengers includes: Each pedestrian target appearing in the video stream data is identified using a pedestrian re-identification model; For each pedestrian target in the video stream data, feature extraction is performed to generate a depth feature vector of the pedestrian target; Calculate the cosine similarity of the depth feature vectors of the different pedestrian targets; Based on a preset association threshold, pedestrian targets with a cosine similarity greater than the association threshold are identified as the same individual in the passenger flow and assigned an ID. The time and location information of individual passengers with the ID appearing under different video health devices are integrated to construct the discrete spatiotemporal trajectory point sequence.
[0011] Furthermore, the multi-dimensional feature attributes include dwell time features, spatial activity features, consumption level features, access preference features, and movement pattern features.
[0012] The beneficial effects of this invention are as follows: The AI-based method for identifying and analyzing visitor flow in cultural and tourism towns overcomes the limitations of single data sources in terms of coverage, accuracy, and dimensionality by integrating mobile terminal signaling data, video stream data, and transaction data, achieving a more comprehensive and accurate identification and analysis of visitor flow within the cultural and tourism town. This invention utilizes a pedestrian re-identification model and cross-modal identity association technology to identify, track, and bind data for each individual visitor while protecting user privacy, generating continuous spatiotemporal movement trajectories and providing a refined data foundation for subsequent analysis. This invention extracts features from multiple dimensions such as dwell time, spending level, visit preferences, and movement patterns, and forms representative visitor groups and their profile labels through cluster analysis, which helps to deeply understand the behavioral characteristics and needs of different types of tourists. This invention quantitatively analyzes the attractiveness of different visitor groups to architectural styles, cultural nodes, and business formats by combining a knowledge graph of regional characteristic elements, revealing the intrinsic connection between spatial design and tourist behavior. This invention provides data support for planning and design by outputting results such as spatial heat maps, OD matrices of passenger flow direction, and classic tour routes, which intuitively display passenger flow distribution, flow patterns, and behavioral preferences. Detailed Implementation
[0013] The present application will now be described in further detail with reference to the embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit the invention.
[0014] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present application will now be described in detail with reference to the embodiments.
[0015] This invention provides a method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology, comprising the following steps: S1. Collect multimodal data within the cultural and tourism town. The multimodal data includes mobile terminal signaling data, video stream data, and transaction flow data of individual visitors.
[0016] Mobile terminal signaling data includes each mobile terminal's anonymized device identifier, timestamp sequence, and corresponding base station location information sequence.
[0017] The mobile terminal signaling data is obtained from the communication base stations deployed in the cultural and tourism town.
[0018] The video stream data is a sequence of consecutive image frames containing pedestrian targets.
[0019] The video stream data was captured by video surveillance equipment deployed at key path nodes and points of interest areas within the cultural and tourism town.
[0020] Transaction log data includes anonymized transaction account identifiers, transaction time, transaction amount, and the geographical location information of the transaction merchants.
[0021] Transaction data is recorded by an electronic payment system deployed at commercial service points within the cultural and tourism town.
[0022] As a preferred implementation, multimodal data also includes user-generated content data. This includes user-generated content data related to the cultural and tourism town, obtained through the town's official social media accounts or application programming interfaces.
[0023] User-generated content data includes geotagged posts, comment text, and uploaded images.
[0024] Before collecting multimodal data, a digital archiving process for regional characteristic elements is first carried out.
[0025] Regional characteristic elements are predefined spatial entities or areas that represent the architectural style, cultural connotation, or business format of a cultural and tourism town.
[0026] This digital archiving process categorizes and codes every building, shop, landscape, and cultural facility in the cultural and tourism town through manual annotation or automated information extraction, and records information such as its geographical location, functional attributes, cultural background, and design style, forming a structured knowledge graph of regional characteristic elements.
[0027] S2. Generate a dataset by aligning the multimodal data in time and space.
[0028] Before generating a dataset by aligning multimodal data in time and space, drift correction and redundancy removal are performed on the mobile terminal signaling data, noise filtering and illumination normalization are performed on the video stream data, and outlier detection and format standardization are performed on the transaction flow data.
