Skeletal animation generation method and system

Abstract

The invention relates to a skeletal animation generation method and system, and the method comprises the steps: obtaining generation expectation information in response to an animation generation request of a user side; identifying key expected features of the generated expected information; inputting the skeleton demand features into a skeleton matching model to match skeleton data corresponding to the skeleton demand features; when the skeleton data is received, inputting the skeleton data and the generated animation characteristics into an animation generation model to generate an original skeleton animation; when the original skeletal animation is received, a user side previews and adjusts the original skeletal animation based on a preset plug-in; and outputting the adjusted original skeletal animation as a skeletal animation file based on the format requirement input by the user side. According to the method, by integrating the automation technology and the pre-training model, the time cost and technical requirements of skeletal animation production are reduced, the working efficiency of skeletal animation designers is improved, and the risk that the skeletal animation effect is poor due to human errors of the skeletal animation designers is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of computer graphics, and in particular to a method and system for generating skeletal animation. Background Art

[0002] Skeletal Animation, also known as Bone Animation, simulates the bone structure of organisms in the real world and uses the internal bone system to manipulate the movement of the character model instead of directly operating each vertex of the model.

[0003] Currently, skeletal animation technology has made significant progress and is widely used in game development, movie special effects, virtual reality and other fields. There are many professional software that support skeletal animation on the market, such as Blender, Maya, Unity, etc. These software provide rich bone binding and control functions, allowing animators to efficiently create and manage skeletal animation.

[0004] However, the existing skeletal animation generation methods still often rely on designers to manually operate skeletal animation professional software, resulting in the following defects:

[0005] 1. High time cost: Manual operation of skeletal animation software requires designers to invest a lot of time and energy. From the creation and binding of bones to the setting of key frames, each step requires delicate operation and adjustment, which makes the animation production cycle longer and time-consuming.

[0006] 2. High technical requirements: The generation of skeletal animation requires designers to have professional skills and knowledge: not only do they need to be familiar with the software's operating interface and tools, but they also need to understand the principles and movement laws of bone structure. At the same time, due to the complexity of its operation process, designers are prone to errors when manually operating skeletal animation software, and are prone to missing certain key steps or parameters during operation, resulting in animation effects that are not as expected. Summary of the invention

[0007] In order to solve the above problems, the present application provides a method and system for generating skeletal animation.

[0008] The above-mentioned invention objective of the present application is achieved through the following technical solutions:

[0009] A method for generating skeletal animation, comprising the steps of:

[0010] In response to an animation generation request from a user terminal, obtaining generation expectation information of the user terminal;

[0011] Identifying key desired features for generating desired information, wherein the key desired features include skeleton requirement features and generated animation features;

[0012] Inputting the bone demand feature into a pre-trained bone matching model, so that the bone matching model matches the bone data corresponding to the bone demand feature based on a preset bone database;

[0013] When receiving the bone data output by the bone matching model, inputting it and the generated animation features into the pre-trained animation generation model, so that the animation generation model generates the original bone animation;

[0014] When receiving the original skeletal animation output by the animation generation model, the user terminal is provided with a preset plug-in to preview and adjust the original skeletal animation;

[0015] The adjusted original skeletal animation is output as a skeletal animation file based on the format requirement input by the user end.

[0016] By adopting the above technical solution, the animation generation request of the user side is responded to, the user's generation expectation information is obtained, and the key expected features in this information are identified, including the skeleton requirement features and the generation animation features; the skeleton requirement features are input into the pre-trained skeleton matching model, and the model will quickly match the skeleton data that meets the requirements based on the preset skeleton database; when the skeleton data is received, it is input into the pre-trained animation generation model together with the generation animation features, and the model will automatically generate the original skeleton animation. In order to facilitate the user to view and adjust the animation effect, a plug-in for previewing and adjusting the original skeleton animation is provided for the user side to preview and adjust, and the adjusted original skeleton animation is output as a skeleton animation file that meets the format requirements of the user side; this application reduces the time cost and technical requirements of skeleton animation production by integrating automation technology and pre-trained models, which not only improves the work efficiency of skeleton animation designers, but also reduces the risk of poor skeleton animation effects due to human errors.

[0017] In a preferred example, the present application can be further configured as follows: the step of inputting the bone requirement feature into the pre-trained bone matching model so that the bone matching model matches the bone data corresponding to the bone requirement feature based on a preset bone database comprises the following steps:

[0018] The skeleton matching model searches the preset skeleton database based on the input skeleton requirement features and matches several candidate skeleton data;

[0019] The skeleton matching model evaluates the similarity of the matched skeleton data to obtain the preferred skeleton data;

[0020] The skeleton matching model optimizes and adjusts the preferred skeleton data based on the similarity evaluation results and preset optimization rules.

[0021] By adopting the above technical solution, after the skeleton requirement characteristics are input into the pre-trained skeleton matching model, the skeleton matching model will conduct a comprehensive search of the preset skeleton database based on these characteristics, and quickly match a number of candidate skeleton data with similar requirements. The model then performs a similarity evaluation on these candidate data, and through comparative analysis, determines the preferred skeleton data that best matches the requirement characteristics. In order to further improve the accuracy of the matching, the skeleton matching model will also perform refined optimization and fine-tuning on the preferred skeleton data based on the results of the similarity evaluation and the preset optimization rules, thereby improving the consistency between the final output skeleton data and the user's requirements. The present application realizes the precise matching of skeleton data through the steps of intelligent retrieval, similarity evaluation and optimization and fine-tuning of the skeleton matching model, which not only greatly shortens the time for designers to manually search and match skeleton data, but also significantly improves the accuracy and efficiency of matching. At the same time, the introduction of the optimization and fine-tuning mechanism enables the final output skeleton data to better meet the personalized needs of users and improves the customization level of animation production.

