Skateboarding action recognition and interaction method based on LCSS-topological feature fusion

Abstract

This application relates to a skateboarding motion recognition and interaction method based on LCSS-topological feature fusion. This method combines sensor trajectory analysis, motion curvature extraction, coherent feature calculation, and statistical modeling to construct a multidimensional recognition framework that integrates motion trajectory similarity, topological structure information, and skateboarding motion complexity. The LCSS algorithm has the advantages of low computational complexity, low time alignment requirements, and strong noise resistance, making it suitable for identifying fast, irregular, and nonlinear motion sequences in skateboarding. The introduction of topological features and motion statistical features enables the system to perform more accurate structural modeling and personalized characterization of motion patterns. The introduction of a social feedback mechanism based on motion recognition achieves a quantitative fusion of technical performance and user interactive behavior, providing skateboarding enthusiasts with an integrated platform for skill evaluation, social sharing, and community interaction, expanding the application boundaries of motion recognition technology.

Description

Technical Field

[0001] The present application relates to the field of motion recognition technology, and in particular to a skateboarding motion recognition and interaction method based on LCSS-topological feature fusion. Background Art

[0002] With the continuous advancement of technologies such as human-computer interaction, intelligent sensing, and computer vision, sensor-based motion recognition technology has been widely applied in a variety of fields, including intelligent device control, virtual reality interaction, and rehabilitation medical monitoring. The mainstream technical approaches for motion recognition currently include: classification methods based on machine learning, sequence matching methods based on dynamic time warping (DTW), rule-based methods based on feature thresholds, and recognition methods based on human posture estimation.

[0003] While the aforementioned technologies have achieved remarkable results in general motion recognition tasks, existing approaches still face significant limitations in skateboarding, a highly dynamic, personalized, and socially interactive application scenario. Machine learning-based methods rely on a large number of labeled samples for model training, making it difficult to cover the diverse and personalized manifestations of skateboarding movements and resulting in limited model generalization. DTW-based methods have high computational complexity when matching long time series and lack robustness to high-frequency jitter and noise common in skateboarding movements. Feature threshold-based methods often employ manually set threshold recognition rules, failing to effectively extract and characterize the spatiotemporal topological structure of skateboarding movements and making it difficult to distinguish between similar but fundamentally different motion patterns. While pose estimation-based methods can extract fine-grained skeletal motion information, they lack the ability to structuredly model the overall trajectory, making it difficult to support the combined needs of trajectory recognition and social sharing in skateboarding. These issues have led to bottlenecks in the integrated application of skateboarding motion recognition and social interaction, making it difficult to provide a systematic solution that is both highly accurate and robust, while also deeply integrated with social scenarios. Summary of the Invention

[0004] Based on this, it is necessary to provide a skateboarding action recognition and interaction method based on LCSS-topological feature fusion to address the above technical problems.

[0005] A skateboarding action recognition and interaction method based on LCSS-topological feature fusion, the method comprising:

[0006] Step S1: obtaining skateboard motion data and preprocessing the skateboard motion data, the skateboard motion data including: three-axis acceleration and angular velocity of the skateboard;

[0007] Step S2: reconstructing the skateboard trajectory by using an integral operation with a damping term based on the preprocessed three-axis acceleration, and extracting multidimensional features from the reconstructed skateboard trajectory; the multidimensional features include: a complete position trajectory sequence, a direction sequence, motion statistical features, and direction change features.

[0008] Step S3: extracting topological feature parameters from the reconstructed skateboard trajectory to obtain topological features.

[0009] Step S4: Based on the skateboard action sequence and the action samples in the preset standard library, the LCSS algorithm is used to perform action matching to obtain an LCSS matching score.

[0010] Step S5: Perform weighted fusion of topological features, motion statistical features and LCSS matching scores to obtain the final action similarity score.

[0011] Step S6: Obtain the skateboarding action recognition result based on the final action similarity score.

[0012] Step S7: Based on the skateboarding action recognition results and the social feedback mechanism, accurate recognition of complex skateboarding actions and personalized social experience are achieved.

[0013] In one embodiment, pre-processing the skateboard motion data includes:

[0014] The skateboard motion data is low-pass filtered and gravity compensated to obtain the processed acceleration and angular velocity. The gravity compensation is used to separate the vertical gravity acceleration component from the original acceleration data through the rotation matrix reflecting the current posture. The gravity compensation formula is:

[0015] ;

[0016] in, is the three-axis acceleration of the skateboard after preprocessing, is the raw acceleration data after filtering, is the rotation matrix reflecting the current posture, θ is the three-dimensional attitude angle of the skateboard, g is the gravity vector, g =[0, 0, -9.81] T , is a matrix or vector multiplication.