[0029] Specifically, drift correction for mobile terminal signaling data is achieved by using a correction model based on historical data statistics to compensate for base station positioning errors caused by multipath effects or signal blockage.
[0030] For the illumination normalization processing of video stream data, the Limit Contrast Adaptive Histogram Equalization (CLAHE) algorithm is used to enhance the stability and consistency of the image under different illumination conditions.
[0031] The steps of generating a dataset by aligning multimodal data in time and space include using a preset global reference clock as a benchmark to synchronize and calibrate the timestamp sequence of mobile terminal signaling data, the image frame timestamp of video stream data, and the transaction time of transaction flow data, and using a geographic information system coordinate system to uniformly map the base station location information sequence, the geographic location of video surveillance equipment, and the geographic location information of transaction merchants to the same geographic space coordinate system to generate a dataset.
[0032] S3. Identify individual passenger flow data from the video stream data in the dataset to construct a discrete spatiotemporal trajectory point sequence of the individual passenger flow data.
[0033] The steps for identifying individual passengers from video stream data in a dataset to construct a discrete spatiotemporal trajectory point sequence for those passengers include: S31. Use a pedestrian re-identification model to identify each pedestrian target appearing in the video stream data.
[0034] S32. Extract features from each pedestrian target in the video stream data to generate a depth feature vector of the pedestrian target.
[0035] S33. Calculate the cosine similarity of the depth feature vectors of different pedestrian targets.
[0036] S34. Based on the preset association threshold, pedestrian targets with cosine similarity greater than the association threshold are identified as the same passenger flow individual and assigned an ID (i.e., a globally unique anonymized passenger flow ID).
[0037] The generation and management of globally unique anonymized passenger flow IDs follow a differential privacy protection mechanism. In each stage of data collection, transmission and analysis, precisely calculated noise is added to individual data to ensure that while outputting macro-statistical analysis results, it is impossible to infer the precise information of any single passenger flow individual, thereby technically protecting the personal privacy and security of tourists.
[0038] S35. Integrate the time and location information of individual passengers with IDs from different video health devices to construct a discrete spatiotemporal trajectory point sequence.
[0039] In this embodiment, the pedestrian re-identification model is a pre-trained, deep learning-based pedestrian re-identification model. Specifically, the deep learning-based pedestrian re-identification model is a ResNet structure that integrates channel attention and spatial attention mechanisms. When extracting deep feature vectors, the model can adaptively focus on the feature regions and feature channels in the image that are most discriminative for identity recognition, thereby improving the recognition accuracy in complex scenes, such as crowd occlusion and pose changes.
[0040] S4. Associate the discrete spatiotemporal trajectory point sequence with the anonymized device identifier in the mobile terminal signaling data, and associate the anonymized transaction account identifier in the transaction flow data corresponding to the anonymized device identifier to generate data records for individual passenger flows.
[0041] Specifically, by performing cross-modal identity association operations, and utilizing a spatiotemporally aligned multi-source data set, based on the principle of spatiotemporal proximity, discrete spatiotemporal trajectory point sequences based on video surveillance data are associated with anonymized device identifiers in mobile terminal signaling data. The anonymized transaction account identifiers in the transaction log data corresponding to the successfully associated anonymized device identifiers are then associated, thereby binding the information of multimodal data to the same globally unique anonymized passenger flow ID to form a multimodal fused individual passenger flow data record.
[0042] The cross-modal identity association operation includes: Construct a matching function with time window and geographic proximity as constraints. For any trajectory point generated from video data, search for mobile terminal signaling data points within its geographic proximity range within the same time window.
[0043] If there is only one mobile terminal signaling data point within the range, the two will be strongly correlated; if there are multiple points, the optimal match will be made based on the similarity of kinematic characteristics such as speed and direction, thereby achieving anonymized anchoring of individual passenger flow identities in both the physical and digital worlds.