[0022] In a preferred example, the present application can be further configured as follows: the animation generation model includes a recognition layer, an integration layer, and a generation layer. When the bone data output by the bone matching model is received, the bone data and the generated animation features are input into the pre-trained animation generation model, so that the animation generation model generates the original bone animation, including the steps of:

[0023] The recognition layer recognizes the bone features corresponding to the bone data, wherein the bone features include bone structure features, bone proportion features, and bone movement features;

[0024] The integration layer integrates the skeleton features and the generated animation features based on the preset integration rules;

[0025] The generation layer generates original skeletal animation based on the comprehensive features obtained by feature integration.

[0026] By adopting the above technical solution, when the bone data output by the bone matching model is received, it is input into the pre-trained animation generation model together with the generated animation features. The animation generation model is composed of a recognition layer, an integration layer and a generation layer. Each layer works together to generate the original bone animation: first, the recognition layer conducts an in-depth analysis of the input bone data to accurately identify key information such as bone structure features, bone proportion features and bone movement features; then, the integration layer integrates these bone features with the generated animation features based on preset integration rules to form comprehensive features containing comprehensive information; the generation layer generates the original bone animation using the comprehensive features generated by the integration layer; this application improves the speed and efficiency of animation generation and enhances the visual effects of the animation and the user experience through the collaborative work of the recognition layer, integration layer and generation layer of the animation generation model.

[0027] In a preferred example, the present application can be further configured as follows: the recognition layer recognizes the bone features corresponding to the bone data, wherein the bone features include bone structure features, bone proportion features, and bone motion features, including the following steps:

[0028] The recognition layer models the topological structure of the skeleton data based on the graph neural network, and uses bones and joints as nodes and edges respectively to capture the skeleton proportion characteristics;

[0029] The recognition layer extracts the continuous motion trajectory and speed changes in the bone data based on the optical flow method, thereby extracting the bone motion features.

[0030] By adopting the above technical solution, the recognition layer uses graph neural networks to accurately model the topological structure of the skeleton data. In this process, bones are regarded as nodes, and joints are regarded as edges connecting nodes, which helps to more accurately capture and understand the proportional relationship between bones, that is, the skeleton proportional characteristics; in order to extract the motion characteristics of bones, the recognition layer uses the optical flow method to effectively extract continuous motion trajectories and speed changes from the skeleton data, so as to accurately capture the motion pattern and dynamic characteristics of the bones; the application of graph neural networks enables the model to accurately capture the proportional relationship between bones, providing accurate bone structure information for subsequent animation generation, and the introduction of the optical flow method enables the model to accurately extract the motion characteristics of bones, improving the natural fluency and realism of bone movement in animation.

[0031] In a preferred example, the present application can be further configured as follows: the generated animation features include motion pattern features and timing features, and the integration layer integrates the skeleton features and the generated animation features based on a preset integration rule, including the steps of:

[0032] The integration layer performs cross-modal feature mapping on the skeleton features and the generated animation features, thereby aligning the skeleton features with the motion pattern features and the timing features;

[0033] The integration layer combines the skeleton features and the generated animation features after cross-modal feature mapping to form comprehensive features based on preset integration rules. The preset integration rules include feature splicing and weighted fusion.

[0034] By adopting the above technical scheme, in order to effectively combine the skeleton features with the generated animation features, the integration layer aligns the skeleton features (including skeleton structure features, skeleton proportion features and skeleton motion features) with the generated animation features (including motion pattern features and timing features) through cross-modal feature mapping to ensure the semantic and spatial consistency of the two. After completing the feature mapping, the integration layer merges the features after cross-modal mapping according to the preset integration rules; the cross-modal feature mapping ensures that the skeleton features and the generated animation features are highly consistent in semantics and space, and provides accurate feature correspondence for the generation of animation, and the feature merging step effectively merges the two features into comprehensive features through scientific integration rules, thereby improving the accuracy and realism of animation generation, which not only optimizes the animation generation process, but also significantly improves the visual effects and user experience of the animation, bringing users a smoother and more natural skeleton animation effect.

[0035] In a preferred example, the present application can be further configured as follows: the step of generating the original skeletal animation based on the comprehensive features obtained by feature integration by the generation layer comprises the following steps:

[0036] The generation layer constructs several key frames of primitive skeletal animation based on comprehensive features;

[0037] The generation layer generates intermediate frames for the gaps between key frames based on the interpolation algorithm, thereby generating the original skeletal animation.

[0038] By adopting the above technical solution, the generation layer constructs several key frames of the original skeletal animation according to the comprehensive features obtained by feature integration. In order to generate a complete and smooth animation, the generation layer adopts an advanced interpolation algorithm to generate intermediate frames for the gaps between the key frames. By constructing key frames based on comprehensive features, the generation layer can improve the consistency of the overall form and movement of the animation with the user's expectations. The application of the interpolation algorithm makes the transition of the animation between key frames more natural and smooth, thereby improving the visual effect of the skeletal animation.

[0039] In a preferred example, the present application can be further configured as follows: the animation generation model further includes an evaluation layer and an optimization layer, and the generation layer generates intermediate frames for gaps between key frames based on an interpolation algorithm, thereby generating the original skeletal animation, and then performing the following steps:

[0040] The evaluation layer quantitatively scores the original skeletal animation based on a predefined reward function, and the items include physical realism and visual smoothness;

[0041] When the score of a sub-item in any item is lower than its corresponding preset threshold, the evaluation layer identifies abnormal frames or abnormal continuous frames based on the quantitative scoring process;

[0042] The optimization layer adjusts the scores of abnormal frames or abnormal continuous frames identified by the evaluation layer to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function.

