Dynamic region division method based on movable coordinate data

Abstract

The invention discloses a dynamic region division method based on movable coordinate data, belongs to the field of big data processing, and aims to realize efficient processing and dynamic region identification of coordinate data. According to the method, firstly, activity coordinate data of a plurality of activity subjects in a target area within a certain time period are collected, abnormal values and noise interference are removed through data preprocessing, and data accuracy is ensured; thirdly, preliminarily dividing the activity area into a plurality of sub-areas according to distribution characteristics of coordinates, performing sub-area fusion in combination with time and space density factors on the basis, and dynamically updating sub-area boundaries so as to adapt to a behavior mode of an activity subject changing along with time; according to the method, the area can be accurately and flexibly divided in real time according to the actual activity condition, the space utilization efficiency is effectively improved, the method can be widely applied to various fields such as indoor space optimization layout, and powerful support is provided for reasonable resource allocation.

Description

Technical Field

[0001] The present invention belongs to the field of big data processing, and specifically relates to a method for regional division based on activity coordinate data. Background Art

[0002] With the rapid development of indoor positioning technologies (such as Wi-Fi positioning, Bluetooth beacons, UWB positioning, image analysis, etc.), it has become increasingly easy to collect the activity coordinate data of indoor personnel. These data can reflect the movement trajectories and residence times of people in the indoor environment in real time, contain the characteristics and laws of social activities among people, and provide an important basis for indoor space management and activity content optimization. Therefore, how to efficiently utilize these positioning data to accurately depict the activity laws and analyze the social characteristics contained in the activity trajectories is an urgent problem to be solved.

[0003] Generally speaking, for social law analysis of personnel activities within a region, it is first necessary to divide the activity region to clarify the personnel participating in various group activities, such as those participating in different game items during children's games. Usually, these are manually delimited in advance, and it is impossible to quickly adjust and adapt to changes in the activity hotspots and interests of children, resulting in the inability to automatically analyze and process. In addition, the manual annotation method has the problems of strong subjectivity and low accuracy, and it is difficult to accurately reflect the actual activity patterns. Especially in scenarios where activity hotspots change frequently, the errors of manual annotation will be further amplified. In recent years, some studies have attempted to perform dynamic region division based on the activity data of indoor personnel. For example, clustering algorithms are used to analyze indoor positioning data to generate dynamic regions. However, these methods usually only focus on the spatial dimension and ignore the changes in the time dimension and activity density, resulting in inaccurate division results. In addition, existing methods often face problems such as low computational efficiency and poor real-time performance when processing large-scale indoor data, and it is difficult to meet the actual application requirements. Especially in the case of a sharp increase in data volume, the algorithm complexity and computational resource consumption become the main bottlenecks.

[0004] Therefore, there is an urgent need for a dynamic region division method that can comprehensively consider space, time, and activity density to better meet the actual needs of indoor scenarios. Summary of the Invention

[0005] The present invention proposes a dynamic region division method based on activity coordinate data, which has the advantages of efficiently processing coordinate data and real-time identifying activity regions. Through steps such as data preprocessing, preliminary sub-region division, and spatio-temporal density fusion, combined with efficient algorithms and real-time data processing technologies, the present invention can achieve precise dynamic division of indoor regions.

[0006] Technical Solution: A dynamic region division method based on activity coordinate data, comprising the following steps: S1. Obtain the continuous coordinate data of multiple target objects within a certain geographical area and a certain data collection cycle within a given time interval, and then obtain the real-time coordinate data of the target objects at a certain moment; S2. Perform data preprocessing on the obtained real-time coordinate data of the target objects; S3. Perform dynamic time window partitioning on the preprocessed coordinate data, and initially divide the geographical area corresponding to the coordinate data of each dynamic time window into multiple sub-regions; S4. Based on the stored initial regional partitioning results and the sub-region overlap situation, perform regional fusion operations to dynamically update the sub-region boundaries; S5. For the results after regional fusion processing, to avoid the regional partitioning results not conforming to the actual region due to sparse data points in the edge regions, according to the density distribution of the target objects in each sub-region, use deep learning to further optimize the regional partitioning and improve the partitioning accuracy.

[0007] S6. After completing the above steps, the final dynamic regional partitioning results can be obtained.

[0008] Further, in S1, the real-time coordinate data includes the ID of the target object, three-dimensional coordinates, and the time information corresponding to the three-dimensional coordinate positions.

[0009] Further, in S1: (1) The method for obtaining continuous coordinate data includes UWB positioning, image processing, sensor positioning, or other technologies.

