Intelligent error detection and repair system in three-dimensional animation production

Abstract

The invention relates to the technical field of three-dimensional animations, in particular to an intelligent error detection and repair system in three-dimensional animation production, which comprises a preprocessing module, an anomaly detection module, a repair processing module and a display terminal. In the action execution process, three-dimensional coordinate information of skeleton nodes is collected, time domain features and frequency domain features of the skeleton nodes are fused to construct a fusion feature vector of an execution action track, the fusion feature vector and a fusion feature vector of a reference execution action track are compared and analyzed, an action deviation comprehensive value is calculated, and abnormal skeleton nodes and associated vertexes thereof are recognized. And based on the voxel units where the abnormal vertexes are located and the structural neighborhoods of the abnormal vertexes, track similarity analysis is performed, a reference vertex set is determined, a repair track is generated, and repair effectiveness is judged, so that intelligent error recognition and natural repair in the three-dimensional animation are realized, and animation production efficiency and quality are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic identification of high-speed air ducts, and in particular to an intelligent error detection and repair system in three-dimensional animation production. Background Art

[0002] In the 3D animation production process, the character model's motion performance is not only related to visual realism and natural interaction, but also directly affects the efficiency and quality of animation rendering. However, currently, when complex character models execute motion, significant errors such as motion trajectory deviation, local vertex jitter, model interlacing, and limb deformation often occur due to skeletal binding errors, unreasonable vertex weight configuration, or kinematic anomalies in the animation drive mechanism. These problems not only undermine the coherence and realism of the character's motion, but also significantly increase the debugging cost of animation production.

[0003] Most existing error detection methods rely on manual frame-by-frame inspection or rule-based static threshold judgments, which are difficult to effectively cope with high frame rates, complex skeletal hierarchies, or nonlinear motion changes. They suffer from high rates of missed detection and misjudgment, poor positioning accuracy, and slow response time. At the same time, mainstream repair strategies generally focus on local geometric adjustments or resetting skeletal parameters, lacking dynamic modeling of vertex temporal trajectories, in-depth mining of frequency domain features, and fusion analysis of skeletal topology and spatial neighborhoods. As a result, the repair process often relies on manual intervention, making it difficult to accurately locate anomalies and adaptively repair them.

[0004] In order to solve the above-mentioned defects, a technical solution is now provided. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides an intelligent error detection and repair system in three-dimensional animation production, which can effectively solve the problems in the prior art.

[0006] To achieve the above objectives, the present invention can be implemented through the following technical solutions:

[0007] The present invention provides an intelligent error detection and repair system for three-dimensional animation production, comprising:

[0008] A pre-processing module for labeling vertices in the character model with space-bone dual indexes;

[0009] The anomaly detection module is used to drive the character model to perform actions in sequence according to the action instructions and their corresponding standard action templates, and collect the three-dimensional coordinate information of each skeleton node in each frame to construct an execution action atlas;

[0010] Based on the execution action graph, a fusion feature vector of the execution trajectory of each skeleton node under the current action instruction is constructed, wherein the fusion feature vector of the execution action trajectory is formed by fusing the representative features of the time domain and the frequency domain;

[0011] Compare and analyze the fused feature vector of the executed motion trajectory with the fused feature vector of the reference executed motion trajectory to obtain the comprehensive value of motion deviation, and use it to determine whether the skeleton node is abnormal;

[0012] Based on the space-bone dual index of the vertex, the vertex corresponding to the bone node judged to be abnormal is marked as an abnormal vertex;

[0013] The repair processing module is used to determine the voxel unit where the abnormal vertex is located, construct a structure-related neighborhood set, extract the three-dimensional trajectory of each vertex in the structure-related neighborhood set, and perform similarity analysis with the three-dimensional trajectory of the abnormal vertex to determine the reference vertex set. Based on this, the repair trajectory of the abnormal vertex is obtained, and then the new action deviation comprehensive value is re-determined to determine whether the repair is successful. If the repair is successful, the repair trajectory is input into the character model as the final result.

