Non-abandoned inheritance display method and system based on artificial intelligence

Abstract

The invention discloses a non-abandoned inheritance display method and system based on artificial intelligence, and particularly relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring dynamic detail time sequence data from a capture system in real time, analyzing track smoothness by using Fourier transform, generating a lossless reference model by using ultrahigh-resolution 3D scanning and image equipment, and performing texture consistency analysis on a difference salient region by combining point cloud registration and a difference heat map to evaluate spatial deviation of a track. Judging the fidelity of dynamic details, comprehensively evaluating the capture accuracy in combination with the smoothness and the fidelity, dividing capture results into accurate capture and inaccurate capture, and archiving the accurately captured data to realize efficient reuse; for data which are not accurately captured, track dynamic complementation is carried out through an autoregression integral moving average model, the accuracy and integrity of non-abandoned dynamic detail capturing are remarkably improved, it is ensured that the display effect is real and coherent, and accurate inheritance and wide propagation of non-abandoned cultural connotation are facilitated.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intangible cultural heritage inheritance display method and system based on artificial intelligence. Background Art

[0002] AI-based intangible cultural heritage display refers to the use of AI technology to digitize and intelligently display and disseminate the content, form and cultural connotation of intangible cultural heritage (ICH) to promote the protection and inheritance of ICH. This method reconstructs the expression of ICH through AI technologies such as machine learning, computer vision and natural language processing, making it more vivid, modern and interactive. For example, through the generated digital models, speech synthesis and virtual reality (VR) technology, users can experience the charm of traditional handicrafts, dance or music in an immersive way. This display method not only improves the dissemination efficiency of ICH, but also reduces geographical and time constraints, allowing more people to access and understand ICH.

[0003] The prior art has the following deficiencies: Intangible cultural heritage projects often have complex dynamic characteristics, such as the delicate movements of a dance or the fine details of a craft production process. During 3D modeling and dynamic rendering, if the algorithm cannot accurately capture these details, the virtual experience will lack authenticity and lose the meaning of the intangible cultural heritage display. Especially in the process of real-time interaction, distortion of movements or changes will cause a sense of disconnection between users and content. In addition, the value of intangible cultural heritage projects lies in their cultural authenticity and uniqueness. For example, each movement of a dance may contain specific cultural meanings, and the production steps of handicrafts embody the essence of traditional skills. If the details are distorted, users will be exposed to incomplete or incorrect forms of expression, which may misunderstand the cultural connotation of intangible cultural heritage and even lead to incorrect cultural inheritance. Summary of the invention

[0004] The purpose of the present invention is to provide an intangible cultural heritage inheritance display method and system based on artificial intelligence to solve the shortcomings of the background technology.

[0005] In order to achieve the above object, the present invention provides the following technical solution: an intangible cultural heritage inheritance display method based on artificial intelligence, comprising the following steps: S1: acquiring time series data of dynamic details from a capture system in real time, wherein the time series data includes trajectory data of dance movements in the production of intangible cultural heritage crafts; S2: Decompose the trajectory data into frequency components using Fourier transform, calculate the proportion of high-frequency noise components, and determine whether the smoothness of the dance movement trajectory in space is consistent based on the abnormal fluctuation of the proportion of high-frequency noise components; S3: When the smoothness of the dance movement trajectory is inconsistent in space, the real scene is modeled through ultra-high-resolution 3D scanning equipment and imaging equipment to generate a lossless benchmark model. The point cloud registration algorithm is used to compare the dance movement trajectory data and the benchmark model data, and the difference value is mapped into a heat map; S4: Divide the heat map into areas with significant differences and areas with insignificant differences. For areas with significant differences, extract the trajectory data of the dance movements and the texture map of the benchmark model, compare the texture matching degree to calculate the texture consistency of the areas with significant differences, and judge the fidelity of the dynamic details. S5: According to the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details, the accuracy of dynamic detail capture is evaluated. According to the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture; S6: Accurately captured dynamic detail data are archived to facilitate reuse in different intangible cultural heritage display projects; for inaccurately captured data, an autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and subsequent trajectories.

[0006] Preferably, in S2, the trajectory data is decomposed into frequency components using Fourier transform, and the proportion of high-frequency noise components is calculated, specifically: Get the time series trajectory data of dance movements from the capture system, apply discrete Fourier transform to the trajectory data sequence, and set the frequency threshold , the low frequency component is , the high frequency component is , respectively calculate the high frequency energy and the total spectral energy , the expression is: ; where X(f) is the complex value of frequency f, f represents the frequency component, N is the length of the time series, and the high-frequency noise ratio is for: .

[0007] Preferably, in S3, judging whether the smoothness of the dance movement trajectory in space is consistent according to the abnormal fluctuation of the high-frequency noise component ratio is as follows: After analyzing the abnormal fluctuation of the high-frequency noise component ratio, the high-frequency noise ratio fluctuation index is generated. The method for obtaining the high-frequency noise ratio fluctuation index is as follows: , where t=1,2,...,Q represents the time point, is the high-frequency noise ratio at the tth moment, sets the sliding window length W, which represents the length of the time series for calculating the standard deviation each time, and the sliding step length Δt, which represents the time interval for each movement of the sliding window, and extracts multiple sliding windows from the time series: ;in: is a time point in the time series The proportion of high-frequency noise at , represents the starting time point of the kth window, for each sliding window , calculate the local standard deviation , the expression is: ;in, is the mean value within the window, and the standard deviation of all windows is calculated in sequence to obtain the volatility time series: ; M is the total number of windows, and the high-frequency noise ratio fluctuation index is defined as the mean of the standard deviations of all sliding windows, expressed as: ; In the formula, is the high-frequency noise proportional fluctuation index.