[0010] (2) When obtaining coordinates, take a certain point within the geographical area as the coordinate origin, and then obtain the real-time coordinate data of the target objects at a certain moment. The obtained real-time coordinate data needs to be able to distinguish the positions of a certain target object at different times.

[0011] (3) The setting of the time interval depends on the activity type of the target object. For the case of a large amount of activity, the time interval should be reduced, and the specific time interval value needs to be selected according to the subsequent data analysis results.

[0012] Further, in S2, the data preprocessing methods include removing duplicate values, missing values, and outliers in the coordinate data, and screening the real-time coordinate data of the target objects that meet the conditions; Duplicate values refer to the situation where, due to the influence of the acquisition device, a certain target object has multiple position coordinates at the same moment. It is necessary to judge and retain the correct coordinate data based on the position coordinates at the previous and subsequent moments for de-duplication processing; Missing values refer to the situation where, due to equipment failures, sensor failures, or network problems, some data fails to be correctly recorded or transmitted; An outlier refers to coordinate data that is identified in the collected coordinate data as exceeding the actual movement ability range of the target object within a set time interval through in-depth analysis of the behavioral characteristics of the target object.

[0013] Screening the data of target objects that meet the conditions means screening the collected coordinate data of target objects based on the preset boundary of the target area to extract the coordinate data of target objects that meet the conditions. Among them, the real-time coordinate data of the target objects that meet the conditions refers to the data whose coordinate data always appears within the target geographical area during the complete data collection cycle, and the target geographical area refers to the overall area that needs to be divided.

[0014] Furthermore, in S2, to improve the accuracy and reliability of the coordinate data, for missing values and outliers, an intelligent coordinate correction module is introduced to further optimize the deduplicated coordinate data. The specific method is as follows: 1) Construct an intelligent coordinate correction model based on the generative adversarial network (GAN), and use a large amount of historical coordinate data for learning and training. Through the adversarial training of the generator and the discriminator, the model automatically scans the coordinate data and identifies abnormal coordinates that do not conform to the behavioral characteristics of the target object, such as coordinate data that exceeds the actual movement ability range of the target object within a set time interval.

[0015] 2) For the identified abnormal coordinates, the model automatically generates more reasonable coordinate values for replacement according to the context information (such as the normal coordinate data at the previous and subsequent moments), so as to achieve efficient optimization of the coordinate data. The corrected coordinate data will be compared and verified with the original data to ensure the accuracy and reliability of the correction result.

[0016] By introducing the intelligent coordinate correction module, not only can the quality of the coordinate data be effectively improved, generating coordinate data closer to the real situation, avoiding complex manual operations, but also it has strong generalization ability, providing a reliable basis for subsequent data analysis and applications, thereby further enhancing the overall performance and reliability of the system.

[0017] Furthermore, the specific steps of S3 are as follows: S3.1. Adopt a dynamic adjustment algorithm based on a sliding window to adaptively adjust the size and step length of the time window to better capture the key features in the time series. The specific method is as follows: S3.1.1. According to the dynamic characteristics of the coordinate data, by analyzing the behavioral patterns of the target object and the fluctuation characteristics of the data, dynamically adjust the size of the time window. During the period when the target object is active frequently or the data changes violently, narrow the size of the time window to capture more details; during the period when the target object is active stably or the data changes gently, appropriately increase the size of the time window to reduce the data processing frequency. The formula for adjusting the window size is as follows: (1) Wherein, w is the size of the current sliding time window, w min and w max are the minimum and maximum values of the window size respectively, and data_variance( D i : i + w ) refers to the variance of the data within the window, indicating the volatility of the data within the window, D [i:i + w is the sequence of data points from the i th to the i + w th data point within the window, θ high is the high data volatility threshold, θ low is the low data volatility threshold; S3.1.2. Dynamically adjust the step size of the time window according to the continuity and change trend of the data. In a period with rapid data change, a smaller step size is adopted to ensure data continuity; in a period with slow data change, a larger step size is adopted to improve the partitioning efficiency. The step size adjustment formula is as follows: (2) Wherein, s is the step size of the current sliding time window, s min and s max are the minimum and maximum values of the step size respectively, and data_rate( D i : i + w ) refers to the change rate of the data within the window, indicating the change speed of the data within the window, D i : i + w is the sequence of data points from the i th to the i + w th data point within the window, θ rate is the threshold of the data change rate.

