Crowd flow prediction method and apparatus
By acquiring spatiotemporal text data and geographic information, and using a pedestrian flow prediction model to predict pedestrian flow, unreasonable data is eliminated, which solves the problem of insufficient response to emergencies in existing technologies and improves the accuracy of pedestrian flow prediction and the adaptability of the model.
Patent Information
- Application Number
- PCT/CN2026/075168
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-03-04
- Filing Date
- 2026-01-27
- Publication Date
- 2026-11-12
AI Technical Summary
Existing methods for predicting pedestrian flow rely too heavily on historical movement data and lack the ability to respond to emergencies, resulting in inaccurate predictions.
By acquiring spatiotemporal text data and using spatiotemporal entity association information to predict pedestrian flow, and combining geographic information and pedestrian flow prediction models, unreasonable pedestrian flow data is eliminated, and model training and parameter optimization are carried out to improve prediction accuracy.
It enables effective response to emergencies, improves the accuracy of crowd flow forecasting and the model's self-optimization capability, and adapts to the crowd flow forecasting needs of different regions.
Smart Images

Figure CN2026075168_12112026_PF_FP_ABST
Abstract
Description
A method and device for predicting pedestrian flow
[0001] Cross-references to related applications
[0002] This application claims priority to Chinese Patent Application No. 202510252368.5, filed on March 4, 2025, entitled "A Method and Apparatus for Predicting Crowd Flow", the entire contents of which are incorporated herein by reference. Technical Field
[0003] This application relates to the field of artificial intelligence technology, and in particular to a method and apparatus for predicting pedestrian flow. Background Technology
[0004] Predicting pedestrian flow within urban spaces can provide support for various scenarios, including public order management (such as crowd control), emergency command (such as preventing stampedes), urban infrastructure layout (such as the layout of public facilities and the identification of urban traffic bottlenecks), large-scale and important events (such as New Year's Eve light shows), management and security for specific groups, and urban traffic management. For example, during New Year's Eve light shows or large-scale sporting events, how to reasonably determine crowd control plans to avoid safety hazards in public places; and in daily traffic management, how to rationally dispatch taxis to meet passenger demand—all require the support of accurate pedestrian flow data. Furthermore, in the field of business intelligence, accurate pedestrian flow data is also needed as a reference in decision-making processes such as store location selection and advertising placement.
[0005] Improving the accuracy of crowd flow prediction is a problem that needs to be solved. Summary of the Invention
[0006] This application provides a method and apparatus for predicting pedestrian flow. By acquiring spatiotemporal text data including events, and utilizing the attribution ability of spatiotemporal entity association information, pedestrian flow prediction can be performed. This allows for full consideration of the impact of sudden events on pedestrian flow when making pedestrian flow predictions, thereby improving the accuracy of pedestrian flow prediction.
[0007] Firstly, embodiments of this application provide a method for predicting pedestrian flow. This method can be applied to pedestrian flow prediction devices, which can be servers, server clusters, personal computers, laptops, smart speakers, robots, mobile phones, maintenance equipment, etc., or components of such devices (e.g., processors, modules, chips, or chip systems), or devices used in conjunction with the pedestrian flow prediction device. It can also be cloud platforms, cloud computing centers, etc. The method includes: acquiring spatiotemporal text data associated with a first region, the spatiotemporal text data including events and information about the time, location, and roles associated with the events; determining spatiotemporal entity association information for the first region based on the spatiotemporal text data, the spatiotemporal entity association information including the association relationships between four types of entities: roles, time, location, and events; and determining the pedestrian flow in the region to be predicted during the predicted time period based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the region to be predicted, using a pedestrian flow prediction model, wherein the region to be predicted belongs to the first region.
[0008] Optionally, obtaining the spatiotemporal text data associated with the first region includes: obtaining the spatiotemporal text data associated with the first region based on social media and / or the Internet.
[0009] Using the methods described above, crowd flow prediction devices can acquire spatiotemporal text data through social media and / or the internet, such as a school A holding a general assembly of all teachers and students from 16:00 to 17:35 today. By utilizing the attribution ability of spatiotemporal entity association information, crowd flow prediction can be made, which allows for full consideration of the impact of sudden events such as general assemblies of all teachers and students or large-scale sports events on crowd flow when making crowd flow predictions, thus improving the accuracy of crowd flow prediction.
[0010] In one possible design, the method further includes: acquiring geographic information of a first region, the geographic information including at least one of street view data, remote sensing data, or base map data of the first region; and determining spatiotemporal entity association information based on spatiotemporal text data, including: determining spatiotemporal entity association information based on spatiotemporal text data and geographic information of the first region.
[0011] By using the above method, considering that different locations correspond to different geographical features, such as different buildings, different areas of interest (AOI), or different points of interest (POI), determining the spatiotemporal entity association information based on the geographical information and spatiotemporal text data of the first region is beneficial for accurately characterizing the locations that may be affected by the event, thereby improving the accuracy of pedestrian flow prediction.
[0012] In one possible design, based on spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted, the pedestrian flow prediction model is used to determine the pedestrian flow in the area to be predicted during the time period. This includes: determining multiple pedestrian flow data points for the area to be predicted during the time period based on spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted, using the pedestrian flow prediction model. Each pedestrian flow data point includes a person and the person's corresponding role, time, location, and event; and determining the pedestrian flow based on the multiple pedestrian flow data points.
[0013] The above design enables pedestrian flow prediction devices to determine multiple pedestrian flow data points in a specific time (i.e., the time period to be predicted) and space (i.e., the area to be predicted) based on spatiotemporal entity association information, and to count the pedestrian flow in a specific time and space based on the multiple pedestrian flow data points.
[0014] In one possible design, the method further includes: determining a reasonableness score for the first flow data, wherein the first flow data belongs to multiple flow data; if the reasonableness score of the first flow data is less than or equal to a first threshold, the first flow data is discarded.
[0015] Through the above design, unreasonable pedestrian flow data can be eliminated by absolutely verifying individual pedestrian flow data, thereby obtaining high-quality pedestrian flow data and improving the accuracy of pedestrian flow prediction.
[0016] In one possible design, the method further includes: determining a difference score between the second flow data and the third flow data, wherein the second flow data and the third flow data belong to multiple flow data and the roles corresponding to the second flow data and the third flow data are different; if the difference score between the second flow data and the third flow data is less than or equal to a second threshold, the second flow data and the third flow data are discarded.
[0017] Through the above design, relative verification of pedestrian flow data for different roles can be performed to eliminate unreasonable pedestrian flow data, thereby obtaining high-quality pedestrian flow data and improving the accuracy of pedestrian flow prediction.