[0008] Preferably, in S3, the point cloud registration algorithm is used to compare the trajectory data of the dance action and the data difference of the benchmark model, and the difference value is mapped into a heat map, specifically: Use the motion capture system to obtain the trajectory point cloud of the dance movement: Pmotion={q1,q2,...,qi}, where qi=(xi,yi,zi) is the three-dimensional coordinate of the trajectory point. Align the dance movement trajectory point cloud Pmotion to the benchmark model point cloud Pbaseline, expressed as: ; Where R is the rotation matrix, t is the translation vector, It represents the result of initializing the transformation of the dance movement trajectory point cloud Pmotion by the rotation matrix R and the translation vector t, that is, the new position after the rotation and translation operations are performed on Pmotion; Fine registration using point-to-point ICP algorithm: Matching point pairs: Find the track point To the reference point The closest point pair of , optimizing the rigid transformation: ; represents the new coordinates obtained by applying the rigid transformation to the trajectory point qi, and m is the number of matching point pairs; The trajectory point cloud after registration , the expression is: ; In the formula, represents the optimal rigid transformation matrix, the rotation matrix calculated by the minimization process is used to transform the trajectory point cloud of the dance movement to a position aligned with the reference model; for each point , calculate its distance to the nearest reference point The Euclidean distance of: , where Indicates the difference between the trajectory point and the baseline model; Map the difference value di of each point to a color value Ci to construct a heat map: Ci = f(di); where f(di) is a color mapping function, and apply the generated color value Ci to the trajectory point cloud Paligned to generate a colored point cloud model.

[0009] Preferably, in S4, divide the heat map into a significantly different region and a non-significantly different region, specifically: Set the high difference threshold as δthreshold and the low difference threshold as δlow, and mark the points whose difference values exceed the threshold; divide the difference values into three levels and distinguish them by colors: low difference green: di ≤ δlow; medium difference yellow: δlow < di ≤ δthreshold; high difference red: di > δthreshold; Perform spatial clustering analysis on the difference points, mark the spatial regions where the high difference points are located, and the clustering result is expressed as: C = {C1, C2,..., Ck}; where Ck is a significantly different region.

[0010] Preferably, in S4, for the significantly different region, extract the trajectory data of the dance movement and the texture map of the reference model, compare the texture matching degree to calculate the texture consistency of the significantly different region, and judge the fidelity of the dynamic details, specifically: Map each trajectory point qi of the dance movement to the surface of the reference model to obtain the surface texture data Tmotion(qi) of the dance movement, extract the texture map Tbaseline(sp) from the reference model, and map it to the corresponding 3D model surface according to the coordinates of the surface point cloud. The expression is: Tbaseline(sp) = Texture Mapping(Pbaseline, sp); where sp is a certain surface point in the reference model, and Tbaseline(sp) is the corresponding texture map data; For each pair of corresponding points qi and sp in the significantly different region, calculate the similarity of their texture images , and the expression is: ; where represents the texture image of the dance movement, represents the texture image of the reference model, is the mean of the local brightness of the image , is the mean of the local brightness of the image , is the standard deviation of the local brightness of the image, representing the contrast of the image , is the standard deviation of the local brightness of the image, representing the contrast of the image is the mean of the local brightness of the image , is the standard deviation of the local brightness of the image, representing the contrast of the image Representing images and images The local covariance of and is a constant; calculate the texture similarity of each point in the area with significant differences and obtain the consistency index of the regional texture , the expression is: ;in, Indicates the area with significant difference. is the number of points in the region; Dynamic detail fidelity refers to the consistency and true restoration of the texture of the dance movement and the texture of the benchmark model in the area where the difference is significant. The calculation expression is: ; In the formula, For dynamic detail fidelity, is the mean square error, and the calculation expression is: ; Where B is the number of comparison points.

[0011] Preferably, in S5, the accuracy of capturing the dynamic details is evaluated according to the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details; The high-frequency noise proportion fluctuation index and dynamic detail fidelity are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model takes the accuracy value label of dynamic detail capture predicted by each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of prediction errors of all accuracy value labels of dynamic detail capture as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The accuracy value of dynamic detail capture is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0012] Preferably, based on the evaluation results, the accuracy of capturing dynamic details is divided into two categories: accurate capture and inaccurate capture. Specifically, the acquired accuracy value of capturing dynamic details is compared with a reference threshold of the accuracy value of capturing dynamic details that is pre-set based on historical data. If the accuracy value of capturing dynamic details is greater than or equal to the reference threshold of the accuracy value, it means that the accuracy of capturing dynamic details is high. In this case, no warning signal is generated, and the accuracy of capturing dynamic details is divided into accurate capture. If the accuracy value of capturing dynamic details is less than the reference threshold of the accuracy value, it means that the accuracy of capturing dynamic details is low. In this case, a warning signal is generated, and the accuracy of capturing dynamic details is divided into inaccurate capture.

[0013] Preferably, in S6, for inaccurate capture, an autoregressive integrated moving average model is used to dynamically complete the missing dynamic details according to the captured previous and next trajectories, specifically: Get a trajectory time series from a capture system: ; where t=1,2,...,D, t is the time frame, is the three-dimensional coordinate of the t-th frame, marking the time frame that is not accurately captured; ; Split the complete trajectory, where the point sequence is known: ; Missing point sequence: ; Autoregression is obtained by linear regression of the current value with the values ​​of the previous time frames: ; In the formula, is the trajectory value of the current time frame, is the autoregressive coefficient, which is used to represent the contribution of the first p trajectory values. for The track value of the time frame, is the error term; the sliding average corrects the current value by the residuals of the previous time frames: ; is the sliding average coefficient, which is used to correct the predicted value. For time The error value at the time; the overall ARIMA model is expressed as: ARIMA (p, d, q); where: p is the autoregressive order, indicating how many previous values ​​are used, d is the difference order, indicating the number of difference operations, and q is the sliding average order, indicating how many residual values ​​are used; For non-stationary sequences, perform d difference operations to obtain a stationary sequence , the expression is: ; Generate initial prediction input data: ; Train the model separately for each trajectory dimension: ; Optimize the model parameters ϕ and θ through maximum likelihood estimation, and predict the value of the missing point based on the trained model: For each missing point Point-by-point prediction replaces the missing points in the original data with the predicted values. The expression is: , represents the final point cloud, It is a newly added trajectory point, which indicates the point added in the process of capturing dynamic details.