[0018] S3.2. For the coordinate data within each dynamic time window, perform preliminary geographical area partitioning respectively, generate a set of sub - regions corresponding to each time period, and store the preliminary area partitioning results.

[0019] ​​​The preliminary regional division method includes, but is not limited to, the K-Means clustering algorithm, the DBSCAN clustering algorithm, the grid division method, and the division method based on the minimum spanning tree, etc. Through the above steps, the present invention can effectively implement the spatio-temporal segmentation processing of coordinate data, dynamically adapt to the changing characteristics of the data, provide more accurate basic data support for subsequent dynamic regional division, and further enhance the flexibility and accuracy of the system.

[0020] Further, the specific steps of S4 are as follows: S4.1. Perform kernel density estimation on the data points in each sub-region. Each set of preliminary division results contains multiple sub-regions, and each sub-region is defined by its boundary coordinates (x1, y1, x2, y2), where x1, y1 are the coordinates of the lower left corner, and x2, y2 are the coordinates of the upper right corner. Perform kernel density estimation (KDE) on the data points in each sub-region and calculate the density value of each data point. The kernel density estimation formula is: (3) Where: is the data point x is the kernel density estimation value of K is the kernel function, which is used to measure the contribution weight of a certain data point x in the neighborhood of the data point x i to the data point x , h is the bandwidth, which is used to control the smoothness of the kernel function, n is the total number of data points in the current sub-region (i.e., the number of samples participating in the density calculation), d is the data dimension.

[0021] Through this formula, for each target point x , calculate the weighted contribution of all data points x i in its neighborhood (the weight is determined by the kernel function K ), and then normalize it through the bandwidth h and the number of samples n to obtain the kernel density estimation value .

[0022] S4.2. According to the kernel density estimation value, divide the data points in the sub-region into high-density regions and low-density regions. The sub-region with a kernel density estimation value higher than the preset density threshold θ high is a high-density region; the sub-region with a kernel density estimation value lower than the preset density threshold θ low is a low-density region.

[0023] S4.3. Arrange the stored preliminary regional division results in chronological order, and compare the coordinates of all sub-regions in the preliminary division results of adjacent two groups in sequence. If there is an area overlap between a certain sub-region in the preliminary regional division result of one group and a certain sub-region in the result of the other group, calculate the kernel density estimate value of the overlapping area. If the overlapping area is greater than or equal to the dynamic fusion threshold, perform the area fusion operation.

[0024] If there is an area overlap between a certain sub-region in the preliminary regional division result of one group and a certain sub-region in the result of the other group, the specific operation steps are as follows: S4.3.1. Calculate the overlapping area R of the two sub-regions overlap : (4) Among them, R overlap is the overlapping area of the two sub-regions, (x1, y1, x2, y2) are the boundary coordinates of the first sub-region, (x1, y1) is the lower left corner coordinate, (x2, y2) is the upper right corner coordinate, (x1′, y1′, x2′, y2′) are the boundary coordinates of the second sub-region, (x1′, y1′) is the lower left corner coordinate, and (x2′, y2′) is the upper right corner coordinate.

[0025] S4.3.2. Calculate the area A of the overlapping area overlap : (5) Among them, A overlap is the area of the overlapping area of the two sub-regions, is the left boundary of the overlapping area, is the left boundary of the overlapping area, is the right boundary of the overlapping area, is the upper boundary of the overlapping area.

[0026] S4.3.3. Use formula (3) to calculate the kernel density estimate value of the overlapping area, and dynamically adjust the fusion threshold θ according to the kernel density estimate value of the overlapping area fuse : (6) Among them, θ fuse is the dynamic fusion threshold, θ high and θ low are the preset high-density threshold and low-density threshold respectively, used to classify the density level, represents the kernel density estimate value within the overlapping area.

[0027] S4.3.4. If the overlapping area A overlap is greater than or equal to the dynamic fusion threshold θ fuse , then perform the region fusion operation; otherwise, directly retain the sub-region and save it as an independent new sub-region in the new region division result.

[0028] In S4.3.4, the steps for performing the region fusion operation are as follows: (1) For the sub-regions that meet the fusion conditions, apply the adaptive density region fusion clustering algorithm for clustering. Dynamically adjust the parameters and MinPts of the adaptive density region fusion clustering according to the kernel density estimation value: (7) Among them, is the neighborhood radius, used to define the neighborhood range of data points, and are the neighborhood radius parameters of the high-density and low-density regions respectively, MinPts is the minimum number of neighborhood points required for a core point, and MinPts large and MinPts small are the minimum number of points parameters of the high-density and low-density regions respectively, is the kernel density estimation value within the overlapping region, θ high is the preset high-density threshold.

