Artificial intelligence-based vr interaction data management method

Abstract

The application relates to the technical field of VR interaction data management, in particular to a VR interaction data management method based on artificial intelligence. The method comprises the following steps: acquiring coordinate point sequences of each joint point and normal fluctuation thresholds of various coordinates respectively; obtaining action change coordinate points; dividing the coordinate point sequences to obtain different compressed data segments; forming two compressed data segments with time containing or equal relationship in the compressed data segments corresponding to two joint points into a to-be-analyzed combination, and then obtaining a data segment combination; acquiring relative position vectors of various coordinates of the data segment combination; judging whether the data segment combination is a linkage combination; acquiring an action mode code of the linkage combination and a fitting slope value of each coordinate of a basic data segment in the linkage combination; and storing the action mode code of the data segment combination, initial coordinate points in the basic data segment and the fitting slope value of each coordinate. The application can improve the compression rate of VR interaction data.

Description

AI-based VR interactive data management methods Technical Field

[0001] This invention relates to the field of VR interactive data management technology, and more specifically to a VR interactive data management method based on artificial intelligence. Background Technology

[0002] VR interactive data refers to the data collection generated when users operate and experience in a virtual environment. Using this collected VR interactive data, commands such as actions and voice input are generated within the virtual environment, ensuring that the movements of characters and objects in the virtual environment match the user's real-world actions, providing an immersive experience. However, due to the large volume and multiple dimensions of VR interactive data, which can generate massive amounts of data in a short period, the collected data often needs to be compressed before storage to improve data storage efficiency and conserve storage resources.

[0003] Existing VR interactive data compression methods mainly extract motion data by using a fixed frame rate. However, since the complexity of user actions varies at different times, a fixed frame rate is usually difficult to achieve good compression results under different actions. For example, the compression is insufficient when there are few changes in actions, while the accuracy of the compressed actions is insufficient when there are many changes in actions. This not only makes it difficult to ensure the efficiency of storage resource utilization, but may also affect further data analysis. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a VR interactive data management method based on artificial intelligence, the specific technical solution of which is as follows:

[0005] One embodiment of the present invention provides a VR interactive data management method based on artificial intelligence, the method comprising:

[0006] The three-dimensional coordinates of each key point at each moment are collected within a preset time period, and the coordinates of each key point are formed into a coordinates sequence; the changes of various coordinates of each coordinate point in a coordinates sequence over time are analyzed to obtain the normal fluctuation threshold of various coordinates.

[0007] Based on the normal fluctuation threshold of various coordinates, all coordinate points are filtered to obtain coordinate points of motion change; the coordinate point sequence is divided according to the coordinate points of motion change to obtain different data segments to be compressed;

[0008] Two data segments whose times are inclusive or equal, corresponding to two key points, are combined into an analysis group; the parts of the two data segments whose times are equal are extracted from the analysis group and denoted as the first data segment and the second data segment, and then combined into a data segment group.

[0009] Based on the differences between various coordinates of coordinate points at the same time in the first and second data segments, the relative position vectors of various coordinates of the data segment combination are obtained; based on the relative position vectors of various coordinates of a data segment combination, it is determined whether the data segment combination is a linked combination.

[0010] The relative position vectors of various coordinates in the linkage combination are encoded to obtain the action pattern code; any data segment in the linkage combination is selected as the basic data segment and the fitting slope value of each coordinate in the basic data segment is obtained respectively; the action pattern code of the linkage combination, the initial coordinate point in the basic data segment and the fitting slope value of each coordinate are stored.

[0011] Preferably, the normal fluctuation thresholds of various coordinates are obtained by analyzing the changes of various coordinates of each coordinate point in a coordinate point sequence over time, including:

[0012] To obtain a time-series coordinate sequence, the coordinates of each point in the coordinate sequence of a key point are arranged in chronological order. The coordinates in the time-series coordinate sequence corresponding to this type of coordinate are smoothed using the exponential moving average method, resulting in a smoothed time-series coordinate sequence. The slope value of each data point in the smoothed time-series coordinate sequence is calculated. The slope values ​​of each data point in the smoothed time-series coordinate sequence are arranged in chronological order to obtain a slope value sequence. The slope values ​​in the slope value sequence are clustered to obtain clustering results. Based on the clustering results, the smoothed time-series coordinate sequence is divided into different coordinate data segments. Linear fitting is performed on each coordinate in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate to obtain a fitted coordinate value for each coordinate in each coordinate data segment. The normal fluctuation threshold for this type of coordinate is obtained based on the difference between the fitted coordinate values ​​of each coordinate in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate. Similarly, the normal fluctuation thresholds for other types of coordinates are obtained.