[0044] S5. Fill and optimize the discrete spatiotemporal trajectory point sequence to reconstruct the complete spatiotemporal movement trajectory of each individual passenger.
[0045] As a preferred implementation method, spatiotemporal interpolation and path smoothing are first performed on the multimodal fusion of individual passenger flow data records. Then, a trajectory estimation algorithm based on Kalman filtering is called to fill and optimize the discrete spatiotemporal trajectory point sequence, reconstructing a complete spatiotemporal movement trajectory with continuity and high accuracy for each individual passenger flow within the cultural and tourism town.
[0046] The trajectory estimation algorithm based on Kalman filtering uses a state vector that includes the position coordinates and velocity components of individual passenger flow data. Its observation vector is derived from discrete spatiotemporal points in multimodal fusion passenger flow data records. Through iterative cycles of prediction and update, the algorithm optimally estimates the state of individual passenger flow data at the next moment, effectively filtering out observation noise and filling in missing trajectory segments.
[0047] S6. Extract multi-dimensional feature attributes of each passenger flow individual from the complete spatiotemporal movement trajectory and data records, and then form multiple passenger flow groups through cluster analysis and configure corresponding group profile labels.
[0048] Multidimensional features include dwell time, spatial activity, consumption level, access preferences, and movement patterns.
[0049] The duration of stay is the total time that the individual visitor spends within the cultural and tourism town.
[0050] Spatial activity characteristics are the number of points of interest visited and the activity radius of the individual in the visitor flow.
[0051] The consumption level characteristics are the cumulative transaction amount and average single transaction amount of this individual customer in the electronic payment system.
[0052] The visitor preference feature is the distribution of the dwell time of the individual visitor in different types of points of interest areas.
[0053] The movement pattern feature is the main movement path type of the individual passenger flow. The extraction of the movement pattern feature is achieved by first spatially gridding the complete spatiotemporal movement trajectory of each individual passenger flow, converting the continuous coordinate sequence into a discrete access sequence of the point of interest region, and then applying the n-gram model analysis method to extract the frequently occurring continuous access subsequence as the movement pattern feature of the individual passenger flow.
[0054] User-generated content data is used as a supplementary basis for extracting individual sentiment and interest tags of customers.
[0055] The multi-dimensional features of all individual passengers are organized into a high-dimensional feature space. An unsupervised machine learning clustering algorithm, such as the density scan-based DBSCAN algorithm, is called to perform cluster analysis on all individual passengers in the high-dimensional feature space. Individual passengers whose feature vectors are close in space are grouped into the same passenger group, and a group profile label is generated for each passenger group.
[0056] Group profile tags include, for example, “high-value overnight tourists,” “pop-up tourist groups,” or “local leisure and consumption groups.”
[0057] After performing cluster analysis, the process also includes generating an automated group profile label. Specifically, by calculating the centroid value of the characteristic attributes of each cluster, which is the average value of all individuals in the group in various dimensions such as stay time and consumption level, and matching these centroid values with a preset semantic rule base, the system automatically assigns the group a human-readable group profile label that best summarizes its core characteristics.
[0058] S7. Calculate and analyze the complete spatiotemporal movement trajectories of multiple visitor groups. The analysis results include spatial heat maps and OD matrices of visitor flow directions within the cultural and tourism town, the attractiveness of the regional characteristics of the cultural and tourism town to visitor groups, and the classic tour routes of the cultural and tourism town.
[0059] Based on the complete spatiotemporal movement trajectory data of different passenger groups, spatial heat maps and OD matrices of passenger flow direction are generated within the cultural and tourism town.
[0060] Spatial heat maps are used to visualize the areas where different customer groups congregate in space.
[0061] The OD matrix is used to quantify the intensity of passenger flow transfer between different points of interest.
[0062] The generation of the passenger flow direction OD matrix is to first define all points of interest within the cultural and tourism town as sets of O points (starting points) and D points (arrival points), then traverse the complete spatiotemporal movement trajectory of each individual passenger flow, and count the frequency of passenger flow from any O point to any D point within a unit of time, finally forming an N×N matrix, where N is the total number of points of interest, and the element values in the matrix represent the passenger flow intensity between point pairs.