[0029] (2) For the data points within the overlapping region, judge whether they are core points according to their density values and the number of points within the neighborhood. If a point is a core point, start from this point and expand the clustering through the density reachable relationship.

[0030] (3) For the sub-clusters formed after the adaptive density region fusion clustering, judge whether they meet the fusion conditions. If they meet the fusion conditions (such as the density distribution of the sub-clusters is continuous and the kernel density estimation value is high), then merge these sub-clusters into a new sub-region; if they do not meet the fusion conditions, then retain these sub-clusters as independent sub-regions.

[0031] S4.4. For the merged sub-regions, dynamically adjust their boundaries according to the kernel density estimation value and save the newly generated sub-regions in the new region division result. The method for dynamically adjusting the sub-region boundaries is as follows: Compare the four corner coordinates (x1, y1, x2, y2) and (x 1’ , y 1’ , x 2’ , y 2’ ) of the two sub-regions, where (x1, y1) and (x2, y2) are the lower left and upper right coordinates of the first sub-region respectively, and (x 1’ , y1’ ), and (x 2’ , y 2’ ) are the lower left coordinate and the upper right coordinate of the second sub-region respectively. The minimum and maximum values in the x and y directions are extracted respectively to generate the boundary coordinates of the new sub-region (x min , y min , x max , y max ): x min = min(x1, x 1’ ), y min = min(y1, y 1’ ); x max = max(x2, x 2’ ), y max = max(y2, y 2’ ); Each coordinate of the fused sub-region contains a unique region number and the fused four-corner coordinates.

[0032] S4.5. After completing the region fusion operation for the adjacent two groups of preliminary region division results, further perform the region fusion operation on all sub-regions in the new region division result. This region fusion operation is directly based on the new region division result without repeating the storage of intermediate results, thus effectively saving storage space and ensuring the accuracy and consistency of the new region division result.

[0033] Furthermore, the specific steps of S5 are as follows: S5.1. From all the coordinate data after preprocessing, filter out the coordinate data existing in each sub-region (the updated sub-region in S4), and draw the heat map of each sub-region. The color distribution in the heat map reflects the density distribution and its change of the target object, and the heat map will be used as the input data of the deep learning model.

[0034] S5.2. Use the deep learning model to perform multi-scale analysis on the heat map to capture the local and global features in the heat map. Multi-scale analysis can effectively handle the uneven density distribution of coordinate data and accurately identify high-density regions and low-density regions.

[0035] S5.3. Introduce an attention mechanism in the deep learning model so that it can automatically focus on the key regions in the heat map and improve the accuracy of density distribution judgment. The attention mechanism optimizes the feature extraction process by assigning weights to highlight important regions and suppress unimportant regions.

[0036] S5.4. Optimize the boundaries of sub-regions according to the processing results of the deep learning model: If there are two or more high-density regions in a certain sub-region, retain the original regional boundaries and make comprehensive adjustments according to the edges of multiple high-density regions; if there is only one high-density region in a certain sub-region, re-divide the regional boundaries according to the edge of this high-density region.

[0037] Beneficial effects: Compared with the prior art, the present invention not only significantly improves the computing efficiency, but also can adapt to changes in activity patterns in real time, avoiding the subjectivity and inaccuracy of manual annotation. For example, in kindergartens, the present invention can dynamically adjust the division of game areas according to the activity hotspots and interest changes of children, ensuring the efficient use of space resources and improving the activity experience of children at the same time. In addition, this technology can also be widely applied to indoor scenarios such as shopping malls and warehouses, providing strong support for space planning, crowd flow management, resource allocation, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flowchart of the method of the present invention; Figure 2 is a comparison diagram of the preliminary area division result and the manually marked result of the embodiment of the present invention; Figure 3 is a density distribution heat map of the preliminary area division result of the embodiment of the present invention; Figure 4 is a diagram of the final area division result of the embodiment of the present invention; Figure 5 is a comparison diagram of the final area division result and the manually marked result of the embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0039] The technical solution of the present invention will be described in detail below through embodiments, but the protection scope of the present invention is not limited to the described embodiments.

[0040] Embodiment 1 Figure 1 is a flowchart of a dynamic area division method based on activity coordinate data of the present invention. This embodiment is a case of dynamic area division of a game area for children's role-playing games using this method.