[0043] By adopting the above technical solution, the evaluation layer conducts detailed quantitative scoring of the original skeletal animation according to the predefined reward function. The scoring items cover multiple key aspects such as physical realism and visual fluency. If the sub-item score in any item is lower than the preset threshold, the evaluation layer activates the abnormal recognition mechanism to accurately locate the abnormal frame or abnormal continuous frame. The optimization layer makes targeted adjustments to the abnormal frame or abnormal continuous frame identified by the evaluation layer according to the optimization mechanism corresponding to the reward function until its score is increased to above the preset threshold, thereby improving the overall quality and look and feel of the animation; through the scoring and abnormal recognition of the evaluation layer, the abnormal frame or abnormal continuous frame in the skeletal animation is discovered and located in time, and the optimization layer makes precise adjustments to the abnormal frame or abnormal continuous frame based on the scientific optimization mechanism, which significantly improves the physical realism and visual fluency of the animation.

[0044] In a preferred example, the present application can be further configured as follows: after the optimization layer adjusts the scores of the abnormal frames or abnormal continuous frames identified by the evaluation layer to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function, the following steps are performed:

[0045] The abnormal frames or abnormal continuous frames whose scores are adjusted to be above the corresponding preset threshold are input into the evaluation layer for quantitative scoring of each item;

[0046] If the same abnormal frame or the same abnormal continuous frame output by the evaluation layer is received again, the number of iterations of the abnormal frame or the abnormal continuous frame is obtained;

[0047] If the number of iterations does not reach the preset maximum number of iterations, the optimization layer adjusts the score of the abnormal frame or abnormal continuous frames to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function. If the number of iterations reaches the preset maximum number of iterations, the abnormal frame or abnormal continuous frames is sent to the user end.

[0048] By adopting the above technical solution, after the optimization layer adjusts the score of the abnormal frame or abnormal continuous frame identified by the evaluation layer to above the preset threshold, the adjusted abnormal frame or abnormal continuous frame will be input into the evaluation layer again for quantitative scoring of each item to verify the optimization effect; if the evaluation layer outputs the same abnormal frame or the same abnormal continuous frame again, the optimization layer obtains the number of iterations of the abnormal frame or the abnormal continuous frame, and determines whether the number of iterations reaches the preset maximum number of iterations; if not, the optimization layer will continue to make further adjustments to the abnormal frame or abnormal continuous frame based on the optimization mechanism corresponding to the reward function until its score stabilizes above the preset threshold; if the number of iterations has reached the preset maximum number of iterations, the optimization layer will continue to make further adjustments to the abnormal frame or abnormal continuous frame based on the optimization mechanism corresponding to the reward function until its score stabilizes above the preset threshold; if the number of iterations has reached the preset maximum number of iterations, the optimization layer will continue to make further adjustments to the abnormal frame or abnormal continuous frame based on the optimization mechanism corresponding to the reward function, until its score stabilizes above the preset threshold; if the number of iterations has reached the preset maximum number of iterations, the optimization layer will continue to make further adjustments to the abnormal frame or abnormal continuous frame based on the optimization mechanism corresponding to the reward function, until the ... If the problem is not completely solved after the maximum number of iterations, the optimization layer will send the abnormal frame or abnormal continuous frames to the user end for manual intervention or further processing by the user; through the above steps, the optimization layer can not only effectively adjust the abnormal frames or abnormal continuous frames identified by the evaluation layer, but also improve the stability and reliability of the adjustment effect through iterative verification; at the same time, this iterative optimization method not only improves the adaptability and robustness of the animation generation model, but also can minimize the need for manual intervention, improve the efficiency and automation of animation generation, and send the problems that cannot be solved completely automatically to the user end for manual processing in a timely manner.

[0049] The second object of the invention is achieved by the following technical solutions:

[0050] A skeletal animation generation system, comprising:

[0051] An information acquisition module, used for responding to an animation generation request from a user terminal and acquiring generation expectation information of the user terminal;

[0052] A feature recognition module, used to identify key desired features for generating desired information, wherein the key desired features include skeleton requirement features and generated animation features;

[0053] A first input module is used to input the bone demand feature into a pre-trained bone matching model, so that the bone matching model matches the bone data corresponding to the bone demand feature based on a preset bone database;

[0054] The second input module is used for, when receiving the skeleton data output by the skeleton matching model, inputting the skeleton data and the generated animation features into the pre-trained animation generation model, so that the animation generation model generates the original skeleton animation;

[0055] A plug-in scheduling module is used to allow the user end to preview and adjust the original skeletal animation based on a preset plug-in when receiving the original skeletal animation output by the animation generation model;

[0056] The output module is used to output the adjusted original skeletal animation into a skeletal animation file based on the format requirement input by the user end.

[0057] By adopting the above technical solution, the information acquisition module is used to respond to the animation generation request of the user terminal and obtain the generation expectation information of the user terminal; the feature recognition module is used to identify the key expected features of the generation expectation information, and the key expected features include the skeleton requirement features and the generation animation features; the first input module is used to input the skeleton requirement features into the pre-trained skeleton matching model, so that the skeleton matching model matches the skeleton data corresponding to the skeleton requirement features based on the preset skeleton database; the second input module is used to input the skeleton data output by the skeleton matching model and the generation animation features into the pre-trained animation generation model when receiving the skeleton data, so that the animation generation model generates the original skeleton animation; the plug-in scheduling module is used to allow the user terminal to preview and adjust the original skeleton animation based on the preset plug-in when receiving the original skeleton animation output by the animation generation model; the output module is used to output the adjusted original skeleton animation as a skeleton animation file based on the format requirement input by the user terminal.