[0018] In one possible design, the method further includes: training a pedestrian flow prediction model based on multiple pedestrian flow data used to determine pedestrian flow, location information of the area to be predicted, time period to be predicted, and spatiotemporal entity association information.
[0019] Through the above design, high-quality pedestrian flow data obtained by eliminating unreasonable pedestrian flow data can be used to train the pedestrian flow prediction model, thereby fine-tuning the parameters of the pedestrian flow prediction model. This enables the self-optimization and iteration of the pedestrian flow prediction model, which is beneficial to improving the accuracy of pedestrian flow prediction. Furthermore, after the pedestrian flow prediction equipment or model is migrated across regions, it can be better adapted to the pedestrian flow prediction environment of the migrated region, which is beneficial to improving the migration and expansion capabilities of the pedestrian flow prediction equipment or model across regions.
[0020] Secondly, embodiments of this application provide an apparatus that performs the function of the method described in the first aspect. This function can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the described function, such as an input / output module and a processing module.
[0021] In one possible design, the device can be a chip or an integrated circuit.
[0022] In one possible design, the device includes a memory and a processor, the memory for storing instructions executed by the processor, and when the instructions are executed by the processor, the device can perform the method of the first aspect.
[0023] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program or instructions, which, when executed by a processor, can implement the method described in the first aspect.
[0024] Fourthly, embodiments of this application also provide a computer program product, including a computer program or instructions, which, when executed by a processor, can implement the method described in the first aspect.
[0025] Fifthly, embodiments of this application also provide a chip system including a processor, which is coupled to a memory for storing programs or instructions. When the program or instructions are executed by the processor, the method described in the first aspect can be implemented.
[0026] The technical effects achievable by the second to fifth aspects mentioned above are similar to those achievable by the first aspect mentioned above, and will not be repeated here. Attached Figure Description
[0027] Figure 1 is a schematic diagram of the application of the pedestrian flow prediction provided in this embodiment of the application to an antenna tuning scenario;
[0028] Figure 2 is a schematic diagram of the pedestrian flow prediction method provided in an embodiment of this application;
[0029] Figure 3 shows a spatiotemporal entity association diagram provided in an embodiment of this application;
[0030] Figure 4 is a schematic diagram of the pedestrian flow prediction processing procedure provided in an embodiment of this application;
[0031] Figure 5 is a schematic diagram of the antenna optimization process based on pedestrian flow prediction provided in an embodiment of this application;
[0032] Figures 6 and 7 are schematic diagrams of the structure of the device provided in the embodiments of this application. Detailed Implementation
[0033] To facilitate understanding by those skilled in the art, some terms used in this application are explained below.
[0034] 1) Large model.
[0035] Large models, also known as artificial intelligence (AI) large models, refer to "large parameter" models trained using massive amounts of data and powerful computing capabilities. These models typically possess high versatility and generalization ability, and can be applied to fields such as natural language processing, image recognition, and speech recognition. They can be categorized into large language models (LLM), visual large models, multimodal large models, and basic large models. The smallest processing unit in a large model can be called a token.
[0036] Taking large language models as an example, these models aim to understand and generate human language. They are trained on massive amounts of text data to learn the ability to understand and generate human language, and can perform a wide range of tasks, including text summarization, translation, sentiment analysis, and more. The core idea of large language models is to learn patterns and structures of natural language through large-scale unsupervised training, which to some extent simulates the human language cognition and generation process.
[0037] 2) Prompt.
[0038] Cue words are instructions, questions, or hints used to guide large models in generating specific outputs. They typically include context, input, or examples. Using cue words can improve the quality of specific outputs generated by large models.
[0039] 3) Geographic information.
[0040] Geographic information can include street view data, remote sensing data, or base map data, etc.
[0041] 4) Spatiotemporal text data.
[0042] Spatiotemporal text data includes information about events and the time, place, and roles associated with them. In other words, spatiotemporal text data can describe events and the time, place, and roles associated with them.
[0043] 5) Spatiotemporal entity association information.
[0044] Spatiotemporal entity association information can be represented in various forms, such as spatiotemporal entity association tables, spatiotemporal entity association databases, or spatiotemporal entity association graphs. Spatiotemporal entity association graphs can also be called spatiotemporal entity relationship graphs or spatiotemporal knowledge graphs. Spatiotemporal entity association information includes four types of entities—roles, time, location, and events—and the relationships between these entities. For example, prompt words can guide a large model to extract the four types of entities (roles, time, location, and events) and their relationships from multi-source data (such as spatiotemporal text data), storing this information in a semantically encoded embedding format. The embedded encoding can then be used to reconstruct the original semantics from a vocabulary.
[0045] 6) Pre-trained model.
[0046] A pre-trained model can refer to a deep learning model that has been pre-trained in unsupervised mode on a large-scale dataset. For example, the pedestrian flow prediction model in this application embodiment can be a pre-trained model.
[0047] 7) Spatiotemporal fine-tuning.
[0048] Spatiotemporal fine-tuning can refer to constructing contextual dialogues using high-quality pedestrian flow data corresponding to roles, times, locations, and events, and adjusting some parameters of the pre-trained model.
[0049] Pedestrian flow prediction within urban spaces can provide support for scenarios such as public order management (e.g., crowd control), emergency command (e.g., preventing stampedes), urban infrastructure layout (e.g., public facility layout, or identification of urban traffic congestion points), large-scale and important events (e.g., New Year's Eve light shows), management and support for specific groups, and urban traffic management. Currently, various types of sensors and location information from mobile devices can be deployed to collect pedestrian flow information in real time, and pedestrian flow data for a future period can be predicted based on the collected information. For example, the following scheme A or scheme B is currently used for pedestrian flow prediction.
[0050] Option A: Based on the characteristics of pedestrian flow, weather, holidays, and weekdays within a first preset time period, a Long Short-Term Memory (LSTM) network model is used to obtain the continuous variation pattern of pedestrian flow over the first preset time period. Based on this continuous variation pattern, and the weather, holiday, and weekday characteristics of the time period to be predicted, the target pedestrian flow for the time period to be predicted is determined. However, this option relies heavily on historical data, resulting in a lack of real-time update capabilities. Complete reliance on historical data makes it difficult to cope with unexpected events (such as a surge in concert attendance).
[0051] Option B: By acquiring historical pedestrian movement trajectory data of the block and inputting this data into a pre-trained block pedestrian flow prediction model, the future pedestrian flow prediction result of the block is obtained from the output of the block pedestrian flow prediction model. This technique relies excessively on historical pedestrian movement trajectory data, performs poorly in new scenarios or sudden events, and has high complexity in handling spatiotemporal migration.
[0052] As the above introduction shows, current methods for predicting pedestrian flow have limitations. For example, they rely too heavily on historical movement trajectory data, lack contingency plans for major sporting events and other emergencies, and lack online data update capabilities, leading to inaccurate pedestrian flow predictions.