[0014] The present invention also provides an intangible cultural heritage inheritance display system based on artificial intelligence, including a dynamic detail capture module, a trajectory smoothness analysis module, a difference analysis module, a dynamic detail fidelity analysis module, a dynamic detail capture accuracy assessment module, and a data archiving and display module; Dynamic detail capture module: acquires time series data of dynamic details from the capture system in real time, wherein the time series data includes trajectory data of dance movements in the production of intangible cultural heritage crafts; Trajectory smoothness analysis module: Use Fourier transform to decompose trajectory data into frequency components, calculate the proportion of high-frequency noise components, and judge whether the smoothness of the dance movement trajectory in space is consistent based on the abnormal fluctuation of the proportion of high-frequency noise components; Difference analysis module: When the smoothness of the dance movement trajectory is inconsistent in space, the real scene is modeled through ultra-high-resolution 3D scanning equipment and imaging equipment to generate a lossless benchmark model. The point cloud registration algorithm is used to compare the dance movement trajectory data with the data difference of the benchmark model, and the difference value is mapped into a heat map; Dynamic detail fidelity analysis module: The heat map is divided into areas with significant differences and areas with insignificant differences. For areas with significant differences, the trajectory data of the dance movements and the texture map of the benchmark model are extracted, and the texture consistency of the areas with significant differences is calculated by comparing the texture matching degree to determine the fidelity of the dynamic details. Dynamic detail capture accuracy evaluation module: evaluates the accuracy of dynamic detail capture based on the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details. Based on the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture; Data archiving and display module: Accurately captured dynamic detail data is archived to facilitate reuse in different intangible cultural heritage display projects; for inaccurate capture, the autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and subsequent trajectories.

[0015] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention captures the trajectory data of intangible cultural heritage dance in real time and uses Fourier transform to analyze the smoothness. It further generates a benchmark model for the trajectory with inconsistent smoothness through ultra-high resolution 3D scanning, and uses a point cloud registration algorithm to locate the difference and generate a difference heat map. The texture consistency calculation and dynamic detail fidelity evaluation of the area with significant differences ensure the authenticity of the capture results and the restoration of cultural connotations. The polynomial regression model is used to comprehensively analyze the trajectory smoothness and fidelity, accurately evaluate the accuracy of dynamic detail capture, and then realize accurate data archiving and intelligent completion of inaccurate data, significantly improving the overall quality of dynamic detail capture and display.

[0016] 2. While ensuring the high fidelity of the dynamic details of the intangible cultural heritage projects, the present invention realizes the systematic storage and efficient reuse of the captured data, meeting the needs of multi-scenario intangible cultural heritage display. In addition, the present invention repairs the inaccurate captured data through intelligent completion technology, making it more coherent and complete, and providing technical support for the protection and inheritance of intangible cultural heritage. At the same time, the accuracy classification and early warning mechanism of dynamic detail capture enhances the intelligence and reliability of the system, provides important guarantees for the virtualization and digital application of intangible cultural heritage display, and promotes the innovative development of intangible cultural heritage projects in the field of digital technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 The figure is a flow chart of the method of the present invention.

[0019] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Example 1, please refer to Figure 1 and Figure 2 As shown, the intangible cultural heritage inheritance display method based on artificial intelligence described in this embodiment includes the following steps: S1: acquiring time series data of dynamic details from a capture system in real time, wherein the time series data includes trajectory data of dance movements in the production of intangible cultural heritage crafts; S2: Decompose the trajectory data into frequency components using Fourier transform, calculate the proportion of high-frequency noise components, and determine whether the smoothness of the dance movement trajectory in space is consistent based on the abnormal fluctuation of the proportion of high-frequency noise components; S3: When the smoothness of the dance movement trajectory is inconsistent in space, the real scene is modeled through ultra-high-resolution 3D scanning equipment and imaging equipment to generate a lossless benchmark model. The point cloud registration algorithm is used to compare the dance movement trajectory data and the benchmark model data, and the difference value is mapped into a heat map; S4: Divide the heat map into areas with significant differences and areas with insignificant differences. For areas with significant differences, extract the trajectory data of the dance movements and the texture map of the benchmark model, compare the texture matching degree to calculate the texture consistency of the areas with significant differences, and judge the fidelity of the dynamic details. S5: According to the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details, the accuracy of dynamic detail capture is evaluated. According to the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture; S6: Accurately captured dynamic detail data are archived to facilitate reuse in different intangible cultural heritage display projects; for inaccurately captured data, an autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and subsequent trajectories.

[0022] In S1, the time series data of dynamic details are obtained in real time from the capture system, and the time series data includes the trajectory data of the dance movements in the intangible cultural heritage craft production, specifically: The capture system needs to meet the requirements of high precision, real-time performance, and adaptability to complex dynamic characteristics. Common systems include: Optical motion capture system: such as Vicon and OptiTrack. By placing reflective markers (such as reflective balls) on the dancer, multiple high-speed cameras are used to capture the three-dimensional coordinate changes of these markers. Suitable for high-precision dance motion capture. Inertial motion capture system: such as Xsens and Perception Neuron. The dancer's limb movement data is captured through inertial sensors, without relying on external cameras, and is suitable for scenes with low space requirements. Depth camera and RGB camera: such as Microsoft Kinect and Intel RealSense. Use depth sensors and image processing algorithms to directly capture human skeleton data, suitable for low-cost real-time capture applications.

[0023] Identify the key parts of the dance movements (such as joints, fingertips, toes), and place markers on these parts. The more markers, the more detailed the trajectory data captured. For markerless capture systems, ensure that the device's field of view can fully cover the dancer's movement area. Multiple cameras are arranged in a circle to form a three-dimensional capture space covering the stage. Calibrate the camera position and angle to ensure that there are no blind spots in the capture area. Use calibration tools (such as calibration rods or calibration plates) to calibrate the capture system to ensure the consistency of the coordinate system of each camera.

[0024] The dancer performs according to the set movements, and the capture system records the 3D coordinates or skeleton data of each marker point in real time. The data format is usually a time series, and each frame contains the position data of all marker points. For example: ;in Indicates the timestamp, is the three-dimensional coordinate of the nth marker point. The time series data is converted into trajectory data, and the missing parts of the data are supplemented by interpolation algorithms. A trajectory curve is generated to intuitively display the motion path of the dance action.