[0058] In summary, the present application includes at least one of the following beneficial technical effects:

[0059] 1. This application reduces the time cost and technical requirements of skeletal animation production by integrating automation technology and pre-trained models, which not only improves the work efficiency of skeletal animation designers, but also reduces the risk of poor skeletal animation effects due to human errors;

[0060] 2. This application achieves accurate matching of skeletal data through intelligent retrieval, similarity evaluation, and optimization and fine-tuning of the skeletal matching model, which not only greatly shortens the time for designers to manually search and match skeletal data, but also significantly improves the accuracy and efficiency of matching; at the same time, the introduction of the optimization and fine-tuning mechanism enables the final output skeletal data to better meet the personalized needs of users and improve the customization level of animation production;

[0061] 3. The application of graph neural network enables the model to accurately capture the proportional relationship between bones, providing accurate bone structure information for subsequent animation generation. The introduction of optical flow method enables the model to accurately extract the movement characteristics of bones, improving the natural fluency and realism of bone movement in animation;

[0062] 4. Cross-modal feature mapping ensures that the skeletal features and the generated animation features are highly consistent in semantics and space, providing accurate feature correspondence for animation generation. The feature merging step effectively merges the two features into comprehensive features through scientific integration rules, improving the accuracy and realism of animation generation. It not only optimizes the animation generation process, but also significantly improves the visual effects and user experience of the animation, bringing users a smoother and more natural skeletal animation effect.

[0063] 5. This application constructs key frames based on comprehensive features, and the generation layer can improve the consistency of the overall form and action of the animation with the user's expectations. The application of the interpolation algorithm makes the transition between key frames more natural and smooth, improving the visual effect of the skeletal animation;

[0064] 6. This application uses the scoring and anomaly recognition of the evaluation layer to timely discover and locate abnormal frames or abnormal continuous frames in the skeletal animation. The optimization layer uses a scientific optimization mechanism to accurately adjust the abnormal frames or abnormal continuous frames, which significantly improves the physical realism and visual smoothness of the animation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of an embodiment of a method for generating skeletal animation of the present application;

[0066] Figure 2 This is a flowchart of an implementation of step S30 in an embodiment of a method for generating skeletal animation of the present application;

[0067] Figure 3 This is a flowchart of an implementation of step S40 in an embodiment of a method for generating skeletal animation of the present application;

[0068] Figure 4 This is an implementation flow chart after step S432 in an embodiment of a method for generating skeletal animation of the present application;

[0069] Figure 5 This is an implementation flow chart after step S435 in an embodiment of a skeletal animation generation method of the present application. DETAILED DESCRIPTION

[0070] The following is combined with Figure 1-5 This application is described in further detail.

[0071] In one embodiment, if Figure 1 As shown, the present application discloses a method for generating skeletal animation, which specifically includes the following steps:

[0072] S10: In response to the animation generation request of the user terminal, obtaining generation expectation information of the user terminal;

[0073] In this embodiment, the user terminal is the party that initiates the animation generation request; the animation generation request is a request issued by the user terminal for the purpose of generating an animation that meets its expectations, and the animation generation request includes the user's specific requirements and preferences for the animation; the generation expectation information is detailed information provided by the user terminal about the animation that it expects to generate;

[0074] Specifically, in response to an animation generation request for generating skeletal animation sent by a user end, generation expectation information of the skeletal animation generation expectation is obtained.

[0075] S20: identifying key expected features for generating expected information, wherein the key expected features include skeleton requirement features and generated animation features;

[0076] In this embodiment, the key expected features are features in the generation expected information that have important limitations on the generation of skeletal animation, including skeletal requirement features and generated animation features; skeletal requirement features are features that describe the skeletal structure or morphology of the character in skeletal animation; generated animation features are features that describe the characteristics of the animation itself, including the style, action type, duration, etc. of the animation;

[0077] Specifically, the key expected features in the expected generation information that have important limitations on the generation of skeletal animation are identified, including skeletal requirement features that describe the skeletal structure or form of the character in the skeletal animation and generated animation features that describe the characteristics of the animation itself.

[0078] S30: Inputting the bone requirement feature into the pre-trained bone matching model, so that the bone matching model matches the bone data corresponding to the bone requirement feature based on a preset bone database;

[0079] In this embodiment, the skeleton matching model is a pre-trained machine learning model, which is specifically used to match the corresponding skeleton data in a preset skeleton database according to the skeleton requirement characteristics; the skeleton database is a database containing various pre-created skeleton data for matching by the skeleton matching model;

[0080] Specifically, the bone requirement features are input into the pre-trained bone matching model, which will quickly match the bone data that meets the requirements based on the preset bone database.

[0081] S40: when receiving the skeleton data output by the skeleton matching model, inputting the skeleton data and the generated animation features into the pre-trained animation generation model, so that the animation generation model generates the original skeleton animation;

[0082] In this embodiment, the animation generation model is a pre-trained machine learning model, which is specifically used to generate original skeletal animation according to skeletal data and generated animation features; the original skeletal animation is a preliminary animation generated by the animation generation model;

[0083] Specifically, after receiving the skeleton data, it is input into a pre-trained animation generation model together with the generated animation features, and the model automatically generates the original skeleton animation.

[0084] S50: When receiving the original skeletal animation output by the animation generation model, the user terminal previews and adjusts the original skeletal animation based on a preset plug-in;

[0085] In this embodiment, the preset plug-in is a software tool used to provide additional functions or support during the animation generation process, specifically used to allow the user to preview and adjust the original skeletal animation;

[0086] Specifically, in order to facilitate users to view and adjust animation effects, a plug-in for previewing and adjusting original skeletal animation is provided for the user to preview and adjust.

[0087] S60: Outputting the adjusted original skeletal animation into a skeletal animation file based on the format requirement input by the user end;

[0088] In this embodiment, the skeletal animation file is a finally generated animation file that meets the user's format requirements;

[0089] Specifically, the adjusted original skeletal animation is output as a skeletal animation file that meets the format requirements of the user end.