[0053] Based on this, embodiments of this application provide a method and apparatus for predicting pedestrian flow, aiming to improve the accuracy of pedestrian flow prediction. The embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0054] The pedestrian flow prediction method provided in this application can accurately predict pedestrian flow in specific areas of a city (such as tourist attractions, large shopping malls, residential areas, etc.) at any given time per unit time, and can be applied to multiple fields such as antenna optimization, urban management, and emergency command.
[0055] The pedestrian flow prediction method provided in this application embodiment can be executed by a pedestrian flow prediction device. This pedestrian flow prediction device can be a server, server cluster, personal computer, laptop, smart speaker, robot, mobile phone, maintenance equipment, or other equipment, or a component of the equipment (such as a processor, module, chip, or chip system), or a device that is matched with the pedestrian flow prediction device, or a cloud platform, cloud computing center, etc.
[0056] Taking the application of pedestrian flow prediction methods to antenna optimization, with the pedestrian flow prediction device being an operation and maintenance device as an example, Figure 1 is a schematic diagram of the application of pedestrian flow prediction to antenna optimization scenarios provided in this application embodiment. The operation and maintenance device can obtain spatiotemporal text data associated with a first region from a search server; additionally, it can also obtain geographical information of the first region. This geographical information can be obtained from an external source (e.g., a map server, which offers stronger dynamic update capabilities) or pre-configured within the operation and maintenance device. The operation and maintenance device can also determine spatiotemporal entity association information based on spatiotemporal text data (or spatiotemporal text data and geographical information), and predict pedestrian flow in a specific time and space (e.g., a specific region during a specific time period) based on this spatiotemporal entity association information. After determining the pedestrian flow, the operation and maintenance device can form a dynamic decision plan based on the pedestrian flow and send it to the digital antenna. The digital antenna uses the decision plan to remotely and dynamically adjust the antenna downtilt angle and azimuth angle, etc. For example, if the operation and maintenance device determines that the pedestrian flow in a specific time and space is greater than a set threshold, the resulting decision plan could be to adjust the orientation of some antennas in the digital antenna, causing more antennas to face the specific area where the pedestrian flow is predicted.
[0057] Furthermore, in the description of this application, terms such as "first" and "second" are used only to distinguish multiple objects and are not used to limit the size, content, order, timing, priority, or importance of the multiple objects. For example, "first flow data" and "second flow data" do not indicate a difference in priority or importance between the two sets of flow data.
[0058] In this application embodiment, the number of nouns, unless otherwise specified, refers to "singular nouns or plural nouns," that is, "one or more." "At least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, or B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. For example, A / B means: A or B. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c means: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.
[0059] Figure 2 is a schematic diagram of a pedestrian flow prediction method provided in an embodiment of this application. The method includes:
[0060] S201: The pedestrian flow prediction device acquires spatiotemporal text data associated with the first area. This spatiotemporal text data includes events, as well as information about the time, location, and roles associated with those events.
[0061] In one possible implementation, people flow prediction devices can be implemented via social media (such as...) Searching for spatiotemporal text data associated with the first region using methods such as portals, search engines, news websites, etc.
[0062] Taking town B as an example in the first region, a user publicly posted on the portal website that a general meeting of all teachers and students will be held at school A on March 9, 2025, from 16:00 to 17:35. School A is located in town B. The population flow prediction device can obtain the spatiotemporal text data through the portal website: School A will hold a general meeting of all teachers and students on March 9, 2025, from 16:00 to 17:35.
[0063] For example, if a user publicly posts on social media that celebrity XX will hold a concert at Stadium C on March 10, 2025, from 19:00 to 21:00, and Stadium C is located in town B, then the crowd prediction device can obtain the spatiotemporal text data through social media: Celebrity XX will hold a concert at Stadium C on March 10, 2025, from 19:00 to 21:00.
[0064] It should be noted that, in the embodiments of this application, the time associated with the event can be a point in time, such as 17:00 on February 28, 2025, or a time period, such as 19:00-21:00 on March 10, 2025.
[0065] The pedestrian flow prediction device can directly search for spatiotemporal text data associated with the first region, or it can search for such data through other devices (e.g., a search server), thus saving on processing resources. For example, a search server can be deployed to retrieve and store spatiotemporal text data associated with the first region from information published on social media and / or the internet, from which the pedestrian flow prediction device can obtain the associated spatiotemporal text data.
[0066] In some embodiments, in order to reduce data redundancy and improve the efficiency of pedestrian flow prediction, the pedestrian flow prediction device can also set time filtering conditions, and the acquired spatiotemporal text data can meet the time filtering conditions.
[0067] For example, the time filtering condition can be that the time recorded in the spatiotemporal text data is no earlier than the current time. Taking the current time as 10:00 on February 24, 2025 as an example, then the time recorded in the obtained spatiotemporal text data should be no earlier than 10:00 on February 24, 2025.
[0068] Alternatively, the time filtering condition can be that the time recorded in the spatiotemporal text data is no earlier than 24 hours before the current time. Taking the current time as 10:00 on February 24, 2025 as an example, then the time recorded in the obtained spatiotemporal text data should be no earlier than 10:00 on February 23, 2025.
[0069] Understandably, a role can also be called an identity, etc. A character can have one or more roles (or identities). For example, a character can be a bank employee or a student's parent.
[0070] In addition, the spatiotemporal text data associated with the first region obtained by the pedestrian flow prediction device may be multiple, and multiple spatiotemporal text data can correspond to multiple events. This application embodiment does not limit the number of spatiotemporal text data obtained by the pedestrian flow prediction device.
[0071] For example: the spatiotemporal text data acquired by the pedestrian flow prediction device includes: spatiotemporal text data 1 and spatiotemporal text data 2. Spatiotemporal text data 1 includes event 1 and the time, location and role associated with event 1; spatiotemporal text data 2 includes event 2 and the time, location and role associated with event 2.
[0072] S202: The pedestrian flow prediction device determines the spatiotemporal entity association information of the first area based on the spatiotemporal text data associated with the first area. This spatiotemporal entity association information includes the relationships between four types of entities: roles, time, location, and events.
[0073] After acquiring spatiotemporal text data, the pedestrian flow prediction device can extract the relationships between four types of entities: roles, time, location, and events, thereby determining the spatiotemporal entity relationship information.
[0074] For example, after acquiring spatiotemporal text data, the pedestrian flow prediction device can call a first model to extract the relationships between four types of entities—roles, time, location, and events—from the spatiotemporal text data, thereby generating spatiotemporal entity relationship information. The first model can be a large model, such as a large language model or a multimodal large model. The first model can be trained (or pre-trained) based on a large number of spatiotemporal text data samples and the relationship samples of the four types of entities (roles, time, location, and events) corresponding to those samples, and can support the extraction of relationships between these four types of entities based on spatiotemporal text data.