[0025] Preprocess the data and use filtering algorithms (such as Kalman filtering) to remove high-frequency noise and ensure smooth trajectories. Convert the captured coordinates to a unified reference coordinate system for subsequent analysis and modeling. Transmit the captured time series data to the backend server or rendering engine in real time through data transmission protocols (such as UDP or WebSocket). Ensure low latency for data transmission (such as within 5 milliseconds) to support real-time dynamic rendering and analysis.

[0026] Verify the accuracy of the capture system for key movements (such as complex body changes and rapid movement switching). Check whether the time series data has obvious breakpoints or missing points to ensure the trajectory is consistent. By comparing the captured data with the benchmark model of dance movements, calculate the position error and trajectory deviation, and adjust the capture device parameters.

[0027] The trajectory data of dance movements is output in real time and provided to subsequent analysis and rendering modules for dynamic modeling, movement evaluation and cultural display. The data files are usually stored in a standard format (such as BVH, FBX or JSON) for cross-platform use.

[0028] S2: Use Fourier transform to decompose the trajectory data into frequency components, calculate the proportion of high-frequency noise components, and judge whether the smoothness of the dance movement trajectory in space is consistent based on the abnormal fluctuation of the proportion of high-frequency noise components.

[0029] The time series trajectory data of dance movements are obtained from the capture system, represented as key point trajectories in three-dimensional space, for example: ; where t is the time frame and n is the number of key points. Perform denoising on the trajectory data, such as using a sliding average or Kalman filter to remove outliers and invalid frames. Convert the trajectory into a single-dimensional data sequence in all directions to facilitate Fourier transform processing.

[0030] Apply a discrete Fourier transform (DFT) to the trajectory data sequence: ; where f represents the frequency component, N is the length of the time series, and X(f) is the complex value of frequency f. is the kernel function of Fourier transform, which calculates the amplitude (modulus) of the spectrum , the expression is: ; In the formula, They are the real and imaginary parts of the spectrum amplitude respectively, and a spectrum curve is drawn to show the energy distribution of different frequencies.

[0031] Set the frequency threshold according to the sampling rate and capture trajectory characteristics , low frequency components: , representing the main motion characteristics of the motion trajectory. High frequency components: , which usually represents noise and fast jitter in the trajectory data. Calculate the high-frequency noise ratio and high-frequency energy and the total spectral energy , the expression is: ; where X(f) is the complex value of frequency f, f represents the frequency component, N is the length of the time series, and the high-frequency noise ratio is for: . The higher it is, the more noise components there are in the trajectory and the worse the trajectory smoothness is.

[0032] After analyzing the abnormal fluctuation of the high-frequency noise component ratio, a high-frequency noise ratio fluctuation index is generated. The method for obtaining the high-frequency noise ratio fluctuation index is as follows: After analyzing the abnormal fluctuation of the high-frequency noise component ratio, the high-frequency noise ratio fluctuation index is generated. The method for obtaining the high-frequency noise ratio fluctuation index is as follows: , where t=1,2,...,Q represents the time point, is the high-frequency noise ratio at the tth moment, sets the sliding window length W, which represents the length of the time series for calculating the standard deviation each time, and the sliding step length Δt, which represents the time interval for each movement of the sliding window, and extracts multiple sliding windows from the time series: ;in: is a time point in the time series The proportion of high-frequency noise at , represents the starting time point of the kth window, for each sliding window , calculate the local standard deviation , the expression is: ;in, is the mean value within the window, and the standard deviation of all windows is calculated in sequence to obtain the volatility time series: ; M is the total number of windows, and the high-frequency noise ratio fluctuation index is defined as the mean of the standard deviations of all sliding windows, expressed as: ; In the formula, is the high-frequency noise proportional fluctuation index.

[0033] The larger the high-frequency noise ratio fluctuation index, the more significant the fluctuation amplitude of the high-frequency noise ratio, the greater the change in the smoothness of the trajectory in different time periods, and the presence of sudden or irregular trajectory disturbances. This shows that the smoothness of the dance movement trajectory is inconsistent in space, and obvious jitter, jump or abnormal curves may appear in the trajectory in some time periods. For the display of intangible cultural heritage dance, such a trajectory will weaken the coherence of the movement and the accurate expression of cultural connotations.

[0034] On the contrary, when the high-frequency noise ratio fluctuation index is smaller, it means that the change amplitude of the high-frequency noise ratio is lower, the smoothness of the trajectory in the time series tends to be consistent, and the overall movement trajectory shows stable and coherent dynamic characteristics. This shows that the dance movement trajectory is smoother and consistent in space. Such data can better restore the cultural connotation of dance movements and provide high-quality dynamic detail support for the display of intangible cultural heritage.

[0035] S3: When the smoothness of the dance movement trajectory is inconsistent in space, the real scene is modeled through ultra-high-resolution 3D scanning equipment and imaging equipment to generate a lossless benchmark model. The point cloud registration algorithm is used to compare the dance movement trajectory data and the data difference of the benchmark model, and the difference value is mapped into a heat map.

[0036] When the smoothness of dance movements is inconsistent in space, use structured light scanners, laser scanners, or multi-view photogrammetry equipment to capture high-resolution 3D data of the dancer. Use high-frame-rate cameras to synchronously record dance movements and capture dynamic texture data.

[0037] Apply point cloud filtering algorithms (such as voxel grid filtering) to remove noise points. Ensure that the coordinate system of the reference model is consistent with the coordinate system of the dance motion capture to facilitate subsequent registration. Save the generated point cloud in a standard format (such as PLY or OBJ).

[0038] Use the motion capture system to obtain the trajectory point cloud of the dance movement: Pmotion={q1,q2,...,qi}, where qi=(xi,yi,zi) is the three-dimensional coordinate of the trajectory point. Filter and time-synchronize the trajectory data to make it consistent with the sampling frequency of the benchmark model.

[0039] Align the dance motion trajectory point cloud Pmotion to the benchmark model point cloud Pbaseline to compare the differences. Use a rigid transformation algorithm (such as ICP initial alignment) to roughly align the point cloud: ; Where R is the rotation matrix, t is the translation vector, It represents the result of initializing the transformation of the dance movement trajectory point cloud Pmotion by the rotation matrix R and the translation vector t, that is, the new position after the rotation and translation operations are performed on Pmotion.