[0090] In one embodiment, if Figure 2 As shown, step S30 includes the steps of:

[0091] S31: The skeleton matching model searches a preset skeleton database based on the input skeleton requirement features and matches a number of candidate skeleton data;

[0092] S32: the skeleton matching model performs similarity evaluation on the matched skeleton data to obtain preferred skeleton data;

[0093] S33: The skeleton matching model optimizes and adjusts the preferred skeleton data based on the similarity evaluation result and the preset optimization rules;

[0094] In this embodiment, the candidate skeleton data is a number of skeleton data similar to the requirements retrieved by the skeleton matching model in a preset skeleton database according to the input skeleton requirement characteristics; the similarity evaluation is a quantitative evaluation of the matching candidate skeleton data by the skeleton matching model, the purpose of which is to determine which skeleton data is most similar to the input skeleton requirement characteristics, and the similarity evaluation is based on multiple factors, including morphological similarity, motion matching, etc.; the preferred skeleton data is the skeleton data selected by the skeleton matching model as the most similar to the input requirement after the similarity evaluation; the preset optimization rule is a preset rule or algorithm, which is specifically used to guide the skeleton matching model to optimize and adjust the preferred skeleton data, including morphological adjustment, motion optimization, joint position fine-tuning, etc.;

[0095] Specifically, after the bone requirement characteristics are input into the pre-trained bone matching model, the bone matching model will conduct a comprehensive search of the preset bone database based on these characteristics, and quickly match several candidate bone data with similar demand characteristics. The model then performs a similarity evaluation on these candidate data, and through comparative analysis, determines the preferred bone data that best matches the demand characteristics. In order to further improve the accuracy of the match, the bone matching model will also perform refined optimization and fine-tuning on the preferred bone data based on the results of the similarity evaluation and the preset optimization rules, thereby improving the consistency of the final output bone data with user requirements.

[0096] In one embodiment, the animation generation model includes a recognition layer, an integration layer, and a generation layer, such as Figure 3 As shown, step S40 includes the steps of:

[0097] S41: The recognition layer identifies the bone features corresponding to the bone data, where the bone features include bone structure features, bone proportion features, and bone movement features;

[0098] S42: the integration layer integrates the skeleton features and the generated animation features based on a preset integration rule;

[0099] S43: The generation layer generates the original skeletal animation based on the comprehensive features obtained by feature integration;

[0100] In this embodiment, the recognition layer is the first part of the animation generation model, which is responsible for identifying the bone features corresponding to the bone data, including bone structure features, bone proportion features and bone motion features; the bone structure features are features that describe the overall structure and morphology of the bones, including the number, shape and connection mode of the bones; the bone proportion features are features that describe the proportional relationship between the various parts of the bones, including the length, width and joint position of the bones; the bone motion features are features that describe the characteristics of the bones during the movement process, including the rotation angle, movement trajectory and speed of the joints; the integration layer is the second part of the animation generation model, which is responsible for integrating the bone features extracted by the recognition layer and the generated animation features provided by the user end. The integration process is based on a preset integration rule, which effectively combines these two features together to provide comprehensive features for the generation layer; the comprehensive features are features obtained after processing by the integration layer, which contain all necessary information of the bone features and the generated animation features; the generation layer is the third part of the animation generation model, which is responsible for generating the original bone animation based on the comprehensive features;

[0101] Specifically, when the bone data output by the bone matching model is received, it is input into the pre-trained animation generation model together with the generated animation features. The animation generation model consists of a recognition layer, an integration layer and a generation layer. Each layer works together to generate the original bone animation: first, the recognition layer conducts an in-depth analysis of the input bone data to accurately identify key information such as bone structure features, bone proportion features, and bone movement features; then, the integration layer integrates these bone features with the generated animation features based on preset integration rules to form comprehensive features that contain comprehensive information; the generation layer uses the comprehensive features generated by the integration layer to generate the original bone animation.

[0102] In one embodiment, step S41 includes the steps of:

[0103] S411: The recognition layer models the topological structure of the skeleton data based on the graph neural network, and uses bones and joints as nodes and edges respectively to capture the skeleton proportion features;

[0104] S412: The recognition layer extracts the continuous motion trajectory and speed change in the skeleton data based on the optical flow method, thereby extracting the skeleton motion features;

[0105] In this embodiment, the graph neural network is a neural network for processing graph structure data. In the graph neural network, data is represented as a collection of nodes (vertices) and edges (edges), where nodes represent entities (such as bones) and edges represent relationships between entities (such as joint connections); the topological structure is a structure that describes the connection relationship between nodes and edges in the graph, specifically representing the connection mode between bones and joints; nodes are entities or objects in the graph neural network, specifically, bones are used as nodes; edges are relationships or connections in the graph neural network, specifically, joints are used as edges that connect different bone nodes; the optical flow method is a computer vision technology used to estimate the movement of objects in an image sequence, specifically for extracting the motion trajectory and speed change of bones in continuous time frames; the continuous motion trajectory describes the motion path of bones in continuous time frames. By extracting the continuous motion trajectory by the optical flow method, the motion pattern and trend of bones can be understood;

[0106] Specifically, the recognition layer uses graph neural networks to accurately model the topological structure of the skeleton data. In this process, bones are regarded as nodes, and joints are regarded as edges connecting nodes, which helps to more accurately capture and understand the proportional relationship between bones, that is, the skeleton proportional characteristics; in order to extract the motion characteristics of bones, the recognition layer uses the optical flow method to effectively extract continuous motion trajectories and speed changes from the skeleton data, thereby accurately capturing the motion patterns and dynamic characteristics of the bones; the application of graph neural networks enables the model to accurately capture the proportional relationship between bones, providing accurate skeleton structure information for subsequent animation generation, and the introduction of the optical flow method enables the model to accurately extract the motion characteristics of bones, improving the natural fluency and realism of skeleton movement in animation.