[0063] On the other hand, a correlation analysis model is constructed between regional characteristic elements and visitor behavior. The correlation analysis model quantifies the attractiveness of regional characteristic elements to a specific visitor group by calculating indicators such as the average length of stay, revisit rate, and consumption conversion rate in areas with specific regional characteristic elements.
[0064] The association analysis model incorporates a time decay factor. When calculating the attractiveness of specific regional characteristics to visitor groups, it assigns lower weights to earlier user behavior data and higher weights to more recent user behavior data. This allows the analysis results to dynamically reflect the latest trends in tourist preferences. Knowledge graphs serve as the core object of association analysis. The process involves motion pattern mining using sequence pattern mining algorithms, such as the GSP algorithm. From the spatiotemporal movement trajectory sequences of various visitor groups, classic tour routes with a frequency exceeding a preset support threshold are identified—these are considered strongly correlated visit sequences. The spatial coupling relationship between different classic tour routes and regional characteristic elements is then analyzed. The criteria for determining strongly correlated visit sequences include not only the support threshold but also a confidence threshold. Support is defined as the proportion of trajectories containing the visit sequence to the total number of trajectories, while confidence is defined as the conditional probability of visiting subsequent nodes after visiting preceding nodes, ensuring that the mined motion patterns have statistical significance and predictive value.
[0065] As a preferred implementation method, the AI-based method for identifying and analyzing visitor flow in cultural and tourism towns of the present invention further includes step S7: generating decision support information for the construction of regionally distinctive scenic spots.
[0066] Specifically, step S7 includes: S71. Based on the aforementioned analysis results, potential problems in the current spatial layout of cultural and tourism towns are automatically identified.
[0067] Potential problems include: whether the carrying capacity of popular tourist areas is saturated, whether there are "cold areas" with significant tourist flow but lack of commercial facilities, whether regional characteristic elements with high attraction potential are not fully utilized, and whether there are cross-conflicts in the tour routes between different tourist groups.
[0068] The capacity saturation of high-traffic areas is identified by comparing the real-time passenger flow density of the area with preset safety and comfort thresholds. When the real-time passenger flow density continuously exceeds the comfort threshold for a certain period of time, or momentarily exceeds the safety threshold, the area is determined to be saturated, and an early warning is triggered.
[0069] S72. Call a rule-based reasoning expert system or a generative adversarial network model to generate optimized suggestions for the construction or renovation of regionally distinctive scenic spots based on the identified potential problems and in combination with preset planning and design constraints. The optimization suggestions are output in a structured data format, and their content includes: Recommendations for the types of new or adjusted commercial service facilities and their optimal spatial locations; It is recommended to strengthen the delineation of regional cultural characteristic theme areas and their spatial scope; And suggested themed tour routes designed to divert different visitor groups.
[0070] The specific application methods of Generative Adversarial Network (GAN) models are as follows: The generator is responsible for generating candidate spatial layout optimization maps or modifying parameter sets based on the input potential problem codes and planning constraints. The discriminator is responsible for learning from a large amount of excellent and real-world scenic area planning case data, and scoring the candidate schemes generated by the generator to determine their similarity to real-world excellent cases. Through adversarial training, the generator is eventually able to produce highly realistic optimization suggestions that conform to planning principles.
[0071] Finally, a visual, interactive analysis interface is used to display the generated spatial heat map, OD matrix, group profile, correlation analysis results, and optimization suggestions in a layer overlay, data drill-down, and 3D modeling manner. This allows planning decision-makers to perform interactive parameter adjustments and simulations to evaluate the potential impact of different optimization suggestions on passenger flow.
[0072] The AI-based method for identifying and analyzing visitor flow in cultural and tourism towns is a comprehensive data processing and analysis method. It constructs an analysis method that integrates multi-source heterogeneous data fusion, individualized spatiotemporal trajectory reconstruction of visitor flow, in-depth characterization of multi-dimensional tourist profiles, and mining of the correlation between behavioral preferences and spatial elements.