[0040] Use the point-to-point ICP (Iterative Closest Point) algorithm for fine alignment: Matching point pairs: Find the track point To the reference point The closest point pair of . Optimize rigid transformation: ; represents the new coordinates obtained by applying the rigid transformation to the trajectory point qi, and m is the number of matching point pairs.

[0041] The trajectory point cloud after registration , the expression is: ; In the formula, represents the optimal rigid transformation matrix, the rotation matrix calculated by the minimization process is used to transform the trajectory point cloud of the dance movement to a position aligned with the reference model; for each point , calculate its distance to the nearest reference point The Euclidean distance of: , where Indicates the difference between this trajectory point and the baseline model.

[0042] Map the difference value di of each point to the color value Ci and construct a heat map: Ci=f(di); where f(di) is the color mapping function, and the commonly used color gradient (such as blue-green-red) represents the size of the difference: blue: small difference. Red: large difference. Apply the color value Ci to the trajectory point cloud Paligned to generate a colored point cloud model.

[0043] S4: Divide the heat map into areas with significant differences and areas with insignificant differences. For areas with significant differences, extract the trajectory data of the dance movements and the texture map of the benchmark model, compare the texture matching degree to calculate the texture consistency of the areas with significant differences, and judge the fidelity of the dynamic details.

[0044] Observe the difference heat map and locate the areas where the trajectory is significantly different from the baseline model. For areas with large difference values, further analyze the accuracy of the captured data or recalibrate the baseline model. The difference heat map is output in 3D visualization form, which is convenient for dynamic visualization in the intangible cultural heritage display. At the same time, save the difference statistics (such as mean, standard deviation) for subsequent optimization analysis.

[0045] Set the high difference threshold to , marking points where the difference value exceeds the threshold: It can be set in the following ways: For example, the mean μd plus k times the standard deviation σd: ; Usually k is 2 or 3, or a fixed value is selected according to specific application requirements.

[0046] The difference values are divided into multiple levels (such as low, medium, and high differences) and distinguished by colors: Low difference (green): di ≤ δlow; Medium difference (yellow): δlow < di ≤ δthreshold; High difference (red): di > δthreshold; Perform spatial clustering analysis on the difference points (such as DBSCAN clustering), mark the spatial regions where the high-difference points are located, and the clustering result is expressed as: C = {C1, C2,..., Ck}; where Ck is a region with significant differences.

[0047] Map each trajectory point qi of the dance movement to the surface of the reference model to obtain the surface texture data Tmotion(qi) of the dance movement. Extract the texture map Tbaseline(sp) from the reference model and map it to the corresponding 3D model surface according to the coordinates of the surface point cloud. The expression is: Tbaseline(sp) = Texture Mapping(Pbaseline, sp); where sp is a certain surface point in the reference model, and Tbaseline(sp) is the corresponding texture map data; For each pair of corresponding points qi and sp in the region with significant differences, calculate the similarity of their texture images , and the expression is: ; where, represents the texture image of the dance movement, represents the texture image of the reference model, is the mean of the local brightness of the image , is the mean of the local brightness of the image , is the standard deviation of the local brightness of the image, representing the contrast of the image , is the standard deviation of the local brightness of the image, representing the contrast of the image is the mean of the local brightness of the image , is the standard deviation of the local brightness of the image, representing the contrast of the image represents the local covariance of the image and the image , and are constants; Calculate the texture similarity of each point in the region with significant differences to obtain the consistency index of the regional texture , and the expression is: ; where, represents the region with significant differences, is the number of points within the region; The dynamic detail fidelity refers to the consistency and true restoration degree of the texture of the dance movement and the texture of the reference model within the region with significant differences. The calculation expression is: ; In the formula, For dynamic detail fidelity, is the mean square error, and the calculation expression is: ; Where B is the number of comparison points.

[0048] High fidelity: If the fidelity index is close to 1, it means that the texture of the dance movement is highly consistent with the texture of the baseline model in the area of ​​significant difference, and the dynamic details are restored with high fidelity.

[0049] Medium fidelity: If the fidelity index is between 0.6 and 1, it means that the texture is fairly consistent, but there are still slight differences and dynamic details may be lost in some parts.

[0050] Low fidelity: If the fidelity index is lower than 0.6, it means that the texture of the dance movements is significantly different from the baseline model, and the dynamic details are severely distorted, affecting user experience and cultural heritage.

[0051] S5: According to the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details, the accuracy of dynamic detail capture is evaluated. According to the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture.

[0052] The high-frequency noise proportion fluctuation index and dynamic detail fidelity are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model takes the accuracy value label of dynamic detail capture predicted by each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of prediction errors of all accuracy value labels of dynamic detail capture as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The accuracy value of dynamic detail capture is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0053] The accuracy value of dynamic detail capture is obtained by obtaining the corresponding function expression from the comprehensive feature vector training data of the trained machine learning model: ; In the formula, is the output function of the model, is the high-frequency noise proportional fluctuation index, For dynamic detail fidelity, The accuracy value captured for dynamic details.

[0054] The acquired accuracy value of dynamic detail capture is compared with the reference threshold of the accuracy value of dynamic detail capture pre-set according to historical data. If the accuracy value of dynamic detail capture is greater than or equal to the reference threshold of the accuracy value, it means that the accuracy of dynamic detail capture is high. In this case, no warning signal is generated, and the accuracy of dynamic detail capture is classified as accurate capture; if the accuracy value of dynamic detail capture is less than the reference threshold of the accuracy value, it means that the accuracy of dynamic detail capture is low. In this case, a warning signal is generated, and the accuracy of dynamic detail capture is classified as inaccurate capture.

[0055] S6: Accurately captured dynamic detail data are archived to facilitate reuse in different intangible cultural heritage display projects; for inaccurately captured data, an autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and subsequent trajectories.

[0056] The dynamic detail data is classified according to the attributes of the intangible cultural heritage items for quick retrieval and application: Intangible cultural heritage categories: such as dance, craft production, music performance, etc. Action types: such as gestures, dance steps, process steps, etc. Dynamic characteristics: distinguish between smooth movements (such as rotation) and complex movements (such as fast jumping).