[0075] As an example, the spatiotemporal text data includes: School A held a general meeting of all teachers and students on March 9, 2025, from 16:00 to 17:35. Based on this spatiotemporal text data, the event can be extracted as: General Meeting of All Teachers and Students, Time: March 9, 2025, from 16:00 to 17:35, Location: School A, Roles: Teachers and Students. The above event, time, location, and roles are interrelated.
[0076] In some embodiments, to further improve the accuracy of the first model in extracting the relationships between the four types of entities—roles, time, location, and events—prompt words can be used to guide the first model in extracting the relationships between these four types of entities.
[0077] Here is a possible prompt: Please extract the relationships between four types of entities—role, time, location, and event—based on the spatiotemporal text data. A sample extraction example: Celebrity XX held a concert at Stadium C on January 1, 2024, from 19:00 to 21:00. The corresponding events are: Concert, Time: January 1, 2024, 19:00-21:00, Location: Stadium C, Roles: Celebrity, Student, Enterprise / Institution Staff, Security Personnel, etc. These events, times, locations, and roles are interconnected. Note: In this step, your task is to extract the four types of entities—role, time, location, and event—and the relationships between them. The resulting data should have a similar structure to the previous example, but it does not need to be exactly the same.
[0078] After the crowd flow prediction device acquires the spatiotemporal text data, in addition to inputting the spatiotemporal text data into the first model, it can also input the prompt word into the first model. The prompt word guides the first model to extract the relationship between the four types of entities: role, time, location, and event from the spatiotemporal text data.
[0079] In some embodiments, the first model may also learn or store descriptive information of each role in order to accurately extract spatiotemporal text data, including the roles associated with the event. The following are examples of descriptive information for some roles provided in embodiments of this application:
[0080] Students: Your travel pattern is usually regular, mainly consisting of traveling between school or educational institutions and home. However, you may also have part-time work, internships, or club activities, which will add extra travel time and may result in irregular schedules.
[0081] Teachers: Your travel patterns typically depend on the location of your workplace and your shift schedule. For example, you may attend teacher training, academic conferences, or other educational activities daily, which can affect your travel frequency and destinations.
[0082] For workers: Your travel patterns typically depend on the location of your workplace and your shift schedule. For example, you may need to commute to a factory, construction site, or office every day, or move between different locations as needed. You may also have overtime or shift work, which will change your travel time.
[0083] Doctors: As a doctor, your travel patterns typically involve traveling between hospitals, clinics, or other medical facilities. You may also frequently travel for business due to emergencies or conferences, even after night shifts. Furthermore, attending medical conferences or training may increase international or long-distance travel.
[0084] Retirees: Your travel patterns are typically quite flexible, depending on your personal interests and schedule. You might visit parks, gyms, libraries, or other community centers daily or every few days. Occasionally, you might take trips or visit relatives and friends, which are usually decided on a whim.
[0085] For office workers: Your travel pattern mainly revolves around commuting, typically traveling between home and the office within a fixed timeframe. You may also need to travel for business or attend work-related meetings, which will increase irregular travel schedules. In addition, working overtime or returning home late will also affect your daily travel patterns.
[0086] In some embodiments, considering that different locations correspond to different geographical features, such as different buildings, different areas of interest (AOI), or different points of interest (POI), the pedestrian flow prediction device can also determine spatiotemporal entity association information based on spatiotemporal text data and the geographical information of the first region.
[0087] For example, the first model is pre-trained based on spatiotemporal text data and geographic information samples, as well as the relationship samples of the four types of entities corresponding to the spatiotemporal text data samples: roles, time, location, and events. It supports determining the relationship of the four types of entities based on spatiotemporal text data and geographic information.
[0088] The geographic information of the first region can be pre-configured in the pedestrian flow prediction device, or it can be obtained by the pedestrian flow prediction device from a map server. For example, the pedestrian flow prediction device can send an identifier of the first region to the map server, and the map server can send the geographic information of the first region to the pedestrian flow prediction device after receiving the identifier. In this embodiment, the method by which the pedestrian flow prediction device obtains the geographic information of the first region is not limited.
[0089] The geographic information of the first area may include at least one of the following: street view data, remote sensing data, or base map data of the first area. Street view data may include multimedia information such as photos and videos of various geographical locations. Remote sensing data may be geographic imagery from satellites or drones, containing detailed ground information. Base map data may be structured geographic information system (GIS) data, such as building layouts and road networks.
[0090] Taking the spatiotemporal entity association diagram (or spatiotemporal entity relationship diagram) as an example, Figure 3 shows a possible spatiotemporal entity association diagram provided by an embodiment of this application. It includes population distribution (or role distribution), such as students, teachers, workers, doctors, retirees, office workers, etc.; events, such as summer vacation, winter vacation, open day, registration day, sports meet, conference, National Day, etc.; activity patterns, such as going to work, attending class, seeking medical treatment, traveling, etc., where activity patterns can also be categorized as events; buildings, such as parks, squares, schools, residences, shopping malls, banks, hospitals, etc.; and maps, such as street views, road networks, POIs, AOIs, etc., where maps or buildings can be used to represent locations. The arrows in Figure 3 represent association relationships. Taking the role of "student" as an example, the locations associated with this role include schools and a certain POI on the map; the events associated with this role include attending class and winter vacation; and the roles associated with this role include teachers. It is understood that the spatiotemporal entity association diagram can also include time associated with events, which is not shown in Figure 3.
[0091] S203: The pedestrian flow prediction device determines the pedestrian flow in the area to be predicted during the period to be predicted based on the spatiotemporal entity association information of the first area, the time period to be predicted, and the location information of the area to be predicted, using a pedestrian flow prediction model. The area to be predicted belongs to the first area.
[0092] After determining the spatiotemporal entity association information, the pedestrian flow prediction device can call the pedestrian flow prediction model. The spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted are input into the pedestrian flow prediction model for processing, and the pedestrian flow in the area to be predicted during the time period output by the model are obtained.
[0093] In one possible implementation, the pedestrian flow prediction model can be a large model. The pedestrian flow prediction model can be pre-trained based on a large amount of spatiotemporal entity association information, the time period to be predicted, the location information samples of the area to be predicted, and the corresponding pedestrian flow samples. It can support outputting the pedestrian flow of the area to be predicted during the time period based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted.
[0094] The time period to be predicted can refer to the time period during which pedestrian flow prediction needs to be performed.
[0095] The area to be predicted can refer to the spatial location where pedestrian flow prediction is required.