[0017] In one embodiment, reconstructing the trajectory of the skateboard using an integral operation with a damping term based on the preprocessed three-axis acceleration includes:

[0018] The pre-processed three-axis acceleration is integrated by the integration operation with the introduction of the damping term to obtain the plate velocity; the integration operation with the introduction of the damping term is:

[0019] ;

[0020] in, for t Skateboard speed at the moment, for Skateboard speed at the moment, is the damping coefficient, is the three-axis acceleration of the skateboard after preprocessing; is the integration time interval.

[0021] The skateboard displacement is obtained by integrating the skateboard speed and the skateboard trajectory is reconstructed.

[0022] In one embodiment, the topological features include: curvature change rate, number of key points, and number of closed-loop structures. The curvature change rate is:

[0023] ;

[0024] in, for t Skateboard speed at the moment, For skateboard tracks t The curvature value at the moment, is a matrix or vector multiplication.

[0025] The number of key points is the number of elements in the key point set; the key point set is:

[0026] ;

[0027] in, K is a set of key points, including the moments when the curvature is greater than the threshold, is the curvature threshold.

[0028] The number of closed-loop structures is:

[0029] ;

[0030] in, is the number of closed-loop structures, used to measure topological characteristics, for k -dimensional cyclic group, which represents the number of rings in the topology, for k dimensional boundary group, indicating the number of boundaries in the topology, is the k-th dimension homology group.

[0031] In one embodiment, based on the skateboarding action sequence and the action samples in the preset standard library, the LCSS algorithm is used to perform action matching to obtain the LCSS matching score, including:

[0032] Set the matching tolerance; the matching tolerance is used to measure the similarity between the user action and the standard action.

[0033] According to the skateboard action sequence, action samples in the preset standard library, and matching tolerance, the LCSS dynamic programming algorithm is used to obtain the LCSS matching similarity; the LCSS matching similarity is:

[0034] ;

[0035] in, Indicates the first i Action and j The LCSS matching similarity of action samples, X_i, Y_j Respectively represent the first i The first and j The feature vector of an action, Indicates the matching tolerance.

[0036] Normalize the LCSS matching similarity and get the LCSS matching score:

[0037] ;

[0038] in, is the LCSS matching score, m 、 n They are the lengths of skateboard action sequences and action sequences in the preset standard library respectively.

[0039] In one embodiment, step S7 includes: connecting to a social module on a preset App terminal, and the user shares the skateboarding action recognition result in the form of a personalized action card on the platform. The system automatically generates a social score based on the action recognition score and social interaction; the social score is:

[0040] ;

[0041] in, Represents the social score.

[0042] The system optimization is judged based on the social score. If the convergence condition is not met, the system returns to step S4 to continue training and optimization.

[0043] If the convergence conditions are met, social function expansion and community mechanisms are used to achieve social function expansion; social function expansion and community mechanisms are used to introduce challenge task recommendations, skateboarding skill communities and skill labeling, ranking systems and personalized dynamic card generation modules based on skateboarding action recognition results, scoring history and interaction preferences, to build a social network among skateboarding enthusiasts with action as the medium, thereby improving user stickiness and platform activity.

[0044] In one embodiment, social function expansion and community mechanism are used to implement social function expansion, including:

[0045] Mark jump height and rotation angle using 3D trajectory heatmap.

[0046] The technical score is determined based on the LCSS matching score, the number of closed-loop structures, and the number of key points. The action level is divided according to the technical score. The technical score is:

[0047] ;

[0048] in, Indicates technical score, represents the LCSS matching score, represents the number of closed-loop structures, Indicates the number of key points.

[0049] The weighted sum of the technical score and the social score gives a comprehensive score:

[0050] ;

[0051] in, For the final rating, The weight of the technical score in the final score.

[0052] The key frames satisfying the curvature change rate greater than the preset curvature change rate value and the speed greater than the preset speed value are intercepted, and the stylized action posters are generated using the GAN network.

[0053] Active users within 1km are displayed through geo-fencing, and skaters with matching actions are recommended based on LCSS similarity and preset thresholds.

[0054] Link the LBS database to mark the venue-specific actions, and unlock the cultural badge by completing the closed-loop structure number β_k ≥ 3 actions.

[0055] Set up technical achievements, social achievements, virtual props and physical rights to form a complete social incentive system.

[0056] In one embodiment, the specific process of the challenge task recommendation module includes:

[0057] Get the action feature vector of the current user.

[0058] Calculate the Euclidean distance between the current user's action feature vector and all challenge feature vectors in the platform challenge library.

[0059] The Euclidean distances are sorted from small to large, and the Top-N challenges are selected for push. If a similar challenge record already exists for the current user's action feature vector, the imitation competition recommendation process is entered.

[0060] In one embodiment, the specific process of the personalized dynamic card generation module includes:

[0061] After the skateboard action is successfully recognized, a skateboard action card is automatically generated containing a trajectory diagram, action GIF demonstration, scoring dimension radar chart and challenge status. Users can share it to the App platform and social media with one click; the skateboard action card comes with a dynamic QR code, which supports others to scan the code to participate in the same action challenge.