[0107] In one embodiment, generating animation features includes motion pattern features and timing features, and step S42 includes the steps of:

[0108] S421: the integration layer performs cross-modal feature mapping on the skeleton features and the generated animation features, thereby aligning the skeleton features with the motion pattern features and the timing features;

[0109] S422: The integration layer combines the skeleton features and the generated animation features after cross-modal feature mapping based on a preset integration rule to form a comprehensive feature, wherein the preset integration rule includes feature splicing and weighted fusion;

[0110] In this embodiment, the motion mode feature is a feature that describes the motion mode of the character in the animation, including the type of motion (such as walking, running, jumping, etc.), the rhythm of the motion, the amplitude of the motion, etc.; the timing feature is a feature that describes the passage of time and the order of events in the animation, including the duration of the animation, the time point of the key frame, the time interval between actions, etc.; cross-modal feature mapping is a technology for aligning the feature spaces of different modalities (such as skeletal features and generated animation features) so that they can be compared and combined in the same space; alignment is to transform the feature spaces of different modalities so that the similarities and differences between them can be measured on the same scale; feature splicing is a feature merging method, specifically connecting two or more feature vectors directly together to form a longer feature vector, specifically used to merge skeletal features and generated animation features into comprehensive features; weighted fusion is a feature merging method, specifically by assigning different weights to each feature vector and weighted summing them to form a new feature vector;

[0111] Specifically, in order to effectively combine the skeleton features with the generated animation features, the integration layer aligns the skeleton features (including skeleton structure features, skeleton proportion features and skeleton motion features) with the generated animation features (including motion pattern features and timing features) through cross-modal feature mapping to ensure the semantic and spatial consistency of the two. After completing the feature mapping, the integration layer merges the features after cross-modal mapping according to the preset integration rules; the cross-modal feature mapping ensures that the skeleton features and the generated animation features are highly consistent in semantics and space, and provides accurate feature correspondence for the generation of animation. The feature merging step effectively merges the two features into comprehensive features through scientific integration rules, thereby improving the accuracy and realism of animation generation, which not only optimizes the animation generation process, but also significantly improves the visual effects and user experience of the animation, bringing users a smoother and more natural skeleton animation effect.

[0112] In one embodiment, step S43 includes the steps of:

[0113] S431: The generation layer constructs several key frames of the original skeletal animation based on the comprehensive features;

[0114] S432: the generation layer generates intermediate frames for gaps between key frames based on an interpolation algorithm, thereby generating original skeletal animation;

[0115] In this embodiment, a key frame is a frame that defines an important moment or action change in an animation, and usually contains key information such as a character's posture, position, and expression; an interpolation algorithm is a mathematical method used to estimate the value of an unknown point between two or more known points, and is specifically used to generate an intermediate frame between key frames, thereby making the animation smoother and more coherent; an intermediate frame is a frame located between two key frames, and specifically, the generation of the intermediate frame is based on the interpolation algorithm and is inferred according to the information of the key frames;

[0116] Specifically, the generation layer constructs several key frames of the original skeletal animation according to the comprehensive features obtained by feature integration. In order to generate a complete and smooth animation, the generation layer adopts advanced interpolation algorithms to generate intermediate frames for the gaps between key frames. By constructing key frames based on comprehensive features, the generation layer can improve the consistency of the overall form and movement of the animation with the user's expectations. The application of the interpolation algorithm makes the transition of the animation between key frames more natural and smooth, thereby improving the visual effect of the skeletal animation.

[0117] In one embodiment, the animation generation model also includes an evaluation layer and an optimization layer, such as Figure 4 As shown, after step S432, the following steps are performed:

[0118] S433: the evaluation layer performs quantitative scoring of the original skeletal animation based on a predefined reward function, wherein the items include physical realism and visual fluency;

[0119] S434: When the score of a sub-item in any item is lower than its corresponding preset threshold, the evaluation layer identifies abnormal frames or abnormal continuous frames based on the quantitative scoring process;

[0120] S435: The optimization layer adjusts the scores of the abnormal frames or abnormal continuous frames identified by the evaluation layer to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function;

[0121] In this embodiment, the predefined reward function is a mathematical function used to score the animation according to specific standards or indicators, including multiple sub-items related to physical realism and visual fluency, each of which has a corresponding scoring standard and threshold; physical realism is a standard for describing whether the movement and interaction of objects in the animation conform to the laws of physics. As one of the important indicators for evaluating the quality of animation, it ensures that the animation looks natural and conforms to the audience's physical intuition; visual fluency is a standard for describing whether the transition between actions and scenes in the animation is smooth and coherent, which is crucial to providing a pleasant viewing experience, and ensures that the animation has no abrupt jumps or freezes; the quantitative scoring of the sub-items is a kind of evaluation The reward function is an algorithm or strategy for optimizing the animation according to the scoring result of the reward function, including adjusting animation parameters, recalculating key frames or intermediate frames, etc., to improve the scores of abnormal frames or abnormal continuous frames; the corresponding preset thresholds are thresholds corresponding to preset values ​​for each sub-item;

[0122] Specifically, the evaluation layer conducts detailed quantitative scoring of the original skeletal animation based on a predefined reward function. The scoring items cover multiple key aspects such as physical realism and visual fluency. If the score of any sub-item is lower than the preset threshold, the evaluation layer activates the abnormality recognition mechanism to accurately locate the abnormal frame or abnormal continuous frame. The optimization layer makes targeted adjustments to the abnormal frames or abnormal continuous frames identified by the evaluation layer according to the optimization mechanism corresponding to the reward function until their scores are raised above the preset threshold, thereby improving the overall quality and look and feel of the animation. Through the scoring and abnormality recognition of the evaluation layer, the abnormal frames or abnormal continuous frames in the skeletal animation are discovered and located in time. The optimization layer makes precise adjustments to the abnormal frames or abnormal continuous frames based on the scientific optimization mechanism, significantly improving the physical realism and visual fluency of the animation.

