Human flow estimation device and human flow estimation method

Abstract

To estimate a human flow in consideration of the interest of a user.SOLUTION: A human flow estimation device includes: a preprocessing unit for converting a predetermined region into grids, and creating human flow data obtained by time-series shaping movement information on a user in the grid region; an extraction unit for extracting a digital persona for a region corresponding to the human flow data, including a predetermined data source of a user, using a previously trained persona estimation model; an integration unit for creating extended human flow data obtained by integrating the human flow data and the extracted digital persona by using a predetermined matching method using each feature amount; and an estimation unit for estimating a human flow of the region, using a simulation method for estimating the human flow by simulation, with the extended human flow data as input.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a people flow estimation device and a people flow estimation method. [Background technology]

[0002] Conventionally, there are techniques for simulating people flow.

[0003] For example, there is a technology related to a people flow visualization system that can provide information about people flow accurately according to the user's intentions (see Patent Document 1). This technology discloses that data including the location, unique identifier, and time of a mobile terminal device is acquired, and people flow data indicating the movement of the person is generated based on the data, and displayed as a trajectory on a map.

[0004] There is also a technology that allows users to add field observation data to field data, thereby supporting more precise behavioral pattern analysis (see Patent Document 2). This technology discloses that measurement data from sensors is acquired and aggregated to generate field data, observation data based on field observations by users is input, and integrated data is generated by integrating the observation data with the field data.

[0005] There is also a technology related to an information presentation device for behavior control (see Patent Document 3). This technology discloses a feature in which group behavior information includes information on the movement speed and flow rate of people as the behavior of people present in a measurement area. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2023-22623 [Patent Document 2] Japanese Patent Application Publication No. 2020-107171 [Patent Document 3] Patent Publication No. 2018-229810 Summary of the Invention [Problem to be solved by the invention]

[0007] Previously, mass human traffic simulation technology modeled the flow of people in specific areas using people's movement data, such as mobile phone location information. However, the attribute information of the simulated people flow data mainly focused on users' demographic information (gender, age, place of residence, etc.). Meanwhile, detailed information on individual users' interests and preferences, such as purchasing data or social media posts, was rarely integrated. As a result, the information that could be derived from the simulation was limited.

[0008] In consideration of the above, the present invention aims to enable estimation of people flow taking into account the interests of users. [Means for solving the problem]

[0009] In order to achieve the above object, the people flow estimation device of the present invention includes a pre-processing unit that grids a specified area and creates people flow data by shaping the movement information of users in the gridded area into a time series; an extraction unit that uses a pre-trained persona estimation model to extract a digital persona for the area corresponding to the people flow data, the digital persona being a digital persona including a specified data source of the user; an integration unit that creates extended people flow data by integrating the people flow data and the extracted digital persona using a specified matching method that uses the feature amounts of each; and an estimation unit that uses the extended people flow data as input and estimates people flow in the area using a simulation method for estimating people flow by simulation. [Effects of the Invention]

[0010] According to the present invention, it is possible to obtain an effect of enabling estimation of people flow taking into account the interests of users. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram showing the configuration of a people flow estimation device. [Figure 2] Figure 2 shows an example of a comparison between the conventional method and the simulation of this method. [Figure 3] Figure 3 shows gridded data for a specific area. [Figure 4] FIG. 4 is an example of user movement information used as people flow data. [Figure 5] FIG. 5 is an example of an image showing the movement of a user relative to a section of a region. [Figure 6] FIG. 6 is a flowchart showing the processing in the people flow estimation device. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0013] FIG. 1 is a block diagram showing the configuration of a pedestrian flow estimation device 100. As shown in FIG. 1, the pedestrian flow estimation device 100 includes a data storage unit 102, a preprocessing unit 110, an extraction unit 112, an integration unit 114, an estimation unit 116, and a presentation unit 118. The hardware configuration of the pedestrian flow estimation device 100 is realized by a computer including a CPU (Central Processing Unit), a ROM (Read Only Memory) storing programs for implementing each processing routine, a RAM (Random Access Memory) for temporarily storing data, a memory serving as a storage means, and a network interface. Note that a GPGPU or accelerator may be used instead of the CPU depending on the suitability of each process, and a computing unit appropriate for the process may be used as appropriate. In particular, the use of a GPGPU or accelerator is preferable for processes related to learning and inference. Note that the preprocessing unit 110 may be provided externally to the pedestrian flow estimation device 100 as an external device. The presentation unit 118 may be an external terminal to the pedestrian flow estimation device 100.

[0014] The data storage unit 102 stores map data including the area to be estimated, movement information of users in the area, information collected about users in the area, a pre-trained persona estimation model, and various information required for simulations to estimate people flow.

[0015] Figure 2 shows an example of a comparison between a conventional simulation and the simulation of this method. The simulation process of this embodiment roughly consists of the following three steps: (1) shaping people flow data, (2) merging people flow data with data from other sources (digital personas) (a unique processing process), and (3) simulating people flow.