[0041] The specific steps are as follows: Step 1. Obtain the real-time coordinate data of all children during the game time period: Taking the location of the front door of the classroom as the origin of the coordinate axis, deploy multiple UWB base stations above the classroom. Each child wears a UWB tag with sufficient battery during the game. Using the trilateration method, collect the continuous activity coordinate data of each child at fixed time intervals (every 1 second) during the game period to ensure the continuity and accuracy of the data. The obtained data includes time, child ID, and their corresponding three-dimensional coordinates (x, y, z).

[0042] Step 2: Perform data preprocessing on all the real-time coordinate data of the obtained children, including removing duplicate values, missing values, and outliers in the coordinate data, and screening the coordinate data of target objects that meet the conditions. The specific steps are as follows: (1) Ensure that the data format is unified, and all data column names should be time, id, x, y, z. At the same time, use the deduplication function to deduplicate the records with the same values in the time and id columns and the same coordinate data (x, y) in the original data, and handle the missing data by zero filling, and ensure that there is still enough data for dynamic area division after deduplication and filling.

[0043] (2) In the collected coordinate data, define the coordinate data with the distance difference between adjacent time points exceeding the reasonable movement level of children (moving more than 1.5 meters within 1 second) as outliers. Establish an intelligent coordinate correction model based on LSTM-GAN. The generator uses a 3-layer LSTM network to learn the movement patterns of children, and the discriminator is a convolutional neural network. Define a dynamic outlier threshold: trigger outlier detection when the displacement speed between two adjacent points exceeds 1.5 m / s (considering the maximum running speed of children). The model analyzes the trajectory context of the previous and next 10 seconds, generates a smooth trajectory that conforms to kinematics for replacement, and verifies the rationality of the correction result through a residual network.

[0044] (3) Screen out the coordinate data of target objects that meet the conditions according to the boundaries of the target area. Since the classroom space is limited and only the area within the classroom is used for game area division, screen all the coordinate data of children by giving the value ranges of x and y, and finally screen out all the coordinate data of children within the classroom as the data for subsequent area division.

[0045] Step 3: Divide the preprocessed coordinate data into coordinate data of multiple fixed time periods according to the time series, and perform preliminary area division on each group. The specific steps are as follows: (1) The activity duration of children in the game area is 40 minutes. Sort the preprocessed coordinate data according to time, set the initial window size to 4 minutes, and the initial step size to 2 minutes. Use a dynamic adjustment algorithm based on a sliding window to adaptively adjust the size and step size of the time window.

[0046] During periods when children are highly active or data changes rapidly, automatically adjust the window size to 3 minutes and the step size to 1 minute; during periods when children's activities are stable or data changes smoothly, automatically increase the window size to 5 minutes and the step size to 2 minutes.

[0047] (2)Customize the neighborhood radius (eps) and minimum number of points (min_samples) of the DBSCAN clustering algorithm. In this embodiment, the neighborhood radius selected for the division of children's game areas is 0.6, and the minimum number of points is 180.

[0048] (3)For the coordinate data within each dynamic time window, apply the DBSCAN clustering algorithm to perform a preliminary regional division, merge clusters with an overlap degree > 65%, eliminate small clusters with a number of points < 50, generate the preliminary dynamic sub-region results corresponding to each time period, and save the coordinates of each divided sub-region as a JSON file.

[0049] Step Four: Based on the coordinate data in each group of JSON files saved from the preliminary regional division, perform regional fusion and dynamically update the sub-region boundaries. The specific steps are as follows: (1)Parse the JSON file storing the preliminary regional division results, extract the coordinates of each sub-region in each group of divisions, group the coordinate data by time point, and generate a dictionary indexed by time point.

[0050] (2)For each sub-region, use a high kernel density estimation with a bandwidth of 0.45 meters (based on the average shoulder width of children and safety margins) to divide the data points within the sub-region into high-density regions and low-density regions.

[0051] If there is an area overlap between a certain sub-region in the preliminary regional division results of a certain group and a certain sub-region in the results of another group, calculate the kernel density estimation value of the overlapping area; If the overlapping area is greater than or equal to the dynamic fusion threshold (dynamically adjusted according to the kernel density estimation value within the overlapping area, the threshold for the high-density area (≥ 1.2 people / ㎡) is 30%, the medium-density area (0.8 - 1.2 people / ㎡) is 35%, and the low-density area is 40%) and the continuous overlap ≥ 3 groups of windows (about 5 - 7 minutes), then perform the regional fusion operation; otherwise, directly retain the sub-region and save it as an independent new sub-region in the new regional division results.