[0062] In one embodiment, the ranking and points system specifically includes: determining comprehensive points based on the user's recognition accuracy, challenge completion, and interaction popularity, and establishing a points ranking list; comprehensive points are used to unlock customized action skins, skill tutorials, and virtual badges, incentivizing users to form a "skill-sharing-feedback" closed loop; comprehensive points are:

[0063] ;

[0064] in, represents the comprehensive score, M_final represents the final action similarity score; Challenge_Success_Rate Indicates the completion rate of the challenge task; Interaction_Index represents the comprehensive interaction score (weighted sum of likes, comments, and reposts); 、 、 They represent the weight coefficients of the final action similarity score, challenge task completion rate, and comprehensive interaction score respectively.

[0065] One of the above technical solutions has the following advantages and beneficial effects:

[0066] The above-mentioned skateboarding motion recognition and interaction method based on LCSS-topological feature fusion combines sensor trajectory analysis, motion curvature extraction, homology feature calculation and statistical modeling to construct a multi-dimensional recognition framework that integrates motion trajectory similarity, topological structure information and skateboarding motion complexity. The LCSS algorithm has the advantages of low computational complexity, low time alignment requirements, and strong noise resistance, making it suitable for the recognition of fast, irregular, and nonlinear motion sequences in skateboarding. At the same time, the introduction of topological features and motion statistical features enables the system to more accurately model the structure and personalize the characterization of motion patterns. In addition, this method introduces a social feedback mechanism based on motion recognition, realizing the quantitative integration of technical performance and user interactive behavior, providing skateboarding enthusiasts with an integrated platform for skill evaluation, social sharing and community interaction, and expanding the application boundaries of motion recognition technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 1 is a flow chart of a skateboarding action recognition and interaction method based on LCSS-topological feature fusion in one embodiment;

[0068] Figure 2 This is a specific flow chart of a skateboarding action recognition and interaction method based on LCSS-topological feature fusion in one embodiment. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0070] LCSS is an algorithm used to measure the similarity between two sequences, focusing on finding the longest subsequence length that can be matched while maintaining their relative order. Unlike DTW, LCSS is insensitive to local noise and the insertion or deletion of non-critical points in the sequence. It features low computational complexity, minimal time alignment requirements, and strong anti-interference capabilities, making it ideal for addressing the rapid switching, high-frequency jitter, and irregular trajectories encountered in skateboarding. Furthermore, by setting a similarity threshold, LCSS can effectively avoid misjudgments caused by minor perturbations while recognizing complex movements, demonstrating greater adaptability and stability in scenarios like skateboarding that require higher spatial and temporal tolerances.

[0071] In one embodiment, Figure 1 As shown in FIG, a skateboarding action recognition and interaction method based on LCSS-topological feature fusion is provided. The method is mainly applicable to the training, challenge and community interaction scenarios of skateboarding enthusiasts. The method includes the following steps:

[0072] Step S1: obtaining skateboard motion data and preprocessing the skateboard motion data, wherein the skateboard motion data includes: three-axis acceleration and angular velocity of the skateboard.

[0073] Specifically, the multimodal sensor module installed on the skateboard collects the three-axis acceleration of the skateboard in real time during its movement. , angular velocity Basic motion data, such as the accelerometer's measurement range is ±1°, the angle sensor's measurement range is ±5°, and the sampling frequency is set to 10Hz to ensure data continuity and accuracy, fully capturing basic skateboarding information. The data is then transmitted to a mobile device or edge computing module for preprocessing. The data collected in this step not only includes the skateboard's motion characteristics but can also be combined with the user's video recordings on the app for subsequent trajectory reconstruction and motion feature analysis.

[0074] The multimodal sensor module includes a three-axis acceleration sensor, a three-axis angle sensor and a GPS module.

[0075] Data preprocessing involves low-pass filtering and gravity compensation of the raw data. The filter coefficient is set to 0.2 to balance the weights of new and historical data and effectively eliminate high-frequency noise. Gravity compensation is also performed using a rotation matrix to eliminate the effects of gravitational acceleration.

[0076] Step S2: reconstructing the skateboard trajectory by using an integral operation with a damping term based on the preprocessed three-axis acceleration, and extracting multidimensional features from the reconstructed skateboard trajectory; the multidimensional features include: a complete position trajectory sequence, a direction sequence, motion statistical features, and direction change features.

[0077] Specifically, accurate 3D trajectory reconstruction is achieved by introducing a velocity integration and position update method with a damping coefficient. The damping coefficient is set to 0.95 to suppress error accumulation during the integration process.

[0078] Calculate motion statistical features, including total displacement, maximum velocity, average velocity, and direction change characteristics, to fully describe the kinematic characteristics of the action.

[0079] By employing an improved feature extraction algorithm and a multi-dimensional feature fusion strategy, the method not only provides a more comprehensive description of the motion characteristics through the combination of position trajectory sequences, topological features, and motion statistics, but also effectively suppresses the accumulation of integral errors by introducing a damping coefficient (0.95). In particular, the innovative introduction of topological feature analysis, by calculating the distribution of key points (curvature > 0.5) and continuous homology features, can capture the essential structural characteristics of skateboarding movements.