[0013] Preferably, the normal fluctuation threshold of this type of coordinate is obtained based on the difference between the coordinate fitting values ​​of each coordinate and each coordinate in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate, including:

[0014] The absolute value of the difference between a coordinate in the smoothed time-series coordinate sequence corresponding to this type of coordinate and the coordinate fitting value of this coordinate is taken as the fluctuation amplitude of this coordinate; the maximum value of the fluctuation amplitude in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate is obtained; the mean of the maximum values ​​of the fluctuation amplitude in all coordinate data segments is taken as the normal fluctuation threshold of this type of coordinate.

[0015] Preferably, all coordinate points are filtered based on normal fluctuation thresholds for various coordinates to obtain coordinate points showing motion changes, including:

[0016] Linear fitting is performed on all coordinates of a smoothed time-series coordinate sequence composed of a certain type of coordinate to obtain the coordinate fitting value of each coordinate. The absolute value of the difference between a coordinate and its coordinate fitting value is taken as the fluctuation amplitude of that coordinate. If the fluctuation amplitude of at least one coordinate in a coordinate point is greater than the normal fluctuation threshold of that type of coordinate, then that coordinate point is the action change coordinate point.

[0017] Preferably, the relative position vectors of various coordinates of the data segment combination are obtained based on the differences between various coordinates of coordinate points at the same time in the first data segment and the second data segment, including:

[0018] The difference between a coordinate at a certain time in the first data segment and the same coordinate at the same time in the second data segment is taken as the relative position of that coordinate at that time. The relative positions of the coordinates at each time are combined to form a relative position vector of that coordinate. Similarly, the relative position vectors of various coordinates of the data segment combination are obtained.

[0019] Preferably, determining whether a data segment combination is a linked combination based on the relative position vectors of various coordinates of a data segment combination includes:

[0020] In a data segment combination, the relative positions of each coordinate at a given moment in the relative position vector corresponding to each coordinate are used to form a relative position data point. These relative position data points are then sorted chronologically to obtain the relative position data sequence of the data segment combination. The relative positions of each coordinate at each moment are calculated from the historical coordinates of the two key points corresponding to the data segment combination, and the relative position data points from each historical moment are combined to form a historical relative position data sequence. A window of the same length as the first and second data segments in the data segment combination is slid across the historical relative position data sequence. The sequence of relative position data points within a window during the sliding process is recorded as the window's relative position sequence. The relative position similarity of the window is calculated based on the Euclidean distance between the relative position data points within the window's relative position sequence and the corresponding relative position data points in the relative position data sequence of the data segment combination. If the relative position similarity of at least one window is greater than a preset threshold, then the data segment combination is considered a linked combination.

[0021] Preferably, the relative position similarity of the window is calculated based on the Euclidean distance between each relative position data point in the window relative position sequence of a window and the relative position data points at the corresponding time in the relative position data sequence of the data segment combination, including:

[0022] Calculate the sum of the Euclidean distances between each relative position data point within the window relative position sequence of a window and the corresponding relative position data points in the relative position data sequence of the data segment combination, denoted as the total distance; calculate the reciprocal of the sum of the total distance and the hyperparameter and normalize it to obtain the relative position similarity.

[0023] The embodiments of the present invention have at least the following beneficial effects: This application collects the three-dimensional coordinate points of each key point at each moment within a preset time period, and forms a coordinate point sequence for each key point. Then, it analyzes the changes of various coordinates of each coordinate point in a coordinate point sequence over time to obtain the normal fluctuation threshold of various coordinates, and then obtains the action change coordinate points. Based on the action change coordinate points, the coordinate point sequence is divided to obtain different data segments to be compressed. Here, the user's action is the position change of the key point, thereby reflecting the complexity of the user's action change. The length of the data segment to be compressed is adaptively adjusted according to the complexity of the user's action change. This can not only retain more accurate action data when the user's action changes frequently, but also improve the compression rate and reduce the waste of storage resources when the user's action is relatively static or changes regularly.