[0052] For sub-regions that meet the fusion conditions, apply the adaptive density region fusion clustering algorithm for clustering.

[0053] According to the kernel density estimates, dynamically adjust the parameters ϵ and MinPts of the algorithm. First, dynamically calculate the key parameters based on the kernel density distribution in the overlapping area: the neighborhood radius ϵ ranges from 0.5 to 0.7 meters (the base value is 0.6 meters, and it expands and contracts dynamically by ±20% according to the density range), and the minimum number of points MinPts is set to 150 - 200 (weighted calculation based on the area and average density of the region). Identify the core points with a density higher than 60% of the average value, then expand the region with the condition that the density gradient < 20%, and finally generate and save the rectangular boundary.

[0054] (4)For the merged sub-regions, dynamically adjust their boundaries according to the kernel density estimates, and save the newly generated sub-regions into the new regional division results. Finally, save the fused regional division results as a new JSON file, and each coordinate of the fused region contains a unique region number and the four corner coordinates (x n 、y n 、x m 、y m ).

[0055] S5. For the regional division results after regional fusion processing, analyze the density distribution of target objects in each sub-region, and use deep learning to optimize the regional division. The specific steps are as follows: (1)Extract the four corner coordinates (x n 、y n 、x m 、y m ) of each sub-region divided in the JSON file, and define the corresponding regional coordinate range and name.

[0056] (2)Use the coordinate data of all preprocessed children to draw a hexagonal heat map, where the color scale is set to 100 and the size of the hexagon is set to 30.

[0057] (3)Construct a two-channel deep network. Channel 1 (local features) is a 3-layer convolutional network with a convolutional kernel size of 3×3 to capture microscopic distributions; Channel 2 (global features) is a dilated convolutional network with a dilation rate of 5 to perceive macroscopic patterns; the feature fusion layer realizes feature interaction through a cross-attention mechanism, and the weights of local features are dynamically adjusted by global features.

[0058] (4)Use the drawn heat map as the input data of the deep learning model, perform multi-scale analysis on the heat map, capture the local and global features in the heat map, and accurately identify high-density regions and low-density regions.

[0059] (5) Optimize the boundaries of the sub-regions according to the processing results of the deep learning model: If there are two or more high-density regions within a certain sub-region, retain the original region boundary and make comprehensive adjustments according to the edges of the multiple high-density regions; if there is only one high-density region within a certain sub-region, re-divide the region boundary according to the edge of the high-density region.

[0060] Figure 4 It is the final region division result diagram of the embodiment of the present invention; Figure 5 It is a comparison diagram of the final region division result of the embodiment of the present invention and the result of manual annotation. Figure 4 and Figure 5 In [diagram], the solid-line box represents the sub-region after the operation of step S5 of the embodiment, and the dashed-line box represents the activity area divided manually according to the boundaries of facilities such as desks, chairs, and cabinets in the kindergarten classroom. Through Figure 5 The comparison in [diagram] shows that the region division result after the operation of step S5 of the embodiment is similar to the result of manual division, proving the effectiveness of using this method to divide regions. At the same time, this method takes into account human mobility and is more accurate than manual annotation according to the layout of facilities in the activity area division.

[0061] Through the above steps, the present invention can achieve dynamic region division of children's role-playing games, providing technical support to optimize children's behavior analysis and game experience.

[0062] Figure 2 It is a comparison diagram of the preliminary region division result of the embodiment of the present invention and the result of manual annotation; the solid-line box represents the sub-region after the operation of step S4 of the embodiment. The dashed-line box represents the activity area divided manually according to the boundaries of facilities such as desks, chairs, and cabinets in the kindergarten classroom. Figure 3 It is a density distribution heat map of the preliminary region division result of the embodiment of the present invention; the solid-line box represents the sub-region after the operation of step S4 of the embodiment. Figure 2 It shows that the S4 region fusion effect is already relatively ideal. However, considering that after the S4 region fusion operation, due to the distribution of scattered points at the edges (data point density distribution problem), there may be a large difference between the space divided by each sub-region and the actual activity area. Therefore, further performing S5 to divide more accurate regions is more accurate. Figure 3 It is a heat map showing the execution of step S5 (display of the process).