[0080] Step S3: extracting topological feature parameters from the reconstructed skateboard trajectory to obtain topological features.

[0081] Specifically, we extract topological features. Key points are detected by calculating curvature values, with a curvature threshold set to 0.5. We also calculate persistent homology features to capture the essential structural characteristics of the action. This method is effective in identifying key turning points in an action.

[0082] Step S4: Based on the skateboard action sequence and the action samples in the preset standard library, the LCSS algorithm is used to perform action matching to obtain an LCSS matching score.

[0083] Specifically, perform LCSS sequence matching. Set the matching threshold to 0.1 and calculate the similarity between the current action sequence and the preset action pattern. The LCSS algorithm can effectively handle problems such as inconsistent sequence lengths and noise interference.

[0084] By replacing the traditional DTW algorithm with the LCSS algorithm, the computational complexity is reduced from O(nm) to O(n), where n is the sequence length. The LCSS algorithm performs sequence matching by setting a matching threshold (preferably 0.1), which better handles noise and outliers in skateboarding, enabling real-time processing on mobile devices.

[0085] Step S5: Perform weighted fusion of topological features, motion statistical features and LCSS matching scores to obtain the final action similarity score.

[0086] Specifically, feature fusion is performed. The LCSS similarity (weight 50%), topological feature similarity (weight 30%), and motion statistical feature similarity (weight 20%) are weighted and fused to obtain the final action similarity score.

[0087] The topological features, motion statistics features and LCSS matching scores are integrated in a weighted fusion manner to construct a multi-dimensional feature vector. The weight coefficient is introduced w _1, w _2. w _3 Assign appropriate importance to different types of features, ultimately generating an interpretable action scoring model. This score can be used as an index of a user's skill spread on social platforms and can also be used to build challenge tasks and leaderboards to increase user engagement.

[0088] The final action similarity score is:

[0089] ;

[0090] in: For the final action similarity score, Score the LCSS similarity, Score topological features, S_stat Score statistical features.

[0091] A weighted combination of feature fusion is used to calculate the similarity score (LCSS similarity score weight 0.5, topological feature similarity score weight 0.3, motion statistical feature similarity score weight 0.2), which not only ensures the accuracy of recognition but also improves the system's robustness to noise.

[0092] Step S6: Obtain the skateboarding action recognition result based on the final action similarity score.

[0093] Specifically, based on the similarity score of the fused features, the current action type is identified, a recognition result is generated, and information such as the identified action type, similarity score, movement duration, and movement distance is output.

[0094] Step S7: Based on the skateboarding action recognition results and the social feedback mechanism, accurate recognition of complex skateboarding actions and personalized social experience are achieved.

[0095] Specifically, based on the recognition results, the following social functions are realized: marking the jump height through the 3D trajectory heat map ( ) and the rotation angle ( ) and other key parameters; Based on technical scoring ( ) is divided into action levels (SS / S / A / B / C), and combined with social scores ( ) Generate a comprehensive score; automatically intercept the curvature κ(t)>0.5 and the speed The key frames of the model are used to generate stylized action posters using the GAN network; active users within 1km are displayed through geo-fencing (GPS accuracy ±3m), based on LCSS similarity ( S_LCSS >0.8) Recommended actions for skaters; Linked to the LBS database to mark venue-specific actions (such as "rail slide"), unlocking requires completion Cultural badges for actions; set technical achievements (5 consecutive days S_tech >0.7) and social achievements (cross-city likes ≥100 times), rewards virtual props ("Topological Streamer" skin) and physical rights (80% off tickets to cooperative venues), forming a complete social incentive system.

[0096] The innovative combination of motion recognition technology and social applications has realized functions including personalized trajectory visualization, AI automatic scoring, personalized poster generation, nearby skater positioning, and local skateboarding culture display. The system adopts a sampling frequency of 10Hz and a filter coefficient of 0.2, which minimizes the data processing burden while ensuring the recognition effect. Experimental results show that the present invention not only achieves significant improvements in technical indicators such as computing efficiency, recognition accuracy, and feature stability, but more importantly, through the innovative integration of social functions, it provides skateboarding enthusiasts with a complete solution that integrates motion recognition, skill improvement, and social interaction. This deep integration of technology and applications has created a new paradigm for intelligent sports social networking.

[0097] The aforementioned skateboarding motion recognition and interaction method based on LCSS-topological feature fusion combines sensor trajectory analysis, motion curvature extraction, coherent feature calculation, and statistical modeling to construct a multidimensional recognition framework that integrates motion trajectory similarity, topological structure information, and skateboarding motion complexity. The LCSS algorithm has the advantages of low computational complexity, low requirements for time alignment, and strong noise resistance, making it suitable for identifying fast, irregular, and nonlinear motion sequences in skateboarding. At the same time, the introduction of topological features and motion statistical features enables the system to more accurately model and personalize motion patterns. Furthermore, this method introduces a feedback mechanism based on motion recognition, achieving a quantitative fusion of technical performance and user interactive behavior. This provides skateboarding enthusiasts with an integrated platform for skill assessment, social sharing, and community interaction, expanding the application boundaries of motion recognition technology and possessing excellent promotional value and commercial potential.