[0096] The area to be predicted belongs to the first area, or it can be a subset of the first area. Alternatively, the first area can be used as the area to be predicted, or a portion of the first area can be used as the area to be predicted. The area to be predicted and the time period to be predicted can be pre-configured in the pedestrian flow prediction device, or determined by the pedestrian flow prediction device in response to a pedestrian flow prediction command input by the user. This command can specify the area to be predicted and the time period to be predicted, or it can be determined by the pedestrian flow prediction device according to a pre-configured pedestrian flow prediction strategy, etc., where the prediction strategy can be used to determine the area to be predicted and the time period to be predicted.
[0097] As an example, the user-input pedestrian flow prediction command includes the time period to be predicted (e.g., 10:00-11:00 on March 1, 2025) and the identifier (e.g., name) of the area to be predicted. After receiving the command, the pedestrian flow prediction device can send the identifier of the area to be predicted to the map server and request its location information. After obtaining the location information from the map server, the device can input the spatiotemporal entity association information, the time period to be predicted, and the location information into the pedestrian flow prediction model for processing, thus obtaining the pedestrian flow in the predicted area during the predicted time period.
[0098] The location information of the area to be predicted may include the latitude and longitude of the area, the name of the location, and / or other geographic attributes that can uniquely identify the location (such as the location of buildings and / or the road layout in the area to be predicted).
[0099] It is important to understand that if the pedestrian flow prediction device has the location information of the area to be predicted, or if the pedestrian flow prediction command entered by the user includes the location information of the area to be predicted, the pedestrian flow prediction device can directly obtain the location information of the predicted area without having to obtain the location information of the area to be predicted through a map server.
[0100] In one possible implementation, when the area to be predicted is the same as the first area, the location information of the area to be predicted may not be input into the pedestrian flow prediction model. That is, the spatial location for pedestrian flow prediction is not restricted. In this case, since the spatiotemporal entity association information is determined based on the spatiotemporal text data associated with the first area, the pedestrian flow in the time period to be predicted is the pedestrian flow data of the first area in the time period to be predicted.
[0101] In some embodiments, to improve the accuracy of pedestrian flow prediction, the pedestrian flow prediction model may also learn or store spatiotemporal context information. This spatiotemporal context information may include the population size and activity characteristics of various groups, and / or historical movement trajectory data of pedestrian flow, etc. The spatiotemporal context information may be specific to a first region or general to all regions; this application does not limit this.
[0102] Taking the spatiotemporal context information including the population size and activity characteristics of various groups as an example, the following are possible examples of population size and activity characteristics for some population categories:
[0103] The population category is "maintenance and cleaning personnel," and the activity characteristics are "regular or scheduled entry and exit, working according to maintenance plans or sudden needs, usually during the day or when residents are out; carrying maintenance or cleaning tools, working in public and private areas," and the population size is "medium, arranged as needed, usually with a relatively fixed number of people per day."
[0104] The group is categorized as "security personnel," and their activities are characterized by "continuous patrolling and monitoring, operating around the clock within the community, managing entrances and exits and vehicle access, handling emergencies, and maintaining community order and security." The group size is described as "medium-sized, with a fixed number of personnel, typically working in daily shifts."
[0105] The population category is "children and teenagers, or students". The activity characteristics are "going to and from school, regular entry and exit times, using playground equipment, playing and exercising in children's playgrounds or sports venues, participating in community activities, interest classes or sports competitions". The population size is "medium, usually accounting for 15-25% of the population, depending on the family structure of the residential area".
[0106] The population category is "elderly people," and their activity characteristics are "walking and socializing, engaging in activities within the community, participating in community activities such as Tai Chi, aerobics, and reading groups, and frequently engaging in activities in fixed public areas such as gardens and squares." The population size is "medium, typically accounting for 10-15% of the population, depending on the family structure of the residential area."
[0107] Taking the historical movement trajectory data of people in the first region included in the spatiotemporal context information as an example, the following are some possible examples of historical movement trajectory data of people:
[0108] Historical movement trajectory of people 1: Person 1, role: a sixth-grade student who leaves community D at 7:10 on weekdays to go to school, travels between community D and school A on road C from 7:10 to 7:30 on weekdays, buys breakfast at breakfast shop E during this period, studies at school A from 7:30 to 17:00 on weekdays, travels between school A and community D on road C from 17:00 to 17:20 on weekdays, and usually studies, eats, and rests at home in community D after 17:20 on weekdays.
[0109] Historical movement trajectory of people 2: Person 2, role: Breakfast shop owner, leaves home in community D at 3:10 on weekdays and heads towards vegetable market E, arrives at vegetable market F at 3:40 on weekdays to buy breakfast ingredients, leaves vegetable market F at 4:00 on weekdays and heads towards breakfast shop E, arrives at breakfast shop E at 4:20 on weekdays to make and sell breakfast, etc.
[0110] In some embodiments, the pedestrian flow prediction device determines the pedestrian flow in the area to be predicted during the time period through a pedestrian flow prediction model. The pedestrian flow prediction model is based on spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted. The output can also be multiple pedestrian flow data points for the area to be predicted during the time period. Each pedestrian flow data point includes a person and the corresponding role, time, location, and event. Different pedestrian flow data points correspond to different people.
[0111] Pedestrian flow prediction devices can obtain the pedestrian flow rate of a region within a predicted time period based on multiple pedestrian flow data points. For example, the pedestrian flow rate can be obtained by counting the number of pedestrian flow data points within a region within a predicted time period. Similarly, pedestrian flow prediction models can be pre-trained based on a large amount of spatiotemporal entity association information, the predicted time period, the location information samples of the predicted region, and corresponding multiple pedestrian flow data samples. This allows for the output of multiple pedestrian flow data points for a region within a predicted time period, based on spatiotemporal entity association information, the predicted time period, and the location information of the predicted region.
[0112] Taking school A as the area to be predicted and the time to be predicted as 16:00-17:00 on March 9, 2025, multiple pedestrian flow data can include the following personnel data:
[0113] Personnel data 1: Person 1, role student, attending the general meeting of all teachers and students in the school's A auditorium from 16:00 to 17:00 on March 9, 2025.
[0114] Personnel data 2: Character 2, breakfast shop owner, attended the general meeting of all teachers and students in the school's A auditorium from 16:00 to 17:00 on March 9, 2025.
[0115] Personnel data 3: Person 3, role student, attending the general meeting of all teachers and students in the school's A auditorium from 16:00 to 17:00 on March 9, 2025.
[0116] Personnel data 4: Person 4, Role Teacher, attending the general meeting of all teachers and students in the school's A auditorium from 16:00 to 17:00 on March 9, 2025.
[0117] The pedestrian flow prediction device can obtain the pedestrian flow of the area to be predicted during the predicted time period based on multiple pedestrian flow data of the area to be predicted during the predicted time period. For example, by counting the number of pedestrian flow data of the area to be predicted during the predicted time period, the pedestrian flow of the area to be predicted during the predicted time period can be determined.