[0123] In one embodiment, if Figure 5 As shown, after step S435, the following steps are performed:

[0124] S436: inputting the abnormal frames or abnormal continuous frames whose scores are adjusted to be above the corresponding preset threshold into the evaluation layer for quantitative scoring of each item;

[0125] S437: If the same abnormal frame or the same abnormal continuous frame output by the evaluation layer is received again, the number of iterations of the abnormal frame or the abnormal continuous frame is obtained;

[0126] S438: If the number of iterations does not reach the preset maximum number of iterations, the optimization layer adjusts the score of the abnormal frame or abnormal continuous frames to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function, and if the number of iterations reaches the preset maximum number of iterations, the abnormal frame or abnormal continuous frames is sent to the user end;

[0127] In this embodiment, the number of iterations is the number of times an abnormal frame or abnormal continuous frames are adjusted during the optimization process, and each adjustment is regarded as an iteration; the preset maximum number of iterations is the maximum number of iterations set for the optimization process. When the number of iterations reaches this limit, the optimization process will stop and send the current result to the user end;

[0128] Specifically, after the optimization layer adjusts the score of the abnormal frame or abnormal continuous frame identified by the evaluation layer to above the preset threshold, the adjusted abnormal frame or abnormal continuous frame will be input into the evaluation layer again for quantitative scoring of each item to verify the optimization effect; if the evaluation layer outputs the same abnormal frame or the same abnormal continuous frame again, the optimization layer obtains the number of iterations of the abnormal frame or the abnormal continuous frame, and determines whether the number of iterations reaches the preset maximum number of iterations; if not, the optimization layer will continue to make further adjustments to the abnormal frame or abnormal continuous frame based on the optimization mechanism corresponding to the reward function until its score stabilizes above the preset threshold; if the number of iterations has reached the preset maximum number of iterations and the problem is still not completely solved, the optimization layer will send the abnormal frame or abnormal continuous frame to the user end for manual intervention or further processing by the user.

[0129] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0130] In one embodiment, a skeletal animation generation system is provided, and the skeletal animation generation system corresponds one-to-one to a skeletal animation generation method in the above embodiment. The skeletal animation generation system includes:

[0131] An information acquisition module, used for responding to an animation generation request from a user terminal and acquiring generation expectation information of the user terminal;

[0132] A feature recognition module, used to identify key desired features for generating desired information, wherein the key desired features include skeleton requirement features and generated animation features;

[0133] A first input module is used to input the bone demand feature into a pre-trained bone matching model, so that the bone matching model matches the bone data corresponding to the bone demand feature based on a preset bone database;

[0134] The second input module is used for, when receiving the skeleton data output by the skeleton matching model, inputting the skeleton data and the generated animation features into the pre-trained animation generation model, so that the animation generation model generates the original skeleton animation;

[0135] A plug-in scheduling module is used to allow the user end to preview and adjust the original skeletal animation based on a preset plug-in when receiving the original skeletal animation output by the animation generation model;

[0136] An output module, used for outputting the adjusted original skeletal animation into a skeletal animation file based on the format requirement input by the user end;

[0137] Optionally, also include:

[0138] A data matching module is used to search a preset bone database based on the input bone requirement characteristics and match a number of candidate bone data;

[0139] A similarity evaluation module is used to perform similarity evaluation on the matched skeleton data to obtain the preferred skeleton data;

[0140] An optimization and adjustment module, used to optimize and adjust the preferred skeleton data based on the similarity evaluation results and preset optimization rules;

[0141] Optionally, also include:

[0142] A recognition layer module, used to recognize the bone features corresponding to the bone data, wherein the bone features include bone structure features, bone proportion features and bone movement features;

[0143] An integration layer module is used to integrate the skeleton features and the generated animation features based on preset integration rules;

[0144] A generation layer module is used to generate original skeletal animation based on comprehensive features obtained by feature integration;

[0145] Optionally, the identification layer module includes:

[0146] The feature capture submodule is used to model the topological structure of the skeleton data based on the graph neural network, and use bones and joints as nodes and edges respectively to capture the skeleton proportion features;

[0147] The feature extraction submodule is used to extract the continuous motion trajectory and speed change in the skeleton data based on the optical flow method, so as to extract the skeleton motion features;

[0148] Optionally, the integration layer modules include:

[0149] A feature alignment submodule is used to perform cross-modal feature mapping on the skeleton features and the generated animation features, so as to align the skeleton features with the motion pattern features and the timing features;

[0150] A feature merging submodule, for merging the skeleton features and the generated animation features after cross-modal feature mapping to form comprehensive features based on preset integration rules, wherein the preset integration rules include feature splicing and weighted fusion;

[0151] Optionally, the generation layer modules include:

[0152] The keyframe construction submodule is used to construct several keyframes of primitive skeletal animations based on comprehensive features;

[0153] The intermediate frame generation submodule is used to generate intermediate frames for the gaps between key frames based on the interpolation algorithm, thereby generating the original skeletal animation;

[0154] Optionally, also include:

[0155] An evaluation layer module is used to quantitatively score the original skeletal animation by item based on a predefined reward function, wherein the items include physical realism and visual fluency; when the score of a sub-item in any item is lower than its corresponding preset threshold, the evaluation layer module is also used to identify abnormal frames or abnormal continuous frames based on the quantitative scoring process;

[0156] An optimization layer module, used to adjust the scores of abnormal frames or abnormal continuous frames identified by the evaluation layer to above a corresponding preset threshold based on an optimization mechanism corresponding to the reward function;

[0157] Optionally, also include:

[0158] An iterative input module is used to input abnormal frames or abnormal continuous frames whose scores are adjusted to be above a corresponding preset threshold into the evaluation layer for quantitative scoring of each item;

[0159] An iteration number acquisition module is used to acquire the iteration number of the abnormal frame or the abnormal continuous frame if the same abnormal frame or the same abnormal continuous frame output by the evaluation layer is received again;

[0160] The iteration number processing module is used to adjust the score of the abnormal frame or abnormal continuous frames to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function if the iteration number does not reach the preset maximum iteration number. If the iteration number reaches the preset maximum iteration number, the abnormal frame or abnormal continuous frames are sent to the user end.