[0057] Use a unified and standardized data format for storage, which is convenient for cross-platform and multi-device use: Trajectory data: Use JSON, BVH or FBX format to record time series trajectory and skeletal animation information. Point cloud data: Use PLY or OBJ format to store high-resolution point cloud models. Texture data: Use PNG or JPEG format to store texture maps to ensure high image fidelity.

[0058] Add detailed meta information to the archived dynamic detail data to facilitate subsequent retrieval and understanding: Project description: Name of the intangible cultural heritage project. Movement source and cultural background (such as a specific dance or ritual). Capture device: Model and parameters of the capture device used (such as frame rate, resolution). Data attributes: Timestamp: Data generation time. Data size and accuracy (such as point cloud density, texture resolution). Quality assessment: High-frequency noise ratio fluctuation index. Dynamic detail fidelity index.

[0059] Establish a dynamic detail database of intangible cultural heritage and classify and store archived data: Relational database: Use MySQL or PostgreSQL to store trajectory data and metadata. Non-relational database: Use MongoDB, which is suitable for storing unstructured data such as point clouds and texture data. Store large-scale point clouds and texture files in a distributed file system (such as HDFS) to improve data reading efficiency.

[0060] Establish an efficient indexing mechanism to facilitate fast query and call: Establish primary indexes by intangible cultural heritage category and action type. Establish auxiliary indexes for dynamic detail features (such as trajectory complexity, texture area). Regularly check the integrity of archived data to ensure that the files are not damaged or lost. Verify whether the meta information is consistent with the dynamic detail data.

[0061] Dynamically update archived data, add new capture results or optimized versions: Version management: Create version numbers for different versions of the same action (such as the initial version and the optimized version). Differential storage: Use incremental storage technology to only record the difference with the baseline data. Provide convenient retrieval tools for archived dynamic detail data: Keyword-based search: such as "dance", "handicraft steps", etc. Feature-based search: Filter data based on trajectory smoothness or action complexity.

[0062] Develop standardized APIs to support the calling of dynamic detail data in different scenarios: Education platform: Provide detailed action decomposition and demonstration for intangible cultural heritage teaching. Virtual display: Provide high-resolution dynamic data for AR / VR applications. Game or animation production: Provide skeletal animation data for developers to quickly call.

[0063] For inaccurate capture, an autoregressive integrated sliding average model is used to dynamically complete the missing dynamic details based on the captured previous and next trajectories. Specifically: Get trajectory time series from capture system ; where t=1,2,...,D, t is the time frame, is the three-dimensional coordinate of the t-th frame, marking the time frame that is not accurately captured; ; Split the complete trajectory, where the point sequence is known: ; Missing point sequence: ; Autoregression is obtained by linear regression of the current value with the values ​​of the previous time frames: ; In the formula, is the trajectory value of the current time frame, is the autoregressive coefficient, which is used to represent the contribution of the first p trajectory values. for The track value of the time frame, is the error term; the sliding average corrects the current value by the residuals of the previous time frames: ; is the sliding average coefficient, which is used to correct the predicted value. For time The error value at the time; the overall ARIMA model is expressed as: ARIMA (p, d, q); where: p is the autoregressive order, indicating how many previous values ​​are used, d is the difference order, indicating the number of difference operations, and q is the sliding average order, indicating how many residual values ​​are used; For non-stationary sequences, perform d difference operations to obtain a stationary sequence , the expression is: ; Generate initial prediction input data: ; Train the model separately for each trajectory dimension: ; Optimize the model parameters ϕ and θ through maximum likelihood estimation, and predict the value of the missing point based on the trained model: For each missing point Point-by-point prediction replaces the missing points in the original data with the predicted values. The expression is: , represents the final point cloud, It is a newly added trajectory point, indicating the point added during the dynamic detail capture process. The completed trajectory is saved in a standardized format (such as JSON, BVH). It is used to reproduce the dynamic details of the intangible cultural heritage display. It provides complete data to support the teaching and research of intangible cultural heritage.

[0064] In this embodiment, after the dance movement trajectory data in the intangible cultural heritage craft production is acquired in real time by the capture system, the trajectory data is frequency decomposed by Fourier transform, the proportion of high-frequency noise components is calculated, and the spatial smoothness of the trajectory is determined by abnormal fluctuation analysis; if the trajectory smoothness is inconsistent, a lossless benchmark model is generated by an ultra-high resolution 3D scanning device, and the difference between the trajectory data and the benchmark model is compared by a point cloud registration algorithm, and the difference value is visualized in the form of a heat map; the heat map is further divided into areas with significant differences and areas with insignificant differences, and trajectory data and texture maps are extracted from the significant areas, and the texture matching degree is calculated to evaluate the fidelity of dynamic details; the accuracy of dynamic detail capture is evaluated in combination with the consistency and fidelity of trajectory smoothness, and the capture results are divided into two categories: accurate capture and inaccurate capture; for accurately captured data, systematic archiving is carried out for reuse, and for inaccurately captured data, an autoregressive integrated sliding average model is used to dynamically complete the missing details through the prediction of the previous and next trajectories, thereby comprehensively improving the quality of dynamic details and reuse value of the intangible cultural heritage display.

[0065] Embodiment 2, the intangible cultural heritage display system based on artificial intelligence described in this embodiment includes a dynamic detail capture module, a trajectory smoothness analysis module, a difference analysis module, a dynamic detail fidelity analysis module, a dynamic detail capture accuracy assessment module and a data archiving and display module; Dynamic detail capture module: acquires time series data of dynamic details from the capture system in real time, wherein the time series data includes trajectory data of dance movements in the production of intangible cultural heritage crafts; Trajectory smoothness analysis module: Use Fourier transform to decompose trajectory data into frequency components, calculate the proportion of high-frequency noise components, and judge whether the smoothness of the dance movement trajectory in space is consistent based on the abnormal fluctuation of the proportion of high-frequency noise components; Difference analysis module: When the smoothness of the dance movement trajectory is inconsistent in space, the real scene is modeled through ultra-high-resolution 3D scanning equipment and imaging equipment to generate a lossless benchmark model. The point cloud registration algorithm is used to compare the dance movement trajectory data with the data difference of the benchmark model, and the difference value is mapped into a heat map; Dynamic detail fidelity analysis module: The heat map is divided into areas with significant differences and areas with insignificant differences. For areas with significant differences, the trajectory data of the dance movements and the texture map of the benchmark model are extracted, and the texture consistency of the areas with significant differences is calculated by comparing the texture matching degree to determine the fidelity of the dynamic details. Dynamic detail capture accuracy evaluation module: evaluates the accuracy of dynamic detail capture based on the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details. Based on the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture; Data archiving and display module: Accurately captured dynamic detail data is archived to facilitate reuse in different intangible cultural heritage display projects; for inaccurate capture, the autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and subsequent trajectories.