[0063] In the method of the present invention, first, the activity coordinate data of multiple active entities within a target area are collected within a certain time period, and outliers and noise interference are removed through data preprocessing to ensure data accuracy. Then, based on the distribution characteristics of the coordinates, the activity area is initially divided into multiple sub-areas. On this basis, the sub-areas are fused by combining time and space density factors, and the boundaries of the sub-areas are dynamically updated to adapt to the behavior patterns of the active entities changing over time. The present invention can accurately and real-time flexibly divide the area according to the actual activity situation, effectively improve the space utilization efficiency, and can be widely applied to many fields such as the optimal layout of indoor spaces, providing strong support for the rational allocation of resources.

[0064] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as a limitation of the present invention itself. Various changes can be made to it in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A dynamic region partitioning method based on activity coordinate data, characterized in that It includes the following steps: S1. Obtain the continuous coordinate data of multiple target objects within a certain geographical area and within a data collection period at given time intervals, and then obtain the real-time coordinate data of the target objects at a certain moment; S2. Perform data preprocessing on the obtained real-time coordinate data of the target objects; S3. Perform dynamic time window partitioning on the preprocessed coordinate data, and preliminarily partition the geographical areas corresponding to the coordinate data in each dynamic time window to obtain multiple sub-regions; S4. Based on the stored preliminary regional partitioning results and sub-region overlap situations, perform regional fusion operations to dynamically update the sub-region boundaries.

2. The dynamic region division method according to claim 1, wherein In S1, the real-time coordinate data includes the ID of the target object, three-dimensional coordinates, and time information corresponding to the three-dimensional coordinate positions.

3. The dynamic region division method according to claim 1, wherein In S2, the data preprocessing methods include removing duplicate values, missing values, and outliers from the coordinate data, and screening out the real-time coordinate data of the target objects that meet the conditions; The real-time coordinate data of the target objects that meet the conditions refers to the data where the coordinate data of the target objects always appears within the scope of the target geographical area during the complete data collection period. The target geographical area refers to the overall area that needs to be regionally partitioned.

4. The dynamic region division method according to claim 1, wherein In S2, for missing values and outliers, an intelligent coordinate correction module is introduced to optimize the coordinate data after duplicate removal. The specific method is as follows: Construct an intelligent coordinate correction model based on a generative adversarial network; the intelligent coordinate correction model automatically scans the coordinate data, identifies abnormal coordinates that do not conform to the behavioral characteristics of the target object, and automatically generates more reasonable coordinate values for replacement for the identified abnormal coordinates.

5. The dynamic area division method according to claim 1, wherein The specific steps of S3 are as follows: S3.

1. Adopt a dynamic adjustment algorithm based on a sliding window to adaptively adjust the size and step length of the time window; S3.

2. For the coordinate data in each dynamic time window, respectively perform preliminary partitioning of the geographical area, generate a set of sub-regions corresponding to each time period, and store the preliminary regional partitioning results.

6. The dynamic region division method according to claim 1, wherein The specific steps of S3.1 are as follows: S3.1.

1. According to the dynamic characteristics of the coordinate data, dynamically adjust the size of the time window. The window size adjustment formula is as follows: （1） Among them, w is the size of the current sliding time window, w min and w max are the minimum and maximum values of the window size respectively. data_variance( D i : i + w ) refers to the variance of the data within the window, indicating the volatility of the data within the window. D [i:i + w is the sequence of data points within the window from the i -th data point to the i + w -th data point. θ high is the high data volatility threshold, θ low is the low data volatility threshold;​ S3.1.

2. According to the continuity and change trend of the data, dynamically adjust the step length of the time window. The step length adjustment formula is as follows: （2） Among them, s is the step size of the current sliding time window, s min and s max are the minimum and maximum values of the step size respectively, and data_rate( D i : i + w ) refers to the change rate of the data within the window, indicating the change speed of the data within the window, D i : i + w is the sequence of data points in the window from the i -th data point to the i + w -th data point, θ rate is the threshold of the data change rate.​​ 7. The dynamic region partitioning method according to claim 1, wherein S4. The specific operation steps are as follows: S4.

1. Perform kernel density estimation on the data points in each sub-region. The kernel density estimation formula is: （3） Wherein: is the kernel density estimate of the data point x , K is the kernel function, used to measure a data point x in the neighborhood of a certain data point x i for the contribution weight of the data point x , h is the bandwidth, used to control the smoothness of the kernel function, n is the total number of data points in the current sub-region (i.e., the number of samples participating in the density calculation), d is the data dimension; S4.

2. Divide the data points in the sub-region into a high-density region and a low-density region according to the kernel density estimate value; the sub-region with a kernel density estimate value higher than the preset density threshold θ high is the high-density region; the sub-region with a kernel density estimate value lower than the preset density threshold θ low is the low-density region; S4.