[0024] Then, the linkage between different joints is analyzed to determine whether the data segment combination is a linkage combination. If it is a linkage combination, it means that the actions of the two joints are linked during the linkage combination period. That is, the two joints have a fixed action pattern during this period. This allows us to identify the user's fixed action pattern and store it by encoding the fixed action pattern. This can effectively reduce the amount of data stored and improve resource utilization efficiency. Attached Figure Description

[0025] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 is a flowchart of a VR interactive data management method based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0027] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a VR interactive data management method based on artificial intelligence proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] The following description, in conjunction with the accompanying drawings, details a specific solution for a VR interactive data management method based on artificial intelligence provided by this invention.

[0030] Example:

[0031] The main application scenario of this invention is as follows: the complexity of user actions when using VR is not constant. When using a fixed frame rate for data compression, the compression accuracy may be insufficient when the action complexity is high, while the compression degree may be insufficient when the action complexity is low, resulting in a waste of storage resources. This will affect the data processing efficiency when further analyzing and processing user data.

[0032] Please refer to Figure 1, which shows a flowchart of a VR interactive data management method based on artificial intelligence provided by an embodiment of the present invention. The method includes the following steps:

[0033] Step S1: Collect the three-dimensional coordinates of each joint at each moment within a preset time period to form a coordinate point sequence for each joint; analyze the changes of various coordinates of each coordinate point in a coordinate point sequence over time to obtain the normal fluctuation threshold of various coordinates.

[0034] The specific scenarios in which VR is used will vary, and the targets of its data collection will also differ. Since VR is currently mainly used in games, education, virtual meetings, etc., most VR interactions are achieved through controllers and gestures. This application takes the monitoring of users' hand movements as an example for analysis.

[0035] Using the sensors inside the motion capture gloves that come with VR, the position data of each joint of the user's hand at each moment is acquired in real time, that is, the three-dimensional coordinates (x, y, z) of each joint at each moment are collected. The data acquisition frequency of the motion capture gloves is adjusted according to the specific application requirements. Here, the data acquisition frequency is set to 100Hz for most scenarios.

[0036] Since this compression process is used for storing user data and has low real-time requirements, it is set to store data once every 1 second during the monitoring process. Thus, the three-dimensional coordinates of each joint point at each moment are collected within a preset time period (the preset time period length is 1 second). The three-dimensional coordinates of each joint point at each moment within 1 second form the joint point coordinate sequence of that joint point.

[0037] To compress data as much as possible while preserving motion accuracy, more motion data needs to be retained when user actions are complex and have low regularity to ensure that the compressed motion is not distorted. Conversely, when user actions are relatively simple or highly regular, only key motion information needs to be extracted to reduce storage costs. Therefore, the size of the compression window needs to be adjusted according to the coordinate changes of each joint.

[0038] Since the frequency of movement may vary at different hand joint positions, it is necessary to analyze the coordinate data of each joint point separately. Here, we will take one joint point as an example for analysis.

[0039] First, based on the overall changes in the user's actions within the coordinate point sequence of a single key point, the threshold for changes in the actions at that key point is analyzed. Since the coordinates of each key point at each monitoring moment are three-dimensional, to facilitate the analysis of the temporal complexity of action changes, the temporal changes of data under each type of coordinate (coordinates in each dimension) are analyzed separately. Therefore, a temporal coordinate sequence is obtained by arranging one type of coordinate of each point in the coordinate point sequence of a single key point in chronological order. For example, the temporal coordinate sequence of the x-axis coordinates of each point in the coordinate point sequence of a single key point is obtained by arranging the x-axis coordinates in chronological order.