[0118] In one possible implementation, to improve the accuracy of pedestrian flow prediction, the pedestrian flow prediction device can also perform absolute and / or relative verification on the obtained pedestrian flow data (such as multiple pedestrian flow data of the area to be predicted in the time period to be predicted output by the pedestrian flow prediction model above), and discard the pedestrian flow data that fails the verification, thereby improving the accuracy of pedestrian flow prediction.
[0119] Absolute verification can refer to verifying the rationality of a single piece of pedestrian flow data and determining the rationality score of the single piece of pedestrian flow data. If the rationality score is less than or equal to the first threshold, the pedestrian flow prediction device can discard the pedestrian flow data.
[0120] Taking the absolute verification of the first flow data as an example, the first flow data belongs to the flow data obtained by the flow prediction device. The flow prediction device can call the first scoring model, input the first flow data into the first scoring model for processing, and obtain the rationality score of the first flow data. If the rationality score of the first flow data is less than or equal to the first threshold, it means that the role corresponding to the first flow data is unreasonable in performing the corresponding event at the corresponding time and place, and the flow prediction device can discard the first flow data.
[0121] For example: Person 1, role student, attends the general meeting of teachers and students in the school's A auditorium from 16:00 to 17:00 on March 9, 2025. The reasonableness score is 95, which is greater than the first threshold of 85. Person 1 can be retained.
[0122] Person Flow Data 2: Character 2, breakfast shop owner, attended the general assembly of teachers and students in the school's A auditorium from 16:00 to 17:00 on March 9, 2025. The reasonableness score is 50, which is less than the first threshold of 85. Person Flow Data 2 is discarded.
[0123] The first scoring model can be pre-trained based on a large number of pedestrian flow data samples and corresponding reasonableness score samples, and can support outputting a reasonableness score for the pedestrian flow data based on the input pedestrian flow data.
[0124] Additionally, it is understandable that if the rationality score of the first flow data is less than or equal to the first threshold, the flow prediction device can discard the first flow data. Alternatively, it can be replaced with: if the rationality score of the first flow data is less than the first threshold, the flow prediction device can discard the first flow data.
[0125] Relative verification can refer to verifying the differences between two sets of pedestrian flow data with different corresponding roles, determining the difference score between the two sets of pedestrian flow data with different corresponding roles, and discarding the two sets of pedestrian flow data with different roles if the difference score is less than or equal to a second threshold.
[0126] Taking the relative verification of the second and third pedestrian flow data as an example, the second and third pedestrian flow data correspond to different roles and both belong to the pedestrian flow prediction device. The pedestrian flow prediction device can call the second scoring model, input the second and third pedestrian flow data into the second scoring model for processing, and obtain the difference score between the second and third pedestrian flow data. If the difference score is less than or equal to the second threshold, it means that the two pedestrian flow data are very similar, but the roles they correspond to are different, and the small difference is unreasonable. The pedestrian flow prediction device can discard the second and third pedestrian flow data in the obtained pedestrian flow data. If the difference score is greater than the second threshold, the pedestrian flow prediction device can retain the second and third pedestrian flow data in the obtained pedestrian flow data.
[0127] The second scoring model can be pre-trained based on a large number of pedestrian data samples consisting of two pedestrian data points corresponding to different roles, and differential scoring samples corresponding to the pedestrian data samples. It can support outputting differential scores based on the input of two pedestrian data points corresponding to different roles.
[0128] It should be noted that if relative and absolute verifications are performed on the pedestrian flow data obtained by the pedestrian flow prediction device, the device can perform relative verification first and then absolute verification, or it can perform absolute verification first and then relative verification, or it can perform relative and absolute verifications simultaneously. This application does not limit this.
[0129] By using relative and / or absolute verification, unreasonable pedestrian flow data obtained from pedestrian flow prediction devices can be discarded, resulting in high-quality pedestrian flow data and thus improving the accuracy of pedestrian flow prediction.
[0130] In some embodiments, high-quality pedestrian flow data can be obtained by removing unreasonable pedestrian flow data, and the pedestrian flow prediction model can be trained to fine-tune the parameters of the pedestrian flow prediction model. This enables the self-optimization and iteration of the pedestrian flow prediction model, which is beneficial to improving the accuracy of pedestrian flow prediction. After the pedestrian flow prediction equipment or model is migrated across regions, it can be better adapted to the pedestrian flow prediction environment of the migrated region, which is beneficial to improving the migration and expansion capabilities of the pedestrian flow prediction equipment or model across regions.
[0131] For example, the pedestrian flow prediction device can use multiple pedestrian flow data obtained after relative and / or absolute verification (i.e., multiple pedestrian flow data used to determine pedestrian flow) as the expected output of the pedestrian flow prediction model, and use the corresponding spatiotemporal entity association map, the time period to be predicted, and the location information of the area to be predicted as the input of the pedestrian flow prediction model to train the pedestrian flow prediction model, adjust the parameters of the pedestrian flow prediction model, and achieve optimization and adjustment of the pedestrian flow prediction model.
[0132] Referring to the schematic diagram of the pedestrian flow prediction process shown in Figure 4, the pedestrian flow prediction method in this embodiment may include four parts: spatiotemporal perception, pedestrian flow generation, verification and correction, and spatiotemporal fine-tuning. Spatiotemporal perception can refer to generating spatiotemporal entity association information based on the geographic information of a first region and the spatiotemporal text data associated with the first region; pedestrian flow generation can refer to generating multiple pedestrian flow data through a pedestrian flow prediction model based on the location information of the region to be predicted, the time period to be predicted (i.e., a specific spatiotemporal period), and the spatiotemporal entity association information; verification and correction can refer to correcting the pedestrian flow data through absolute and relative verification, deleting unreasonable pedestrian flow data, and determining the pedestrian flow; spatiotemporal fine-tuning can refer to training the pedestrian flow prediction model and fine-tuning the parameters of the pedestrian flow prediction model using the verified and corrected pedestrian flow data and the spatiotemporal entity association information corresponding to the pedestrian flow data.
[0133] Taking the application of pedestrian flow prediction to antenna optimization and the pedestrian flow prediction equipment as maintenance equipment as an example, Figure 5 is a schematic diagram of an antenna optimization process based on pedestrian flow prediction provided in an embodiment of this application. The process includes:
[0134] S501: The maintenance equipment obtains the geographic information of the first area from the map server.
[0135] The geographic information may include at least one of the following: street view data, remote sensing data, or base map data of the first area.
[0136] S502: The maintenance equipment obtains the spatiotemporal text data associated with the first region from the search server.