[0098] In one embodiment, the skateboard motion data is pre-processed in step S1, including: low-pass filtering and gravity compensation of the skateboard motion data to obtain processed acceleration and angular velocity; wherein the gravity compensation is used to separate the vertical gravity acceleration component from the original acceleration data through a rotation matrix reflecting the current posture; the gravity compensation formula is:

[0099] ;

[0100] in, is the three-axis acceleration of the skateboard after preprocessing, is the raw acceleration data after filtering, is the rotation matrix reflecting the current posture, θ is the three-dimensional attitude angle of the skateboard, g is the gravity vector, g =[0, 0, -9.81] T , is a matrix or vector multiplication.

[0101] Specifically, a low-pass filter algorithm is used to suppress noise in the raw data, and the filter coefficient α is set to 0.2. To eliminate the interference of gravity acceleration, the acceleration is transformed using the rotation matrix R(θ) of the skateboard posture, and uniformly projected into the skateboard body coordinate system, and the standard gravity vector is eliminated. g = [0, 0, -9.81] T This step ensures that the data reflects the actual motion trajectory of the skateboard rather than external interference, providing accurate input for action recognition.

[0102] Filtering algorithm:

[0103] ;

[0104] in, is the original acceleration data after filtering, is the original acceleration data, is the filter coefficient, , which controls the smoothness of the filter.

[0105] In one embodiment, in step S2, reconstructing the skateboard trajectory by using an integral operation with a damping term based on the preprocessed three-axis acceleration includes: integrating the preprocessed three-axis acceleration by using an integral operation with a damping term to obtain the skateboard velocity; the integral operation with the damping term is:

[0106] ;

[0107] in, for t Skateboard speed at the moment, for Skateboard speed at the moment, is the damping coefficient, is the three-axis acceleration of the skateboard after preprocessing; is the integration time interval.

[0108] The skateboard displacement is obtained by integrating the skateboard speed and the skateboard trajectory is reconstructed.

[0109] Specifically, the pre-processed three-axis acceleration data is used to calculate the speed of the skateboard through integration operation. and displacement A damping term of λ = 0.95 is introduced during the integration process to simulate friction and air resistance in a real environment. The reconstructed skateboard trajectory is used to capture key moments such as jumps, rotations, and landings. This trajectory not only serves as recognition input but also generates action visualization lines on the app, providing interactive feedback to users.

[0110] Velocity and position update equations:

[0111] ;

[0112] ;

[0113] in, For the time point t The skateboard speed at For the time point t The position of the skateboard at is matrix or vector multiplication, is the damping coefficient, simulating friction and air resistance, with a value ranging from 0 to 1, and a typical value of 0.95. Δt is the sampling interval in seconds (for example, 0.1 seconds).

[0114] In one embodiment, the topological features include: curvature change rate, number of key points, and number of closed-loop structures; the curvature change rate is:

[0115] ;

[0116] in, for t Skateboard speed at the moment, For skateboard tracks t The curvature value at the moment, is a matrix or vector multiplication.

[0117] The number of key points is the number of elements in the key point set; the key point set is:

[0118] ;

[0119] in, K is a set of key points, including the moments when the curvature is greater than the threshold, is the curvature threshold.

[0120] The number of closed-loop structures is:

[0121] ;

[0122] in, is the number of closed-loop structures (i.e. k dimensional Betti number), used to measure topological characteristics, for k -dimensional cyclic group, which represents the number of rings in the topology, for k dimensional boundary group, indicating the number of boundaries in the topology, is the k-th dimension homology group.

[0123] Specifically, topological feature parameters are extracted from the reconstructed trajectory, such as the curvature value of the skateboard trajectory at time t , number of key points N_dir , the number of closed-loop structures (Betti number ) to identify the complexity, completeness, and structural patterns of movements. This module identifies turning points and jumps in movements by analyzing points where the curvature exceeds a set threshold (e.g., 0.5).

[0124] For the k dimensional homology group. Its mathematical definition is: Algebraic topology meaning: describes the topological space k The equivalence class of 0-dimensional holes (such as 0-dimensional corresponding to connected components, 1-dimensional corresponding to ring structures, 2-dimensional corresponding to cavities, etc.)

[0125] When analyzing the trajectory of a skateboard,H _1 can capture the closed-loop structure in the trajectory (such as the number of circular sliding paths); is the Betti number in this dimension, which quantifies the persistence of topological features.