[0040] When the position of a joint remains stationary or changes regularly in a certain direction, it can be reproduced based solely on its initial position and movement pattern over that time period. Therefore, it can be divided into a compressed window, and key motion information can be extracted to represent the changes in motion within that window. Because there may be some tremors during hand movements, to avoid misidentifying these slight tremors as changes in hand movements, the exponential moving average method is first used to smooth one coordinate in the time-series coordinate sequence, resulting in a smoothed time-series coordinate sequence. Based on the smoothed time-series coordinate sequence, the slope value at the position of each data point is calculated.

[0041] The specific model for calculating the slope value is as follows:

[0042] ;

[0043] in, This represents the slope value of the i-th data point in the smoothed time-series coordinate sequence corresponding to the x-coordinate of the a-th joint. This represents the value of the i-th data point in the smoothed time-series coordinate sequence corresponding to the x-coordinate of the a-th joint, which is also the smoothed value of the i-th x-coordinate in the time series. Let represent the value of the (i-1)th data point in the smoothed time-series coordinate sequence corresponding to the x-coordinate of the a-th joint. and This represents the monitoring time corresponding to the i-th data point and its adjacent preceding data point. When i=1, the slope value of the first data point is represented by the slope value of the next adjacent data point, i.e. .

[0044] This allows us to obtain the slope value of each data point in the smoothed time-series coordinate sequence corresponding to each coordinate in the coordinate sequence of a key point.

[0045] The slope values ​​of each data point in the smoothed time-series coordinate sequence are arranged in temporal order to obtain a slope value sequence. The slope values ​​in the slope value sequence are then clustered to obtain the clustering results. Based on the clustering results, the smoothed time-series coordinate sequence is divided into different coordinate data segments. The clustering algorithm used is the k-means clustering algorithm, and the optimal k value is determined using the elbow method.

[0046] Furthermore, linear fitting is performed on each coordinate in each data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate to obtain the fitted coordinate value for each coordinate in each data segment. The algorithm for linear fitting is the least squares method. Further, the normal fluctuation threshold for this type of coordinate is obtained based on the difference between each coordinate in each data segment of the smoothed time-series coordinate sequence and the fitted coordinate value for each coordinate. Similarly, the normal fluctuation thresholds for the other two types of coordinates are obtained.

[0047] The normal fluctuation threshold of this type of coordinate is obtained by comparing the fitted values ​​of each coordinate with the fitted values ​​of each coordinate in the smoothed time-series coordinate data segments corresponding to this type of coordinate. The difference between the fitted value and the actual value can be used to represent the fluctuation of each coordinate. Specifically, the absolute value of the difference between a coordinate in the smoothed time-series coordinate sequence and its fitted value is taken as the fluctuation amplitude of that coordinate.

[0048] The specific calculation model for fluctuation amplitude is as follows:

[0049] ,

[0050] in, This represents the fluctuation amplitude of the x-coordinate of the i-th coordinate in the u-th coordinate data segment of the smoothed time-series coordinate sequence corresponding to the a-th key point. Let represent the i-th coordinate in the u-th coordinate data segment of the smoothed time-series coordinate sequence corresponding to the a-th joint, where the coordinate is the x-coordinate. Let represent the fitted coordinate value of the i-th coordinate in the u-th coordinate data segment of the smoothed time-series coordinate sequence corresponding to the a-th joint, where | represents the absolute value sign.

[0051] Furthermore, obtain the maximum fluctuation amplitude in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate; calculate the average of the maximum fluctuation amplitudes in all coordinate data segments as the normal fluctuation threshold for this type of coordinate, which is also the normal fluctuation threshold for the x-coordinate. Then, based on the above steps, the normal fluctuation thresholds for the y-coordinate and z-coordinate can also be obtained.

[0052] Step S2: Based on the normal fluctuation threshold of various coordinates, all coordinate points are filtered to obtain coordinate points of motion change; the coordinate point sequence is divided according to the coordinate points of motion change to obtain different data segments to be compressed.

[0053] The above steps obtained the normal fluctuation threshold for each coordinate in the coordinate point sequence. Since the flexibility of hand joints may vary in different directions, and hand movements may only occur in a certain direction, the fluctuation amplitude of each coordinate in each type of coordinate in the current data to be stored (coordinate point sequence) is calculated separately. Taking the x-coordinate as an example, all coordinates in the smoothed time-series coordinate sequence composed of x-coordinates are linearly fitted using the least squares method according to "time - x-coordinate" to obtain the coordinate fitting value for each coordinate. The absolute value of the difference between the coordinate fitting value and the actual value is the fluctuation amplitude of that coordinate (by default, the first coordinate is the start time of the action, and this process is not analyzed). Similarly, the fluctuation amplitude of each coordinate in the y-coordinate and the fluctuation amplitude of each coordinate in the z-coordinate are obtained.