[0137] Spatiotemporal text data includes information about events and the time, location, and roles associated with those events.
[0138] It should be noted that this application does not limit the order in which the maintenance equipment performs steps S501 and S502. The maintenance equipment may perform step S501 first and then step S502; or it may perform step S502 first and then step S501; or it may perform steps S501 and S502 simultaneously.
[0139] In one possible implementation, the maintenance equipment can acquire the geographic information of the first region and / or the spatiotemporal text data associated with the first region according to a set period. The period for acquiring the geographic information of the first region and the period for acquiring the spatiotemporal text data associated with the first region can be the same or different.
[0140] For example, the maintenance equipment is used to optimize the antennas in the first area. The maintenance equipment can obtain the spatiotemporal text data associated with the first area from the search service on a 24-hour cycle, and obtain the geographic information of the first area from the map server on a 72-hour cycle.
[0141] In another possible implementation, the geographic information of the first area and / or the spatiotemporal text data associated with the first area can also be obtained by the operation and maintenance equipment after receiving the pedestrian flow prediction instruction. The pedestrian flow prediction instruction can come from the user or from other devices.
[0142] For example, the maintenance equipment is used to optimize the antennas in the first area. After receiving the pedestrian flow prediction instruction from the user, the maintenance equipment can obtain the spatiotemporal text data associated with the first area from the search server and the geographic information of the first area from the map server.
[0143] It should be noted that if the geographical information of the first region is pre-configured in the maintenance equipment, step S501 can be omitted.
[0144] S503: The maintenance equipment determines the spatiotemporal entity association information based on geographic information and spatiotemporal text data.
[0145] Among them, the spatiotemporal entity association information includes the association relationships between four types of entities: roles, time, location, and events.
[0146] S504: The operation and maintenance equipment determines multiple pedestrian flow data for the area to be predicted within the predicted time period based on the spatiotemporal entity association information, the location information of the area to be predicted, and the time period to be predicted, using a pedestrian flow prediction model.
[0147] In one possible implementation, if the maintenance equipment does not store the location information of the area to be predicted, it can request the location information of the area to be predicted from the map server by sending the identifier of the area to be predicted. The location information of the area to be predicted may include the latitude and longitude of the area, the location name, and / or other uniquely identifiable geographical attributes (such as the location of buildings and / or road layout in the area).
[0148] S505: The operation and maintenance equipment performs pedestrian flow data verification and correction to determine the pedestrian flow in the area to be predicted during the time period to be predicted.
[0149] The verification and correction of pedestrian flow data includes absolute and / or relative verification. Absolute verification involves validating the reasonableness of a single pedestrian flow data point and determining its reasonableness score. If the reasonableness score is less than or equal to a first threshold, the maintenance equipment can discard that pedestrian flow data. Relative verification involves verifying the differences between two pedestrian flow data points corresponding to different roles and determining their difference score. If the difference score is less than or equal to a second threshold, the maintenance equipment can discard these two pedestrian flow data points corresponding to different roles.
[0150] After verifying and correcting the multiple pedestrian flow data obtained in step S504 and discarding unreasonable pedestrian flow data, the operation and maintenance equipment can determine the pedestrian flow in the area to be predicted during the time period based on the amount of data of the verified and corrected pedestrian flow data.
[0151] S506: The maintenance equipment sends dynamic decision-making plans to the digital antenna based on the flow of people.
[0152] For example, if maintenance personnel determine that the pedestrian flow in the area to be predicted exceeds a set threshold during the predicted time period (i.e., a specific time and space), the resulting decision-making scheme could be to adjust the orientation of some antennas in the digital antenna, so that more antennas are facing the specific area where pedestrian flow is predicted. After receiving this dynamic decision-making scheme, the digital antenna can adjust the orientation of some antennas in the digital antenna, so that more antennas are facing the area where pedestrian flow is predicted.
[0153] S507: The operation and maintenance equipment trains the pedestrian flow prediction model based on multiple pedestrian flow data (i.e., the pedestrian flow data after verification and correction) used to determine pedestrian flow, as well as spatiotemporal entity association information, the location information of the time period to be predicted and the area to be predicted.
[0154] For example, the operation and maintenance equipment can use multiple pedestrian flow data obtained after relative and / or absolute verification (i.e., multiple pedestrian flow data used to determine pedestrian flow) as the expected output of the pedestrian flow prediction model. The corresponding spatiotemporal entity association graph, the time period to be predicted, and the location information of the area to be predicted are used as the input of the pedestrian flow prediction model to train the pedestrian flow prediction model, adjust the parameters of the pedestrian flow prediction model, and achieve optimization and adjustment of the pedestrian flow prediction model.
[0155] The implementation of each step in the antenna optimization process diagram based on pedestrian flow prediction shown in Figure 5 above can be referred to the implementation of each step in the pedestrian flow prediction method shown in Figure 2 above. The repeated parts will not be described again.
[0156] The above mainly describes the solution provided in this application from the perspective of method and process. It is understood that, in order to achieve the above functions, the apparatus may include hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0157] Figure 6 is a schematic diagram of a possible device provided in an embodiment of this application. The device may include units or modules corresponding to all or part of the steps in the above method embodiments, and can be used to execute the steps in the above embodiments. For details, please refer to the relevant descriptions in the above method embodiments.
[0158] As shown in Figure 6, the device 600 includes an input / output module 610 and a processing module 620, wherein the processing module 620 can be a processor or a processing circuit, and the input / output module 610 can be a transceiver unit or an input / output interface. The device 600 can be used to implement the steps performed by the crowd flow prediction device in the above embodiments.
[0159] When device 600 is used to implement the steps performed by the crowd flow prediction device in the above embodiments:
[0160] The input / output module 610 is used to acquire the spatiotemporal text data associated with the first region. The spatiotemporal text data includes information about events and the time, location, and roles associated with the events.
[0161] The processing module 620 is used to determine the spatiotemporal entity association information of the first region based on the spatiotemporal text data. The spatiotemporal entity association information includes the association relationship between four types of entities: role, time, location and event. Based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the region to be predicted, the flow of people in the region to be predicted during the time period is determined by the flow prediction model.
[0162] In one possible design, the input / output module 610 is also used to acquire geographic information of the first region, which includes at least one of street view data, remote sensing data, or base map data of the first region; when the processing module 620 determines the spatiotemporal entity association information of the first region based on the spatiotemporal text data, it is specifically used to determine the spatiotemporal entity association information of the first region based on the spatiotemporal text data and the geographic information of the first region.
[0163] In one possible design, when the processing module 620 determines the pedestrian flow in the area to be predicted during the predicted time period based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted using a pedestrian flow prediction model, it is specifically used to determine multiple pedestrian flow data points in the area to be predicted during the predicted time period based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted using a pedestrian flow prediction model. Each pedestrian flow data point includes a person and the person's corresponding role, time, location, and event; and determines the pedestrian flow based on the multiple pedestrian flow data points.