[0126] In one embodiment, step S4 includes: setting a matching tolerance; the matching tolerance is used to measure the similarity between the user action and the standard action; using the LCSS dynamic programming algorithm based on the skateboard action sequence, the action samples in the preset standard library, and the matching tolerance, to obtain the LCSS matching similarity; wherein the LCSS matching similarity is:

[0127] ;

[0128] in, Indicates the first i Action and j The LCSS matching similarity of action samples, X_i, Y_j Respectively represent the first i The first and j The feature vector of an action, Indicates the matching tolerance.

[0129] Normalize the LCSS matching similarity and get the LCSS matching score:

[0130] ;

[0131] in, is the LCSS matching score, m 、 n They are the lengths of skateboard action sequences and action sequences in the preset standard library respectively.

[0132] Specifically, the longest common subsequence (LCSS) matching of user action sequences and action samples in the preset standard library is performed, and the matching tolerance ε=0.1 is set to measure the similarity between user actions and standard actions. The matching score between actions is calculated through the dynamic programming algorithm, and the confidence level is extracted. ( in The recognition result is the similarity score of all candidate actions. This step implements the core computational process of user action recognition and forms the technical foundation for subsequent social interaction.

[0133] In one embodiment, step S7 includes: accessing a social module on a preset App end, and the user shares the skateboarding action recognition results to the platform in the form of personalized action cards, and the system automatically generates a social score based on the action recognition score and social interaction; performing system optimization judgment based on the social score, and if the convergence condition is not met, returning to step S4 to continue training and optimization; if the convergence condition is met, using social function expansion and community mechanism to achieve social function expansion; social function expansion and community mechanism are used to introduce challenge task recommendations, skateboarding skill communities and skill labeling, ranking systems and personalized dynamic card generation modules based on skateboarding action recognition results, scoring history and interaction preferences, to build a social network among skateboarding enthusiasts with action as the medium, thereby improving user stickiness and platform activity.

[0134] Specifically, by integrating the social module into the app, users can choose to share recognition results on the platform as personalized action cards. The system automatically generates a "social score" based on the action recognition score and social interactions (likes, comments, and reposts). This mechanism expands skateboarding action recognition from a personal skill assessment to an interactive experience with social communication value.

[0135] After each round of recognition and scoring, the system performs an iterative judgment. If the action similarity has stabilized (error convergence threshold ε_1 < 0.001) or the maximum number of iterations (e.g., 100) has been reached, the system outputs the final action recognition and social interaction results. Otherwise, it returns to step S4 to continue training and optimization.

[0136] In one embodiment, social function expansion and community mechanism are used to achieve social function expansion, including: marking jump height and rotation angle through a three-dimensional trajectory heat map; determining a technical score based on the LCSS matching score, the number of closed-loop structures, and the number of key points, and dividing the action level according to the technical score; the technical score is:

[0137] ;

[0138] in, Indicates technical score, represents the LCSS matching score, represents the number of closed-loop structures, Indicates the number of key points.

[0139] The technical score and social score are weighted and summed to obtain a comprehensive score; the social score is:

[0140] ;

[0141] in, Represents the social score.

[0142] The comprehensive score is:

[0143] ;

[0144] in, For the final rating, The weight of the technical score in the final score, usually set to 0.7.

[0145] Intercept keyframes that satisfy the curvature change rate greater than the preset curvature change rate value and the speed greater than the preset speed value, and use the GAN network to generate stylized action posters; display active users within 1km through geo-fencing, and recommend skaters with matching actions based on LCSS similarity and preset thresholds; link the LBS database to mark venue-specific actions, and unlock the number of closed-loop structures required to complete. Cultural badges of action; set technical achievements, social achievements, virtual props and physical rights to form a complete social incentive system.

[0146] In one embodiment, the specific process of the challenge task recommendation module includes: obtaining the action feature vector of the current user; calculating the Euclidean distance between the action feature vector of the current user and all challenge feature vectors in the platform challenge library; sorting the Euclidean distances from small to large, and selecting the Top-N challenges to push; if a similar challenge record already exists for the action feature vector of the current user, then entering the imitation competition recommendation process.

[0147] In one embodiment, the specific process of the personalized dynamic card generation module includes: after the skateboard action is successfully recognized, a skateboard action card containing a trajectory dynamic line diagram, an action GIF demonstration, a scoring dimension radar chart and a challenge status is automatically generated, and the user can share it to the App platform and social media with one click; the skateboard action card comes with a dynamic QR code, which supports others to scan the code to participate in the same action challenge.

[0148] In one embodiment, the ranking and points system specifically includes: determining comprehensive points based on the user's recognition accuracy, challenge completion, and interaction popularity, and establishing a points ranking list; comprehensive points are used to unlock customized action skins, skill tutorials, and virtual badges, incentivizing users to form a "skill-sharing-feedback" closed loop; comprehensive points are:

[0149] ;

[0150] in, represents the comprehensive score, M_final represents the final action similarity score; Challenge_Success_Rate Indicates the completion rate of the challenge task; Interaction_Index represents the comprehensive interaction score (weighted sum of likes, comments, and reposts); 、 、 They represent the weight coefficients of the final action similarity score, challenge task completion rate, and comprehensive interaction score, and their default values ​​are 0.5, 0.3, and 0.2, respectively.