[0016] A challenge with conventional technology is that the source data for people flow simulations is limited to a single set of standard demographic information. For example, when using mobile phone location data, while contract information provides data such as age, gender, and place of residence, information for inferring hobbies and preferences cannot be obtained. As a result, simulation results simply indicate the flow of people but do not provide deep insights into urban development and other areas. Therefore, the method of this embodiment uses digital persona technology to integrate people's location data, which is the basis for people flow simulations, with other data that reveals hobbies and preferences, thereby improving the interpretability of the information obtained from the simulation results. For example, there are countless people who visit Shinjuku on Saturdays at 11:00, but this information can be combined with data from other sources. This integration enables simulation results (or simulation inputs) to be given more precise persona images, such as "a resident of Kichijoji who visits a department store because he or she prefers luxury goods" or "a resident of Omiya who is hardworking and uses a cram school."

[0017] The preprocessing unit 110 grids map data for the area to be estimated and creates people flow data by shaping the movement information of users in the area into a time series. People flow data is basically a collection of location information data in chronological order. Continuous values ​​of time and latitude / longitude information are converted into discrete data that is easy to handle in analysis. For example, it performs operations such as time slicing into 15-minute intervals, or dividing the map into a grid and assigning grid names to the latitude and longitude. It also performs data cleansing, such as removing data that contains obvious positioning errors and data that interferes with analysis.

[0018] 3 shows data in which a specific area is gridded. In shaping the people flow data in the preprocessing unit 110, the transition of movement in the area is shaped in time series using the areas divided by the gridding.

[0019] FIG. 4 shows an example of user movement information used as people flow data. The user movement information is converted into schedule data on an individual / daily basis. This schedule data is then shaped into a time series. FIG. 5 shows an example of an image representing a user's movement within a region. In FIG. 5, numbers are assigned to the grid and the outer edge. The grid is padded around the observation outer edge of the region's region, and the out-of-observation time is interpolated. The time required to return to the observation area is interpolated using the shortest route along the outer edge. For example, the route shown in FIG. 5 provides time series movement data of "7 → 8 → 9 → 15 → 16 → 17 → NaN → NaN → NaN → 29 → 28 → 27 → 26." The preprocessing unit 110 aggregates this user movement data and creates shaped people flow data. Specifically, user movement information (location information) includes mobile phone GPS data, mobile phone connection base station data, automobile probe data, Wi-Fi (registered trademark) packet sensor data, and beacon data. In particular, for macro-level simulations of people flow, mobile phone-related data is used.

[0020] The extraction unit 112 extracts a digital persona for a region corresponding to the created people flow data using a persona estimation model that has been trained in advance. The data source includes information about users belonging to the region, such as demographic information (attribute information), purchase data at stores, usage information on facilities, preference information such as responses to a questionnaire on hobbies and preferences, and posted information on social media, etc.

[0021] We will now explain digital personas. Digital personas are a technology that combines multiple data sources, such as demographic information (statistical data), purchasing data, and social media posts, to accurately depict the interests and preferences of each user. For example, we use the digital persona extraction method described in Reference 1 below. [Reference 1] JP 2023-146836 A

[0022] The technology of Reference 1 uses, as an example, a persona estimation model using an MLP (Multilayer Perceptron) or the like. This technology is a method of applying a persona estimation model to spatial information indicating a region, and estimating information indicating a user's persona, such as behavioral information, purchasing information, and interest information, using attribute information (such as a person's age group, gender, and residential area) as input. In this embodiment, a trained digital persona is used that includes user information estimated using the same method as Reference 1 as a data source.

[0023] The integration unit 114 creates extended people flow data by integrating the created people flow data and the extracted digital persona using a predetermined matching method that uses the respective feature amounts.

[0024] Here is an example of a specific data fusion technique. Fusion is generally performed using a technique known as data fusion. One example is Mahalanobis matching, which uses the similarity of common items (covariates). This method calculates the similarity between the source record and the target record by accumulating the differences in the values ​​of each covariate item, and sequentially combines the most similar records. For example, purchase history data linked to a point card and movement data linked to a transportation IC card both contain information such as age, gender, and place of residence. This common information is used to match the data. Note that the data from other sources to be fused is clustered to ensure anonymity. For example, the purchase data items themselves are clustered, such as "23 years old, male, luxury-oriented."

[0025] Digital personas are also created as abstracted data using technology that precisely depicts the relationships between data. Using extraction technology, the data groups combined using the data fusion method described above are linked in a statistically valid manner with information such as gender, age, and place of residence, as well as purchasing patterns, behavioral patterns, or interest patterns extracted from social media. A set of preprocessing / semantic extraction technologies is used for each data group. Examples of such technologies include converting people flow data into station network data, abstracting purchasing data using pattern mining and clustering, and estimating user characteristics from social media data using natural language processing.

[0026] The estimation unit 116 receives the extended people flow data as input and estimates people flow in the area using a simulation method for estimating people flow through simulation.

[0027] As the simulation method, either the first simulation method or the second simulation method described below is used. Note that which method to use can be set in advance by receiving input from an administrator or user of the people flow estimation device 100. Also, people flow may be estimated using the two simulation methods and averaged.