[0054] Furthermore, the fluctuation amplitude of each coordinate at each coordinate point is compared with the normal fluctuation threshold of each coordinate. If the fluctuation amplitude of at least one coordinate at a coordinate point is greater than the normal fluctuation threshold of that coordinate, then that coordinate point is a coordinate point of motion change.

[0055] Furthermore, each action change coordinate point is used as a segmentation point to divide the coordinate point sequence into different data segments to be compressed. Thus, based on the user's action changes, the coordinate point sequence is divided into different data segments to be compressed. Shorter data segments indicate faster action changes and lower continuity during that period, making the analysis of its linkage more random and yielding less compression benefit. Therefore, data in data segments shorter than 3 units are directly stored.

[0056] Step S3: Combine two data segments to be compressed that have an inclusive or equal time relationship in the data segments to be compressed corresponding to two key points into a combination to be analyzed; extract the parts of the two data segments to be compressed that have equal time in the combination to be analyzed, and record them as the first data segment and the second data segment, and combine them into a data segment combination.

[0057] When a user's hand moves, the position of each joint does not change independently, but rather there is a certain correlation, meaning that different joints exhibit some relatively fixed movement patterns. Therefore, by identifying the repetitive movement patterns of the user based on the linkage between different joints of the user's hand, and merging and storing the linked and changing joint data, the amount of data stored can be effectively reduced compared to storing them separately.

[0058] Based on the above steps, the data segments to be compressed for each joint point within 1 second are obtained. If the data segments to be compressed for different joint points overlap in time interval, that is, if the times in the data segments to be compressed for these two joint points are inclusive or equal, for example, if the time in one data segment to be compressed is equal to the time in another data segment to be compressed, or if the time in one data segment to be compressed includes the time in another data segment to be compressed, the two data segments to be compressed for these two joint points are combined into a combination to be analyzed. There may be linkage changes between the two data segments to be compressed for these two joint points. Thus, the combination to be analyzed between each pair of joint points is obtained. Since hand joints are usually connected in pairs, the combination to be analyzed here only refers to the data corresponding to each pair of joint points. If multiple joint points have the above-mentioned overlapping time intervals, then pairwise analysis is performed separately.

[0059] Furthermore, the lengths of the two data segments to be compressed within a given combination of data points may be the same or different. In this case, for two data segments of different lengths, the portions with equal durations are extracted and designated as the first and second data segments. These first and second data segments are then combined into a single data segment for subsequent analysis. For data segments of equal length, the two data segments are directly designated as the first and second data segments. This process yields the data segment combinations between every two key points. The remaining portions are then used for further analysis, essentially treated as new data segments to be compressed and matched with the data segments from other key points before proceeding with the analysis.

[0060] Step S4: Based on the differences between various coordinates between coordinate points at the same time in the first and second data segments, obtain the relative position vectors of various coordinates of the data segment combination; determine whether the data segment combination is a linked combination based on the relative position vectors of various coordinates of a data segment combination.

[0061] The above steps obtain the data segment combinations between every two key points, and further analysis is performed on these data segment combinations. The relative positional relationship between the first and second data segments at corresponding times within a data segment combination is analyzed.

[0062] Specifically, the difference between a coordinate at a certain moment in the first data segment and a coordinate at the same moment in the second data segment is taken as the relative position of that coordinate at that moment. The relative positions of that coordinate at each moment are combined to form a relative position vector of that coordinate. Similarly, the relative position vectors of various coordinates of the data segment combination are obtained.

[0063] The specific calculation model for relative position is as follows:

[0064] ,

[0065] in, It represents the relative position of the x-axis coordinate at the i-th moment in the first data segment corresponding to joint A in a data segment combination, relative to the x-axis coordinate at the i-th moment in the second data segment corresponding to joint B. In other words, it is the relative position of the x-coordinate at the i-th moment in the first data segment and the second data segment. This represents the x-coordinate of the i-th time point in the second data segment corresponding to joint point B. This represents the x-axis coordinate at the i-th moment in the second data segment corresponding to joint point A.