[0073] The AI-based method for identifying and analyzing visitor flow in cultural and tourism towns, as proposed in this invention, conducts in-depth analysis of visitor flow in these towns and ultimately generates decision support information to guide the planning and construction of distinctive scenic spots. By integrating multi-source data and using AI models to create profiles and analyze the behavior of tourists, this method ultimately serves the planning and optimization of tourist destinations. Its integration and implementation in the specific business application of tourism management and planning has significant potential for widespread application.
[0074] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.
Claims
1. A method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology, characterized in that, Includes the following steps: Multimodal data is collected within the cultural and tourism town, including mobile terminal signaling data, video stream data, and transaction data of individual visitors; The multimodal data is spatiotemporally aligned to generate a data set; Individual passenger flow data are identified from the video stream data in the dataset to construct a discrete spatiotemporal trajectory point sequence of the individual passenger flow data. The discrete spatiotemporal trajectory point sequence is associated with the anonymized device identifier in the mobile terminal signaling data, and the anonymized transaction account identifier in the transaction flow data corresponding to the anonymized device identifier is associated to generate the data record of the individual passenger flow. The discrete spatiotemporal trajectory point sequence is filled and optimized to reconstruct the complete spatiotemporal movement trajectory of each individual passenger flow; Multi-dimensional feature attributes of each passenger flow individual are extracted from the complete spatiotemporal movement trajectory and the data records, and then multiple passenger flow groups are formed through cluster analysis and corresponding group profile labels are configured. For the complete spatiotemporal movement trajectories of multiple visitor groups, the analysis results are calculated and obtained. The analysis results include the spatial heat map and visitor flow direction OD matrix of multiple visitor groups in the cultural and tourism town, the attractiveness of the regional characteristic elements of the cultural and tourism town to the visitor groups, and the classic tour routes of the cultural and tourism town.
2. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 1, characterized in that, The mobile terminal signaling data includes an anonymized device identifier, a timestamp sequence, and a corresponding base station location information sequence for each mobile terminal.
3. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 2, characterized in that, The video stream data is a sequence of consecutive image frames containing pedestrian targets.
4. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 3, characterized in that, The transaction log data includes anonymized transaction account identifiers, transaction time, transaction amount, and the geographical location information of the transaction merchants.
5. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 4, characterized in that, Before generating a dataset by aligning the multimodal data in time and space, the mobile terminal signaling data is subjected to drift correction and redundancy removal, the video stream data is subjected to noise filtering and illumination normalization, and the transaction flow data is subjected to outlier detection and format standardization.
6. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 4, characterized in that, The step of generating a data set by spatiotemporal alignment of the multimodal data includes synchronizing and calibrating the timestamp sequence of the mobile terminal signaling data, the image frame timestamp of the video stream data, and the transaction time of the transaction flow data based on a preset global reference clock, and using a geographic information system coordinate system to uniformly map the base station location information sequence, the geographic location of the video surveillance equipment, and the geographic location information of the transaction merchants to the same geographic space coordinate system to generate the data set.
7. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 1, characterized in that, The step of identifying individual passengers from video stream data in the dataset to construct a discrete spatiotemporal trajectory point sequence for the individual passengers includes: Each pedestrian target appearing in the video stream data is identified using a pedestrian re-identification model; For each pedestrian target in the video stream data, feature extraction is performed to generate a depth feature vector of the pedestrian target; Calculate the cosine similarity of the depth feature vectors of the different pedestrian targets; Based on a preset association threshold, pedestrian targets with a cosine similarity greater than the association threshold are identified as the same individual in the passenger flow and assigned an ID. The time and location information of individual passengers with the ID appearing under different video health devices are integrated to construct the discrete spatiotemporal trajectory point sequence.
8. The method for identifying and analyzing visitor flow in cultural and tourism towns based on AI technology according to claim 1, characterized in that, The multi-dimensional feature attributes include dwell time features, spatial activity features, consumption level features, access preference features, and movement pattern features.