[0151] Specifically, the expansion of social functions and the design of community mechanisms include:

[0152] To enhance the social interaction experience of the skateboarding action recognition system, this method further constructs a multi-level social function expansion mechanism. Based on the user's action recognition results, scoring history and interaction preferences, it introduces challenge task recommendation, skateboarding skill community, ranking system and personalized dynamic card generation module to build a social network among skateboarding enthusiasts with "action" as the medium, thereby improving user stickiness and platform activity.

[0153] (1) Personalized action recommendation and challenge mechanism

[0154] Based on the user's historical recognition records and scoring results, combined with their action type feature vector, the system automatically generates challenge tasks with appropriate difficulty and pushes them to the user to complete. Challenge tasks can be divided into basic skill practice, action difficulty upgrades, and imitation challenges from others. After completing the challenge, the system records the completion rate, action matching degree, and social reactions, forming a user skill growth curve and resume profile. The recommendation algorithm process is as follows: Get the current user's action feature vector ; All challenge feature vectors in the computing platform challenge library Euclidean distance; sort by distance from smallest to largest, select the Top-N challenge push; if If there is a similar challenge record, the "Imitation Competition" recommendation process will be entered.

[0155] (2) Action community and skill labeling

[0156] To build a skill-focused skateboarding community, the system clusters users based on move recognition tags and ratings. For example, community categories like "Basic Move Style," "Spin Masters," and "Advanced Jumpers" can be automatically assigned by the system or voluntarily joined by users. Each community supports interactive activities such as dynamic posting, move voting, and skill relay challenges, fostering a continuously active community ecosystem.

[0157] (3) Action card generation and cross-platform sharing

[0158] After successful recognition, the system can automatically generate a "skateboard action card" containing a trajectory diagram, action GIF demonstration, scoring dimension radar chart and challenge status. Users can share it to the App platform and social media (such as WeChat, Weibo, Xiaohongshu, etc.) with one click. The card comes with a dynamic QR code, which supports others to scan the code to participate in the same action challenge.

[0159] (4) Ranking and points system

[0160] The system generates comprehensive points based on the user's recognition accuracy, challenge completion, interaction popularity and other indicators ( Total_ Score ), and establishes three types of leaderboards: daily, weekly, and monthly. Points can be used to unlock custom action skins, skill tutorials, virtual badges, and more, encouraging users to form a closed loop of "skill-sharing-feedback."

[0161] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0162] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0163] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the scope of the present application, and such modifications and improvements are all within the scope of protection of the present application.

Claims

1. A skateboarding action recognition and interaction method based on LCSS-topological feature fusion, characterized in that: The method comprises: Step S1: Obtain skateboard motion data and pre-process the skateboard motion data, wherein the skateboard motion data includes: the three-axis acceleration and angular velocity of the skateboard, Step S2: reconstructing the skateboard trajectory using an integral operation with a damping term based on the pre-processed three-axis acceleration, and extracting multidimensional features from the reconstructed skateboard trajectory; the multidimensional features include: a complete position trajectory sequence, a direction sequence, motion statistical features, and direction change features; Step S3: extracting topological feature parameters from the reconstructed skateboard trajectory to obtain topological features; Step S4: Based on the skateboard action sequence and the action samples in the preset standard library, the LCSS algorithm is used to perform action matching to obtain an LCSS matching score; Step S5: weighted fusion of topological feature similarity, motion statistical feature similarity and LCSS matching score to obtain the final action similarity score; Step S6: Obtaining the skateboarding action recognition result based on the final action similarity score; Step S7: Based on the skateboarding action recognition results and the social feedback mechanism, accurate recognition of complex skateboarding actions and personalized social experience are achieved; Among them, the skateboard trajectory is reconstructed by using the integral operation with the introduction of the damping term based on the preprocessed three-axis acceleration, including: The pre-processed three-axis acceleration is integrated by the integration operation with the introduction of the damping term to obtain the skateboard speed; the integration operation with the introduction of the damping term is: ； in, for t Skateboard speed at the moment, for t- Skateboard speed at 1 moment, is the damping coefficient, is the three-axis acceleration of the skateboard after preprocessing; is the integration time interval; Integrate the skateboard speed to obtain the skateboard displacement and complete the skateboard trajectory reconstruction; Among them, topological features include: curvature change rate, number of key points, and number of closed-loop structures; The curvature change rate is: ； in: for t Skateboard speed at the moment, For skateboard tracks t The curvature value at the moment, is matrix or vector multiplication; The number of key points is the number of elements in the key point set, and the key point set is: ； in, K is a set of key points, including the moments when the curvature is greater than the threshold, is the curvature threshold; The number of closed-loop structures is: ； in, is the number of closed-loop structures, used to measure topological characteristics, for k -dimensional cyclic group, used to represent the number of rings in the topology, for k dimensional boundary group, which is used to represent the number of boundaries in the topology, For the k dimensional homology group.

2. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 1 is characterized in that: Preprocessing the skateboard motion data includes: The skateboard motion data is low-pass filtered and gravity compensated to obtain processed three-axis acceleration and angular velocity; wherein the gravity compensation is used to separate the vertical gravity acceleration component from the original acceleration data through a rotation matrix reflecting the current posture; the gravity compensation formula is: ； in, is the three-axis acceleration of the skateboard after preprocessing, is the raw acceleration data after filtering, is the rotation matrix reflecting the current posture, θ is the three-dimensional attitude angle of the skateboard, g is the gravity vector, g =[0, 0, -9.81] T , is a matrix or vector multiplication.

3. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 1 is characterized in that: Based on the skateboard action sequence and the action samples in the preset standard library, the LCSS algorithm is used to perform action matching and obtain the LCSS matching score, including: Setting a matching tolerance; the matching tolerance is used to measure the similarity between the user action and the standard action; According to the skateboard action sequence, the action samples in the preset standard library and the matching tolerance, the LCSS dynamic programming algorithm is used to obtain the LCSS matching similarity; wherein the LCSS matching similarity is: ； in, Indicates the first i Action and j The LCSS matching similarity of action samples, X_ i、Y_j Respectively represent the first i The first and j The feature vector of an action, Indicates matching tolerance; The LCSS matching similarity is normalized to obtain the LCSS matching score: ； in, is the LCSS matching score, m 、 n They are the lengths of skateboard action sequences and action sequences in the preset standard library respectively.

4. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 1 is characterized in that: Step S7 includes: By connecting to the social module on the preset App, users can share the skateboarding action recognition results on the platform in the form of personalized action cards. The system automatically generates a social score based on the action recognition score and social interaction. The social score is: ； in, represents the social score; Performing system optimization judgment based on the social score, if the convergence condition is not met, returning to step S4 to continue training optimization; If the convergence conditions are met, social function expansion and community mechanisms are used to achieve social function expansion; the social function expansion and community mechanisms are used to introduce challenge task recommendations, skateboarding skill communities and skill labeling, ranking systems and personalized dynamic card generation modules based on skateboarding action recognition results, scoring history and interaction preferences, to build a social network among skateboarding enthusiasts with action as the medium, thereby improving user stickiness and platform activity.

5. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 4 is characterized in that: Adopt social function expansion and community mechanism to achieve social function expansion, including: Mark jump height and rotation angle through 3D trajectory heat map; The technical score is determined based on the LCSS matching score, the number of closed-loop structures, and the number of key points, and the action level is divided according to the technical score; the technical score is: ； in, Indicates technical score, represents the LCSS matching score, represents the number of closed-loop structures, Indicates the number of key points; The technical score and the social score are weighted and summed to obtain a comprehensive score: ； in, For the final rating, The weight of technical scores in the final score; Intercept key frames that satisfy the curvature change rate greater than the preset curvature change rate value and the speed greater than the preset speed value, and use the GAN network to generate stylized action posters; Use geo-fencing to display active users within 1km and recommend skaters with matching actions based on LCSS similarity and preset thresholds; Link the LBS database to mark the venue-specific actions, and unlock the cultural badge that requires completing the closed-loop structure number β_k ≥ 3 actions; Set up technical achievements, social achievements, virtual props and physical rights to form a complete social incentive system.

6. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 4 is characterized in that: The specific process of the challenge task recommendation module includes: Get the action feature vector of the current user; Calculate the Euclidean distance between the current user's action feature vector and all challenge feature vectors in the platform challenge library; The Euclidean distances are sorted from small to large, and the Top-N challenges are selected for push; if a similar challenge record already exists for the current user's action feature vector, the imitation competition recommendation process is entered.

7. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 4 is characterized in that: The specific process of the personalized dynamic card generation module includes: After the skateboard action is successfully recognized, a skateboard action card containing a trajectory diagram, action GIF demonstration, scoring dimension radar chart and challenge status is automatically generated. Users can share it to the App platform and social media with one click; the skateboard action card comes with a dynamic QR code, which supports others to scan the code to participate in the same action challenge.

8. The skateboarding action recognition and interaction method based on LCSS-topological feature fusion according to claim 4 is characterized in that: The ranking and points system specifically includes: Based on the user's recognition accuracy, challenge completion, and interactive enthusiasm, comprehensive points are determined and a points ranking list is established; the comprehensive points are used to unlock customized action skins, skill tutorials, and virtual badges, encouraging users to form a "skill-sharing-feedback" closed loop; the comprehensive points are: ； in: represents the comprehensive score; M_final represents the final action similarity score; Challenge_ Success_Rate Indicates the completion rate of the challenge task; Interaction_Index represents the comprehensive interaction score; 、 、 They represent the weight coefficients of the final action similarity score, challenge task completion rate, and comprehensive interaction score respectively.

Images