[0066] Thus, the relative position vectors of the x-coordinates of the data segment combinations are obtained respectively. This allows us to obtain the relative position vector of the y-coordinate. The relative position vector of the z-coordinate .

[0067] Furthermore, a relative position data point is formed by taking the relative position vector of each coordinate in a data segment combination at a certain time and forming a relative position data point. The relative position data points are then sorted in chronological order to obtain the relative position data sequence of the data segment combination. The relative position of each coordinate at each time is calculated in the historical coordinate points of the two key points corresponding to the data segment combination, and the relative position data points at each historical time are formed into a historical relative position data sequence. A window with the same length as the first and second data segments in the data segment combination is slid across the historical relative position data sequence. The sequence of relative position data points within a window during the sliding process is recorded as the window relative position sequence of that window. The relative position similarity of the window is calculated based on the Euclidean distance between the relative position data points within the window relative position sequence of a window and the relative position data points at the corresponding time in the relative position data sequence of the data segment combination. If the relative position similarity of at least one window is greater than a preset threshold, then the data segment combination is a linked combination.

[0068] The specific calculation model for relative position similarity is as follows:

[0069] ,

[0070] in, This represents the similarity between the relative position data sequence of the data segments corresponding to joint points A and B and the relative position data points within a window when the window slides on the historical relative position data sequence corresponding to the coordinate points of these two joint points in history; that is, the relative position similarity corresponding to the window. norm represents the normalization operation; n represents the number of relative position data sequences or relative position data points within the window. , and These represent the relative positions of the i-th relative position data point in the relative position data sequence of the data segment combination corresponding to joint point A and joint point B, respectively, corresponding to the relative positions of the x-coordinate, y-coordinate, and z-coordinate. , and These represent the relative positions of the i-th relative position data point in the window relative position sequence corresponding to a window, namely the x-coordinate, y-coordinate, and z-coordinate, respectively. The Euclidean distance between the i-th relative position data point in the relative position data sequence of the data segment combination corresponding to node A and key point B and the i-th relative position data point in the window's relative position sequence is represented. It represents the sum of the Euclidean distances between the relative positions of data points at corresponding times in two sequences, which is also the total distance; This represents a hyperparameter, used to prevent the denominator from being 0.

[0071] It should also be noted that the window sliding step size is the length of the window, and its preset threshold value is 0.5. This is because the normalized relative position similarity values ​​obtained by non-linked and linked combinations will be distributed at both ends of the range [0,1]. Therefore, the preset threshold value is set to 0.5. When it is greater than 0.5, it is considered that the relative position change of the data segment combination corresponding to a certain period of time is similar to that of a corresponding period of time in history. That is, the two key points are in linkage change during that period of time.

[0072] This allows us to determine whether the combination of data segments obtained between each pair of joints is a linked combination.

[0073] Step S5: Encode the relative position vectors of various coordinates of the linkage combination to obtain the action mode code; select any data segment in the linkage combination as the basic data segment and obtain the fitting slope value of each coordinate in the basic data segment; store the action mode code of the linkage combination, the initial coordinate point in the basic data segment, and the fitting slope value of each coordinate.

[0074] The above steps yielded a linkage combination similar to the historical changes in relative position. In such a linkage combination, the motion changes of the two joints are similar and fixed within the time period. Therefore, the relative position vectors of various coordinates of the linkage combination can be encoded to obtain the motion pattern code of the linkage combination. Here, the encoding is a compressed encoding, which is existing technology and will not be elaborated further.

[0075] Furthermore, for the two data segments (the first data segment and the second data segment) in a linkage combination, one data segment can be used as the base data segment for data storage. This is because during decoding, the content of the other data segment can be reconstructed based on the content of the base data segment and the action mode encoding. Therefore, any data segment in a linkage combination is used as the base data segment; linear fitting is performed on various indicators of all coordinate points in the base data segment to obtain the fitting slope value for each indicator; then, the initial coordinate point (the first coordinate point), the fitting slope value for each coordinate of the base data segment in the linkage combination, and the action mode encoding of the linkage combination are stored.