[0164] In one possible design, the processing module 620 is also used to determine the rationality score of the first flow data, wherein the first flow data belongs to multiple flow data; if the rationality score of the first flow data is less than or equal to a first threshold, the first flow data is discarded.
[0165] In one possible design, the processing module 620 is further configured to determine the difference score between the second flow data and the third flow data, wherein the second flow data and the third flow data belong to multiple flow data and the roles corresponding to the second flow data and the third flow data are different; if the difference score between the second flow data and the third flow data is less than or equal to a second threshold, the second flow data and the third flow data are discarded.
[0166] In one possible design, the processing module 620 is also used to train the pedestrian flow prediction model based on multiple pedestrian flow data used to determine pedestrian flow, the location information of the area to be predicted, the time period to be predicted, and the spatiotemporal entity association information.
[0167] In one possible design, when the input / output module 610 acquires the spatiotemporal text data associated with the first region, it is specifically used to acquire the spatiotemporal text data associated with the first region based on social media and / or the Internet.
[0168] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods are possible. Furthermore, the functional modules in each embodiment of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.
[0169] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a device (which may be a personal computer, network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0170] Another apparatus is also provided in this application embodiment. Figure 7 exemplarily shows an apparatus that may include a memory 701, a processor 702, and a bus 703, wherein the memory 701 and the processor 702 are interconnected via the bus 703. The processor 702 can read programs or instructions stored in the memory 701 to implement the crowd flow prediction method described in the above embodiments.
[0171] This application also provides a computer-readable storage medium storing a computer program or instructions for implementing the above-described pedestrian flow prediction method.
[0172] For example, when the computer program or instructions are executed by the processor, the processor can implement the above-mentioned method for predicting pedestrian flow.
[0173] This application also provides a computer program product containing a computer program or instructions, which, when executed by a processor, causes the processor to implement the above-described pedestrian flow prediction method.
[0174] In this application embodiment, the processor can be one or more central processing units (CPUs). If the processor is a CPU, it can be a single-core CPU or a multi-core CPU. The processor can also be one or a combination of several of the following: CPU, general-purpose processor, application-specific integrated circuit (ASIC), digital signal processor, microprocessor, microcontroller, graphics processor, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, artificial intelligence processor, or neural network processor. The processor can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.
[0175] In this application embodiment, the memory may include, but is not limited to, cache, read-only memory (ROM), random access memory, synchronous dynamic random access memory, hard disk or solid-state drive, erasable programmable read-only memory, or compact disc read-only memory (CD-ROM), etc. Memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited to this. The memory in this application embodiment may also be a circuit or any other device capable of implementing storage functions for storing computer programs or instructions, and / or data.
[0176] It is understood that the method steps in the embodiments of this application can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Additionally, the ASIC can reside in a network device or terminal. Alternatively, the processor and storage medium can exist as discrete components in the network device or terminal.
[0177] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are performed entirely or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user equipment, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one network device, terminal, computer, server, or data center to another network device, terminal, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video optical disc; or it can be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both types of storage media.
[0178] In the various embodiments of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of different embodiments are consistent and can be referenced by each other. The technical features of different embodiments can be combined to form new embodiments according to their inherent logical relationship.
[0179] Additionally, it should be understood that in the embodiments of this application, the term "exemplary" is used to indicate that it is an example, illustration, or description. Any embodiment or design scheme described as "exemplary" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplary" is intended to present the concept in a concrete manner.
[0180] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. The order of the process numbers described above does not imply the order of execution; the execution order of each process should be determined by its function and internal logic.
Claims
1. A method for predicting pedestrian flow, characterized in that, include: Obtain spatiotemporal text data associated with the first region, wherein the spatiotemporal text data includes events and information about the time, location, and roles associated with the events; Based on the spatiotemporal text data, determine the spatiotemporal entity association information of the first region. The spatiotemporal entity association information includes the association relationships between four types of entities: role, time, location, and event. Based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted, the pedestrian flow in the area to be predicted during the time period is determined by the pedestrian flow prediction model, wherein the area to be predicted belongs to the first area.
2. The method as described in claim 1, characterized in that, The method further includes: Obtain geographic information of the first area, wherein the geographic information includes at least one of street view data, remote sensing data, or base map data of the first area; The step of determining the spatiotemporal entity association information of the first region based on the spatiotemporal text data includes: Based on the spatiotemporal text data and the geographical information of the first region, the spatiotemporal entity association information is determined.
3. The method as described in claim 1 or 2, characterized in that, The step of determining the pedestrian flow in the area to be predicted during the period to be predicted, based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted, using a pedestrian flow prediction model, includes: Based on the spatiotemporal entity association information, the time period to be predicted, and the location information of the area to be predicted, multiple pedestrian flow data points for the area to be predicted during the time period are determined by the pedestrian flow prediction model. Each pedestrian flow data point includes a person and the person's corresponding role, time, location, and event. The pedestrian flow is determined based on the multiple pedestrian flow data.
4. The method as described in claim 3, characterized in that, The method further includes: Determine the reasonableness score of the first flow data, wherein the first flow data belongs to the multiple flow data; If the rationality score of the first pedestrian flow data is less than or equal to the first threshold, the first pedestrian flow data is discarded.
5. The method as described in claim 3 or 4, characterized in that, The method further includes: Determine the difference score between the second flow data and the third flow data, wherein the second flow data and the third flow data belong to the multiple flow data, and the roles corresponding to the second flow data and the third flow data are different; If the difference score between the second flow data and the third flow data is less than or equal to the second threshold, the second flow data and the third flow data are discarded.
6. The method as described in claim 4 or 5, characterized in that, The method further includes: The pedestrian flow prediction model is trained based on multiple pedestrian flow data used to determine the pedestrian flow, the location information of the area to be predicted, the time period to be predicted, and the spatiotemporal entity association information.
7. The method according to any one of claims 1-6, characterized in that, The acquisition of the spatiotemporal text data associated with the first region includes: Spatiotemporal text data associated with the first region is obtained based on social media and / or the Internet.
8. An apparatus, characterized in that, Includes modules or units for performing the method as described in any one of claims 1-7.
9. An apparatus, characterized in that, include: Processor and memory; The memory stores computer programs or instructions; The processor is configured to invoke a computer program or instruction stored in the memory to execute the method as described in any one of claims 1-7.
10. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, cause the method as described in any one of claims 1-7 to be implemented.
11. A computer-readable storage medium, characterized in that, The storage medium stores a computer program or instructions that, when executed by a processor, cause the method as described in any one of claims 1-7 to be implemented.