[0161] The specific definition of a skeletal animation generation system can be found in the definition of a skeletal animation generation method mentioned above, which will not be repeated here. Each module in the above-mentioned skeletal animation generation system can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0162] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for generating skeletal animation, characterized in that: Includes steps: In response to an animation generation request from a user terminal, obtaining generation expectation information of the user terminal; Identifying key desired features for generating desired information, wherein the key desired features include skeleton requirement features and generated animation features; Inputting the bone demand feature into a pre-trained bone matching model, so that the bone matching model matches the bone data corresponding to the bone demand feature based on a preset bone database; When receiving the bone data output by the bone matching model, inputting it and the generated animation features into the pre-trained animation generation model, so that the animation generation model generates the original bone animation; When receiving the original skeletal animation output by the animation generation model, the user terminal is provided with a preset plug-in to preview and adjust the original skeletal animation; The adjusted original skeletal animation is output as a skeletal animation file based on the format requirement input by the user end.

2. The method for generating skeletal animation according to claim 1, characterized in that: The step of inputting the bone requirement feature into the pre-trained bone matching model so that the bone matching model matches the bone data corresponding to the bone requirement feature based on a preset bone database comprises the following steps: The skeleton matching model searches the preset skeleton database based on the input skeleton requirement features and matches several candidate skeleton data; The skeleton matching model evaluates the similarity of the matched skeleton data to obtain the preferred skeleton data; The skeleton matching model optimizes and adjusts the preferred skeleton data based on the similarity evaluation results and preset optimization rules.

3. The method for generating skeletal animation according to claim 1, characterized in that: The animation generation model comprises a recognition layer, an integration layer and a generation layer. When the skeleton data output by the skeleton matching model is received, the skeleton data and the generated animation features are input into the pre-trained animation generation model so that the animation generation model generates the original skeleton animation, comprising the steps of: The recognition layer recognizes the bone features corresponding to the bone data, wherein the bone features include bone structure features, bone proportion features, and bone movement features; The integration layer integrates the skeleton features and the generated animation features based on the preset integration rules; The generation layer generates primitive skeletal animation based on the comprehensive features obtained by feature integration.

4. The method for generating skeletal animation according to claim 3, characterized in that: The recognition layer recognizes the bone features corresponding to the bone data, wherein the bone features include bone structure features, bone proportion features, and bone movement features, including the steps of: The recognition layer models the topological structure of the skeleton data based on the graph neural network, and uses bones and joints as nodes and edges respectively to capture the skeleton proportion characteristics; The recognition layer extracts the continuous motion trajectory and speed changes in the bone data based on the optical flow method, thereby extracting the bone motion features.

5. The method for generating skeletal animation according to claim 3, characterized in that: The generated animation features include motion pattern features and timing features, and the integration layer integrates the skeleton features and the generated animation features based on a preset integration rule, including the steps of: The integration layer performs cross-modal feature mapping on the skeleton features and the generated animation features, thereby aligning the skeleton features with the motion pattern features and the timing features; The integration layer combines the skeleton features and the generated animation features after cross-modal feature mapping to form comprehensive features based on preset integration rules. The preset integration rules include feature splicing and weighted fusion.

6. The method for generating skeletal animation according to claim 3, characterized in that: The step of generating the original skeletal animation based on the comprehensive features obtained by feature integration at the generation layer comprises the steps of: the generation layer constructing a number of key frames of the original skeletal animation based on the comprehensive features; The generation layer generates intermediate frames for the gaps between key frames based on the interpolation algorithm, thereby generating the original skeletal animation.

7. A method for generating skeletal animation according to claim 6, characterized in that: The animation generation model further includes an evaluation layer and an optimization layer. After the generation layer generates intermediate frames for gaps between key frames based on an interpolation algorithm to generate the original skeletal animation, the following steps are performed: The evaluation layer quantitatively scores the original skeletal animation based on a predefined reward function, and the items include physical realism and visual smoothness; When the score of a sub-item in any item is lower than its corresponding preset threshold, the evaluation layer identifies abnormal frames or abnormal continuous frames based on the quantitative scoring process; The optimization layer adjusts the scores of abnormal frames or abnormal continuous frames identified by the evaluation layer to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function.

8. The method for generating skeletal animation according to claim 7, characterized in that: After the optimization layer adjusts the scores of the abnormal frames or abnormal continuous frames identified by the evaluation layer to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function, the following steps are performed: The abnormal frames or abnormal continuous frames whose scores are adjusted to be above the corresponding preset threshold are input into the evaluation layer for quantitative scoring of each item; If the same abnormal frame or the same abnormal continuous frame output by the evaluation layer is received again, the number of iterations of the abnormal frame or the abnormal continuous frame is obtained; If the number of iterations does not reach the preset maximum number of iterations, the optimization layer adjusts the score of the abnormal frame or abnormal continuous frames to above the corresponding preset threshold based on the optimization mechanism corresponding to the reward function. If the number of iterations reaches the preset maximum number of iterations, the abnormal frame or abnormal continuous frames is sent to the user end.

9. A skeletal animation generation system, characterized by: include: An information acquisition module, used for responding to an animation generation request from a user terminal and acquiring generation expectation information of the user terminal; A feature recognition module, used to identify key desired features for generating desired information, wherein the key desired features include skeleton requirement features and generated animation features; A first input module is used to input the bone demand feature into a pre-trained bone matching model, so that the bone matching model matches the bone data corresponding to the bone demand feature based on a preset bone database; The second input module is used for, when receiving the skeleton data output by the skeleton matching model, inputting the skeleton data and the generated animation features into the pre-trained animation generation model, so that the animation generation model generates the original skeleton animation; A plug-in scheduling module is used to allow the user end to preview and adjust the original skeletal animation based on a preset plug-in when receiving the original skeletal animation output by the animation generation model; The output module is used to output the adjusted original skeletal animation into a skeletal animation file based on the format requirement input by the user end.

Images