[0066] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0067] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0068] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0069] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. The intangible cultural heritage display method based on artificial intelligence is characterized by: It includes the following steps: S1: Obtain time - series data of dynamic details in real - time from a capture system. The time - series data includes trajectory data of dance movements in the intangible cultural heritage craft production; S2: Use Fourier transform to decompose the trajectory data into frequency components, calculate the proportion of high - frequency noise components, and judge whether the smoothness of the dance movement trajectory in space is consistent according to the abnormal fluctuation of the proportion of high - frequency noise components; S3: When the smoothness of the dance movement trajectory in space is inconsistent, model the real scene through an ultra - high - resolution 3D scanning device and imaging device to generate a lossless reference model. Through the point - cloud registration algorithm, compare the data differences between the trajectory data of the dance movement and the reference model, and map the difference values to a heat map; S4: Divide the heat map into regions with significant differences and regions with insignificant differences. For regions with significant differences, extract the trajectory data of the dance movement and the texture map of the reference model, compare the texture matching degree to calculate the texture consistency of the regions with significant differences, and judge the fidelity of the dynamic details; S5: Evaluate the accuracy of dynamic detail capture according to the consistency of the smoothness of the dance movement trajectory in space and the fidelity of the dynamic details. According to the evaluation results, divide the accuracy of dynamic detail capture into two categories: accurate capture and inaccurate capture; S6: Archive the accurately captured dynamic detail data for repeated use in different intangible cultural heritage display projects; for inaccurate capture, use an autoregressive integrated moving average model to dynamically complete the missing dynamic details according to the captured front - and - back trajectories.

2. The intangible cultural heritage display method based on artificial intelligence according to claim 1, characterized in that: In S2, use Fourier transform to decompose the trajectory data into frequency components and calculate the proportion of high - frequency noise components. Specifically: Get the time series trajectory data of dance movements from the capture system, apply discrete Fourier transform to the trajectory data sequence, and set the frequency threshold , the low frequency component is , the high frequency component is , respectively calculate the high frequency energy and the total spectral energy , the expression is: ; where X(f) is the complex value of frequency f, f represents the frequency component, N is the length of the time series, and the high-frequency noise ratio is for: .

3. The intangible cultural heritage display method based on artificial intelligence according to claim 2 is characterized by: In S3, judge whether the smoothness of the dance movement trajectory in space is consistent according to the abnormal fluctuation of the proportion of high - frequency noise components. Specifically: After analyzing the abnormal fluctuation of the high-frequency noise component ratio, the high-frequency noise ratio fluctuation index is generated. The method for obtaining the high-frequency noise ratio fluctuation index is as follows: , where t=1,2,...,Q represents the time point, is the high-frequency noise ratio at the tth moment, sets the sliding window length W, which represents the length of the time series for calculating the standard deviation each time, and the sliding step length Δt, which represents the time interval for each movement of the sliding window, and extracts multiple sliding windows from the time series: ;in: is a time point in the time series The proportion of high-frequency noise at , represents the starting time point of the kth window, for each sliding window , calculate the local standard deviation , the expression is: ;in, is the mean value within the window, and the standard deviation of all windows is calculated in sequence to obtain the volatility time series: ; M is the total number of windows, and the high-frequency noise ratio fluctuation index is defined as the mean of the standard deviations of all sliding windows, expressed as: ; In the formula, is the high-frequency noise proportional fluctuation index.

4. The intangible cultural heritage display method based on artificial intelligence according to claim 3 is characterized by: In S3, through the point - cloud registration algorithm, compare the data differences between the trajectory data of the dance movement and the reference model, and map the difference values to a heat map. Specifically: Use a motion capture system to obtain the trajectory point cloud of the dance movement: Pmotion={q1,q2,...,qi}, where qi=(xi,yi,zi) is the three-dimensional coordinate of the trajectory point. The dance motion trajectory point cloud Pmotion is aligned to the benchmark model point cloud Pbaseline. The expression is: ; Where R is the rotation matrix, t is the translation vector, It represents the result of initializing the transformation of the dance movement trajectory point cloud Pmotion by the rotation matrix R and the translation vector t, that is, the new position after the rotation and translation operations are performed on Pmotion; Fine registration using point-to-point ICP algorithm: Matching point pairs: Find the track point To the reference point The closest point pair of , optimizing the rigid transformation: ; represents the new coordinates obtained by applying the rigid transformation to the trajectory point qi, and m is the number of matching point pairs; The trajectory point cloud after registration , the expression is: ; In the formula, represents the optimal rigid transformation matrix, the rotation matrix calculated by the minimization process is used to transform the trajectory point cloud of the dance movement to a position aligned with the reference model; for each point , calculate its distance to the nearest reference point The Euclidean distance of: , where Indicates the difference between the trajectory point and the baseline model; Map the difference value di of each point to a color value Ci to construct a heat map: Ci = f(di); where f(di) is a color mapping function, and apply the generated color value Ci to the trajectory point cloud Paligned to generate a colored point - cloud model.

5. The intangible cultural heritage display method based on artificial intelligence according to claim 4 is characterized by: In S4, divide the heat map into regions with significant differences and regions with insignificant differences. Specifically: Set a high - difference threshold as δthreshold and a low - difference threshold as δlow, and mark the points whose difference values exceed the threshold; divide the difference values into three levels and distinguish them by colors: low - difference green: di≤δlow; medium - difference yellow: δlow<di≤δthreshold; high - difference red: di>δthreshold; Conduct spatial clustering analysis on the difference points, mark the spatial regions where the high - difference points are located, and the clustering result is expressed as: C={C1,C2,...,Ck}; where Ck is a region with significant differences.