[0076] For example, for a linkage combination corresponding to joints A and B, the initial coordinates of the data segment belonging to joint A, the fitted slope values ​​of various coordinates, and the motion mode code of this linkage combination are stored. During subsequent reconstruction, the coordinates of each data segment belonging to joint A can be reconstructed based on the initial coordinates of the data segment belonging to joint A and the fitted slope values ​​of various coordinates. Then, by combining the coordinates of each data segment belonging to joint A and the motion mode code of this linkage combination, the coordinates of each data segment belonging to joint B can be reconstructed.

[0077] Additionally, if a data segment combination is a non-linked combination, the initial coordinate points of the two data segments to be compressed to which the data segment combination belongs, as well as the fitted slope value of each coordinate, are stored.

[0078] If the data segment combinations of joints A, B, and C in the same time period are all in a linked change, that is, the data segment combination of joints A and C in that time period is a linked combination, the data segment combination of joints B and C in that time period is a linked combination, and the data segment combination of joints A and B in that time period is a linked combination, then any one of these three linked combinations is taken as the basic data segment. Then, the initial coordinate point and the fitting slope value corresponding to various coordinates of the basic data segment are stored, and the action mode code of the two linked combinations containing the basic data segment is also stored. For example, among the six data segments in these three linked combinations, the data segment belonging to joint A is taken as the basic data segment. At this time, the initial coordinate point and the fitting slope value corresponding to various coordinates of the data segment belonging to joint A are stored. For the linked combination corresponding to joints B and A, the action mode code of the linked combination corresponding to joints B and A is stored (the action mode code corresponding to joints B and A in that time period). For the linked combination corresponding to joints C and A, the action mode code of the linked combination corresponding to joints C and A is stored (the action mode code corresponding to joints C and A in that time period).

[0079] In subsequent storage, if a linkage combination with a similar action pattern to an already stored linkage combination appears, such as a linkage combination between joint A and joint B whose relative position change is similar to the action pattern of a previously stored linkage combination between these two joints, then the initial coordinate point of the data segment belonging to joint A, the fitted slope values ​​of various coordinates, and the action pattern code of the already stored linkage combination (for joint B) are directly stored in the subsequent linkage combination.

[0080] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0081] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A VR interactive data management method based on artificial intelligence, characterized in that, The method includes: collecting three-dimensional coordinate points of each joint at each moment within a preset time period, forming a coordinate point sequence for each joint; analyzing the changes of various coordinates of each coordinate point in a coordinate point sequence over time to obtain normal fluctuation thresholds for various coordinates; filtering all coordinate points based on the normal fluctuation thresholds to obtain motion change coordinate points; dividing the coordinate point sequence according to the motion change coordinate points to obtain different data segments to be compressed; combining two data segments to be compressed that have an inclusive or equal time relationship in the data segments to be compressed corresponding to two joints into an analysis combination; and extracting two data segments to be compressed from the analysis combination. The data segments with equal time intervals are denoted as the first data segment and the second data segment, and these segments are combined into a data segment combination. The relative position vectors of various coordinates in the data segment combination are obtained based on the differences between coordinates at the same time points in the first and second data segments. The relative position vectors of various coordinates in a data segment combination are used to determine whether the combination is a linked combination. The relative position vectors of various coordinates in the linked combination are encoded to obtain the action pattern code. Any data segment in the linked combination is selected as the base data segment, and the fitting slope value for each coordinate in the base data segment is obtained. The action pattern code of the linked combination and the base data segment are then combined. The initial coordinate points and the fitted slope values ​​for each coordinate are stored. The step of determining whether a data segment combination is a linked combination based on the relative position vectors of various coordinates in the data segment combination includes: forming a relative position data point from the relative position vectors of each coordinate in the data segment combination at a given time moment, and sorting the relative position data points according to time order to obtain the relative position data sequence of the data segment combination; calculating the relative position of each coordinate at each time moment in the historical coordinate points of the two key points corresponding to the data segment combination, and forming a relative position data sequence from the historical time moments. Historical relative position data sequence; using a window of the same length as the first and second data segments in the data segment combination, slide on the historical relative position data sequence. During the sliding process, the sequence of relative position data points within a window is recorded as the window relative position sequence of that window. The relative position similarity corresponding to the window is calculated based on the Euclidean distance between the relative position data points within the window relative position sequence of a window and the relative position data points at the corresponding time in the relative position data sequence of the data segment combination. If the relative position similarity corresponding to at least one window is greater than a preset threshold, then the data segment combination is a linked combination.