[0028] The first simulation method is a method in which the augmented people flow data (or only the people flow data) is clustered, and the results of simulations performed for each cluster are finally integrated. Specific clustering methods include k-means, HDBSCAN, and distributed representation learning. Additionally, Markov models and cellular automata are assumed as simulation algorithms. When using the first simulation method, the augmented people flow data or only the people flow data included in the augmented people flow data are clustered, a clustering method is applied, simulations are performed for each cluster, and the results for each cluster are integrated to estimate the people flow in the area. As an example, when implemented using a Markov model (multiple Markov), the transition probability matrix is ​​calculated as P(Gj,t+1|Gi,t), and the movement probability between grid points from time t to t+1 is calculated.

[0029] The second simulation method incorporates the augmented people flow data as simulation variables without clustering it. Specifically, it is assumed that simulations will be performed using algorithms such as LSTM, VAE, GAN, and Transformer. When using the second simulation method, a simulation is performed by applying a method that incorporates each element of the augmented people flow data as a simulation variable, and people flow in the area is estimated.

[0030] The presentation unit 118 displays the estimated people flow on a map, and also displays information about the digital personas of the area.

[0031] Next, the operation of the embodiment of the present invention will be described. Fig. 6 is a flowchart showing the processing in the people flow estimation device 100. The CPU, GPGGPU, or accelerator arithmetic unit of the people flow estimation device 100 reads and executes programs and various data from the ROM, and the arithmetic unit performs the processing of each part of the people flow estimation device 100.

[0032] In step S100, the preprocessing unit 110 grids map data of the area to be estimated, and generates people flow data by shaping the movement information of users in the area into a time series.

[0033] In step S102, the extraction unit 112 uses a persona estimation model that has been trained in advance to extract a digital persona for the area corresponding to the created people flow data.

[0034] In step S104, the integration unit 114 creates extended people flow data by integrating the created people flow data and the extracted digital persona using a predetermined matching method that uses the respective feature amounts.

[0035] In step S106, the estimation unit 116 receives the extended people flow data as input and estimates people flow in the area using a preset simulation method.

[0036] In step S108, the presentation unit 118 displays the estimated people flow on a map, and also displays information about the digital personas of the area.

[0037] As described above, the people flow estimation device according to the embodiment of the present invention makes it possible to estimate people flow taking into account the interests of users.

[0038] Furthermore, the technology according to this embodiment incorporates digital personas into people flow simulations, resulting in more insightful simulation results that take into account the interests and preferences of individual users. Furthermore, performing people flow simulations that respond to changes in users' hobbies and preferences can contribute to effective persona marketing. Furthermore, for urban planners and developers, understanding consumer needs, lifestyles, and preferences allows them to develop development plans that accommodate diversity. Understanding changes in people flow as individual users' behaviors and preferences change allows developers to carefully plan long-term visions and strategies.

[0039] The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit and scope of the present invention. [Explanation of symbols]

[0040] 100 People flow estimation device 102 Data storage unit 110 Pretreatment section 112 Extraction part 114 Integration Department 116 Estimation Department 118 Presentation section

Claims

1. a pre-processing unit that grids a predetermined area and generates people flow data by shaping user movement information of the gridded area into time series; an extraction unit that uses a pre-trained persona estimation model to extract a digital persona including a predetermined data source of a user, the digital persona being for an area corresponding to the people flow data; an integration unit that creates extended people flow data by integrating the people flow data and the extracted digital persona using a predetermined matching method that uses the respective feature amounts; an estimation unit that estimates people flow in the area using a simulation method for estimating people flow by simulation, using the extended people flow data as an input; A people flow estimation device comprising:

2. At least one of a first simulation method and a second simulation method is used as the simulation method, When the first simulation method is used, the estimation unit clusters the extended people flow data or only the people flow data included in the extended people flow data, applies a method using clusters, performs a simulation for each cluster, and integrates the results for each cluster to estimate the people flow in the area; 2. The people flow estimation device according to claim 1, wherein, when the second simulation method is used, the estimation unit performs a simulation by applying a method that incorporates each element of the extended people flow data as a simulation variable, and estimates people flow in the area.

3. The people flow estimation device of claim 1, wherein the data source of the digital persona extracted by the extraction unit includes at least predetermined demographic information, predetermined purchasing data, predetermined usage information, predetermined preference information, and predetermined posting information as information about users belonging to the area.

4. Further comprising a presentation unit, The people flow estimation device according to claim 1 , wherein the presentation unit displays an estimated result of the people flow on a map, and also displays information about the digital personas in the area.

5. A predetermined area is gridded, and people flow data is generated by shaping the movement information of users in the gridded area into a time series. extracting a digital persona for a region corresponding to the people flow data, the digital persona including a predetermined data source of the user, using a pre-trained persona estimation model; Creating extended people flow data by integrating the people flow data and the extracted digital persona using a predetermined matching method that uses the respective feature amounts; Estimating the flow of people in the area using a simulation method for estimating the flow of people by simulation, using the expanded people flow data as an input. A people flow estimation method in which processing is performed by a computer.

Images