6. The intangible cultural heritage display method based on artificial intelligence according to claim 5 is characterized by: In S4, for the area with significant difference, the trajectory data of the dance movement and the texture map of the benchmark model are extracted, the texture consistency of the area with significant difference is calculated by comparing the texture matching degree, and the fidelity of the dynamic details is judged, specifically: Map each trajectory point qi of the dance movement to the surface of the benchmark model to obtain the surface texture data Tmotion(qi) of the dance movement. Extract the texture map Tbaseline(sp) from the benchmark model and map it to the corresponding 3D model surface according to the coordinates of the surface point cloud. For each pair of corresponding points qi and sp in the area with significant differences, calculate the similarity of their texture images. , the expression is: ;in, Represents a dance movement texture image, represents the benchmark model texture image, is an image The mean local brightness, is an image The mean local brightness, is an image The standard deviation of local brightness, indicating the image The contrast ratio, is an image The standard deviation of local brightness, indicating the image The contrast ratio, Representing images and images The local covariance of and is a constant; calculate the texture similarity of each point in the area with significant differences and obtain the consistency index of the regional texture , the expression is: ;in, Indicates the area with significant difference. is the number of points in the region; Dynamic detail fidelity refers to the consistency and true restoration of the texture of the dance movement and the texture of the benchmark model in the area where the difference is significant. The calculation expression is: ; In the formula, For dynamic detail fidelity, is the mean square error, and the calculation expression is: ; Where B is the number of comparison points.

7. The intangible cultural heritage display method based on artificial intelligence according to claim 6 is characterized by: In S5, the accuracy of dynamic detail capture is evaluated based on the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details; The high-frequency noise proportion fluctuation index and dynamic detail fidelity are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model takes the accuracy value label of dynamic detail capture predicted by each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of prediction errors of all accuracy value labels of dynamic detail capture as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The accuracy value of dynamic detail capture is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

8. The intangible cultural heritage display method based on artificial intelligence according to claim 7 is characterized by: According to the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture. Specifically: the acquired accuracy value of dynamic detail capture is compared with the reference threshold of the accuracy value of dynamic detail capture pre-set according to historical data. If the accuracy value of dynamic detail capture is greater than or equal to the reference threshold of the accuracy value, it means that the accuracy of dynamic detail capture is high. At this time, no warning signal is generated, and the accuracy of dynamic detail capture is divided into accurate capture; if the accuracy value of dynamic detail capture is less than the reference threshold of the accuracy value, it means that the accuracy of dynamic detail capture is low. At this time, a warning signal is generated, and the accuracy of dynamic detail capture is divided into inaccurate capture.

9. The intangible cultural heritage display method based on artificial intelligence according to claim 1, characterized in that: In S6, for inaccurate capture, an autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and next trajectories. Specifically: Get a trajectory time series from a capture system: ; where t=1,2,...,D, t is the time frame, is the three-dimensional coordinate of the t-th frame, marking the time frame that is not accurately captured; ; Split the complete trajectory into parts, where the known point sequence ; Missing point sequence ; Autoregression is obtained by linear regression of the current value with the values ​​of the previous time frames: ; In the formula, is the trajectory value of the current time frame, is the autoregressive coefficient, which is used to represent the contribution of the first p trajectory values. for The track value of the time frame, is the error term; the sliding average corrects the current value by the residuals of the previous time frames: ; is the sliding average coefficient, which is used to correct the predicted value. For time The error value at the time; the overall ARIMA model is expressed as: ARIMA (p, d, q); where: p is the autoregressive order, indicating how many previous values ​​are used, d is the difference order, indicating the number of difference operations, and q is the sliding average order, indicating how many residual values ​​are used; For non-stationary sequences, perform d difference operations to obtain a stationary sequence , generate initial prediction input data , the expression is: ; Train the model separately for each trajectory dimension: ; Optimize the model parameters ϕ and θ through maximum likelihood estimation, and predict the value of the missing point based on the trained model: For each missing point Point-by-point prediction replaces the missing points in the original data with the predicted values. The expression is: , represents the final point cloud, It is a newly added trajectory point, which indicates the point added in the process of capturing dynamic details.

10. An intangible cultural heritage display system based on artificial intelligence, used to implement the intangible cultural heritage display method based on artificial intelligence according to any one of claims 1 to 9, characterized in that: It includes dynamic detail capture module, trajectory smoothness analysis module, difference analysis module, dynamic detail fidelity analysis module, dynamic detail capture accuracy assessment module and data archiving and display module; Dynamic detail capture module: acquires time series data of dynamic details from the capture system in real time, wherein the time series data includes trajectory data of dance movements in the production of intangible cultural heritage crafts; Trajectory smoothness analysis module: Use Fourier transform to decompose trajectory data into frequency components, calculate the proportion of high-frequency noise components, and judge whether the smoothness of the dance movement trajectory in space is consistent based on the abnormal fluctuation of the proportion of high-frequency noise components; Difference analysis module: When the smoothness of the dance movement trajectory is inconsistent in space, the real scene is modeled through ultra-high-resolution 3D scanning equipment and imaging equipment to generate a lossless benchmark model. The point cloud registration algorithm is used to compare the dance movement trajectory data with the data difference of the benchmark model, and the difference value is mapped into a heat map; Dynamic detail fidelity analysis module: The heat map is divided into areas with significant differences and areas with insignificant differences. For areas with significant differences, the trajectory data of the dance movements and the texture map of the benchmark model are extracted, and the texture consistency of the areas with significant differences is calculated by comparing the texture matching degree to determine the fidelity of the dynamic details. Dynamic detail capture accuracy evaluation module: evaluates the accuracy of dynamic detail capture based on the smoothness consistency of the dance movement trajectory in space and the fidelity of the dynamic details. Based on the evaluation results, the accuracy of dynamic detail capture is divided into two categories: accurate capture and inaccurate capture; Data archiving and display module: Accurately captured dynamic detail data is archived to facilitate reuse in different intangible cultural heritage display projects; for inaccurate capture, the autoregressive integrated moving average model is used to dynamically complete the missing dynamic details based on the captured previous and subsequent trajectories.

Images