2. The VR interactive data management method based on artificial intelligence according to claim 1, characterized in that, The method of analyzing the changes of various coordinates of each coordinate point in a coordinate point sequence over time to obtain the normal fluctuation threshold of various coordinates includes: arranging one type of coordinate of each coordinate point in the coordinate point sequence of a key point in chronological order to obtain a time-series coordinate sequence; smoothing the coordinates in the time-series coordinate sequence corresponding to this type of coordinate using the exponential moving average method to obtain a smoothed time-series coordinate sequence; calculating the slope value of each data point in the smoothed time-series coordinate sequence; arranging the slope values ​​of each data point in the smoothed time-series coordinate sequence in chronological order to obtain a slope value sequence; clustering the slope values ​​in the slope value sequence to obtain clustering results; dividing the smoothed time-series coordinate sequence into different coordinate data segments based on the clustering results; performing linear fitting on each coordinate in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate to obtain the coordinate fitting value of each coordinate in each coordinate data segment; obtaining the normal fluctuation threshold of this type of coordinate based on the difference between the coordinate fitting values ​​of each coordinate in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to this type of coordinate; similarly, obtaining the normal fluctuation threshold of other types of coordinates.

3. The VR interactive data management method based on artificial intelligence according to claim 2, characterized in that, The step of obtaining the normal fluctuation threshold of a coordinate based on the difference between the fitted values ​​of each coordinate and each coordinate in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to that coordinate includes: taking the absolute value of the difference between a coordinate and its fitted value in the smoothed time-series coordinate sequence corresponding to that coordinate as the fluctuation amplitude of that coordinate; obtaining the maximum value of the fluctuation amplitude in each coordinate data segment of the smoothed time-series coordinate sequence corresponding to that coordinate; and taking the average of the maximum values ​​of the fluctuation amplitude in all coordinate data segments as the normal fluctuation threshold of that coordinate.

4. The VR interactive data management method based on artificial intelligence according to claim 1, characterized in that, The step of filtering all coordinate points based on the normal fluctuation threshold of various coordinates to obtain action change coordinate points includes: performing linear fitting on all coordinates of a smoothed time-series coordinate sequence composed of a certain type of coordinate to obtain the coordinate fitting value of each coordinate; taking the absolute value of the difference between a coordinate and the coordinate fitting value of that coordinate as the fluctuation amplitude of that coordinate; if there is at least one coordinate in a coordinate point whose fluctuation amplitude is greater than the normal fluctuation threshold of that type of coordinate, then that coordinate point is an action change coordinate point.

5. The VR interactive data management method based on artificial intelligence according to claim 1, characterized in that, The method of obtaining the relative position vectors of various coordinates of the data segment combination based on the difference between various coordinates of coordinate points at the same time in the first data segment and the second data segment includes: taking the difference between a coordinate of a certain time in the first data segment and the same coordinate of that time in the second data segment as the relative position of that coordinate at that time; forming a relative position vector of that coordinate by the relative positions of the coordinates of that type at each time; and similarly obtaining the relative position vectors of various coordinates of the data segment combination.

6. The VR interactive data management method based on artificial intelligence according to claim 1, characterized in that, The step of calculating the relative position similarity of a window based on the Euclidean distance between each relative position data point in the window relative position sequence of a window and the relative position data point at the corresponding time in the relative position data sequence of the data segment combination includes: calculating the sum of the Euclidean distances between each relative position data point in the window relative position sequence of a window and the relative position data point at the corresponding time in the relative position data sequence of the data segment combination, denoted as the total distance; and obtaining the relative position similarity by taking the reciprocal of the sum of the total distance and the hyperparameter and normalizing it.

Images