3. Arrange the stored preliminary regional partitioning results in chronological order, and successively compare the coordinates of all sub-regions in the preliminary partitioning results of adjacent two groups. If a certain sub-region in the preliminary regional partitioning result of a certain group overlaps with a certain sub-region in the result of another group, calculate the kernel density estimation value of the overlapping region. If the overlapping area is greater than or equal to the dynamic fusion threshold, perform the regional fusion operation; S4.

4. For the merged sub-regions, dynamically adjust their boundaries according to the kernel density estimation value, and save the newly generated sub-regions to the new regional partitioning results; S4.

5. After performing the region fusion operation on the adjacent two sets of preliminary region division results, further perform the region fusion operation on all sub-regions in the new region division result.

8. The dynamic region division method according to claim 1, wherein In S4.3, if there is a region overlap between a certain sub-region in the preliminary region division result of a certain group and a certain sub-region in the result of another group, the specific operation steps are as follows: S4.3.

1. Calculate the overlapping region R of the two sub-regions overlap :[[]]END]] （4） Among them, R overlap is the overlapping region of two sub-regions, (x1, y1, x2, y2) are the boundary coordinates of the first sub-region, (x1, y1) is the lower left corner coordinate, (x2, y2) is the upper right corner coordinate, (x1′, y1′, x2′, y2′) are the boundary coordinates of the second sub-region, (x1′, y1′) is the lower left corner coordinate, and (x2′, y2′) is the upper right corner coordinate; S4.3.

2. Calculate the area A of the overlapping region overlap : （5） Among them, A overlap is the area of the overlapping region of two sub-regions, is the left boundary of the overlapping region, is the left boundary of the overlapping region, is the right boundary of the overlapping region, is the upper boundary of the overlapping region; S4.3.

3. Calculate the kernel density estimate value of the overlapping region, and dynamically adjust the fusion threshold θ according to the kernel density estimate value of the overlapping region fuse : （6） Among them, θ fuse is the dynamic fusion threshold, θ high and θ low are the preset high-density threshold and low-density threshold respectively, used for classifying the density level, represents the kernel density estimation value within the overlapping region; S4.3.

4. If the overlapping area A overlap is greater than or equal to the dynamic fusion threshold θ fuse , then perform the region fusion operation; otherwise, directly retain the sub-region and save it as an independent new sub-region in the new region division result. The steps for performing the region fusion operation are as follows: (1) For the sub-regions that meet the fusion conditions, apply the adaptive density region fusion clustering algorithm for clustering; according to the kernel density estimation value, dynamically adjust the parameters of the adaptive density region fusion clustering and MinPts: （7） Among them, is the neighborhood radius, which is used to define the neighborhood range of data points, and are the neighborhood radius parameters of the high-density and low-density regions respectively. MinPts is the minimum number of neighborhood points required for a core point. MinPts large and MinPts small are the minimum number of points parameters of the high-density and low-density regions respectively, is the kernel density estimate value within the overlapping region, θ high is the preset high-density threshold; (2) For the data points in the overlapping region, judge whether it is a core point according to its density value and the number of points in the neighborhood; if a point is a core point, start from this core point and expand the clustering through the density-reachable relationship. (3) For the sub-clusters formed after the adaptive density region fusion clustering, judge whether they meet the fusion conditions; if they meet the fusion conditions, merge these sub-clusters into a new sub-region; if they do not meet the fusion conditions, retain these sub-clusters as independent sub-regions.

9. The dynamic region division method according to claim 1, wherein The method for dynamically adjusting the sub-region boundary in S4.4 is as follows: Compare the four corner coordinates (x1, y1, x2, y2) and (x 1’ , y 1’ , x 2’ , y 2’ ) of two sub-regions, where (x1, y1) and (x2, y2) are the lower left corner coordinate and the upper right corner coordinate of the first sub-region respectively, and (x 1’ , y 1’ ) and (x 2’ , y 2’ ) are the lower left corner coordinate and the upper right corner coordinate of the second sub-region respectively. Extract the minimum and maximum values in the x and y directions respectively to generate the boundary coordinates (x min , y min , x max , y max ) of the new sub-region: x min = min(x1, x 1’ ), y min = min(y1, y 1’ ); x max = max(x2, x 2’ ), y max = max(y2, y 2’ ).

10. The dynamic region division method according to claim 1, wherein It also includes S5. For the result after the region fusion processing, further optimize the region division by using deep learning according to the density distribution of the target objects in each sub-region.

Images