[0014] Furthermore, the vertices in the character model are marked with space-bone dual indexes. The specific marking process is as follows:

[0015] Construct a three-dimensional rectangular space area containing the entire character model with the center point of the character model as the origin, and record it as the bounding box;

[0016] Divide the bounding box into N equal parts along the x, y, and z axes. x 、N y 、N z Parts, totaling N x ×N y ×N z A small cube, and record it as a voxel unit;

[0017] Label each voxel unit with a triplet index, then extract the 3D coordinates of each vertex in the character model, determine the voxel unit it falls into, and use the triplet index corresponding to the voxel unit as the spatial index of the vertex;

[0018] Based on the bone structure of the character model, extract the center of mass coordinates of each bone node in 3D space, determine the Euclidean distance from each vertex in the character model to the center of mass coordinates of all bones, and use the name and distance value corresponding to the bone with the minimum Euclidean distance as the bone index of the vertex;

[0019] The spatial index and bone index of the vertex form a spatial-bone dual index.

[0020] Furthermore, the fusion feature vector of the execution action trajectory is constructed. The specific construction process is as follows:

[0021] Extract the three-dimensional coordinates (x(t a ), y(t a ), z(t a )), where t a represents the a-th frame moment, a=1, 2, 3…A, A represents the total number of frames of the action, and a three-dimensional sequence is constructed based on this, that is, the time domain trajectory of each skeleton node in the action;

[0022] Perform fast Fourier transform on the time domain trajectory of each skeleton node in the action to obtain spectrum information, which includes the f-th direction of the skeleton node in the x, y, and z directions. r The amplitude at the frequency;

[0023] Extract the amplitudes of the first r main frequencies from the spectrum information and calculate their average amplitude value as the representative feature of the frequency domain in, Represents the average amplitude value of the bone node in the x, y, and z directions respectively;

[0024] The average coordinates of each direction are extracted from the time domain trajectory as the representative features of the time domain (u x ,u y ,u y );

[0025] Fuse the representative features of time domain and frequency domain to construct the fusion feature vector of the execution trajectory

[0026] Furthermore, the process of solving the comprehensive value of action deviation is:

[0027] The difference between the execution action trajectory and the reference execution action trajectory is calculated in the time domain and frequency domain respectively, and the time domain deviation index value and the frequency domain deviation index value are obtained respectively;

[0028] The time domain deviation index value and the frequency domain deviation index value are weighted and calculated to obtain the motion deviation comprehensive value.

[0029] Furthermore, a structure-related neighborhood set is constructed. The specific construction process is as follows:

[0030] For the spatial index of the abnormal vertex in the preset bounding box, determine the voxel unit position where it is located, and establish a spatial neighborhood search range with the voxel unit as the center. Common neighborhood construction methods include: using six neighborhoods or second-order twenty-six neighborhood voxel units as reference areas to construct a candidate spatial neighborhood set of the abnormal vertex;

[0031] Based on the candidate spatial neighborhood set of the abnormal vertex, vertices that are associated with the same skeleton node as the current abnormal vertex or its adjacent skeleton node are screened out to form a structurally related neighborhood set.

[0032] Furthermore, the solution process for determining the reference vertex set is:

[0033] According to the three-dimensional trajectory of each vertex in the structurally relevant neighborhood set and the three-dimensional trajectory of the abnormal vertex, the difference value is determined to obtain the similarity score value;

[0034] Sort the vertices by similarity score from high to low, and select the first 1 / 3 of the vertices as the reference vertex set.

[0035] Furthermore, the process of solving the repair trajectory of abnormal vertices is as follows:

[0036] Based on the reference vertex set, the trajectory is weighted linear interpolated to obtain the repair trajectory of the abnormal vertex. The repair function formula is: Among them, h represents the number of the reference vertex, H represents the total number of reference vertex numbers, Gj represents the three-dimensional trajectory of the repaired abnormal vertex, Gs h represents the three-dimensional trajectory of the h-th reference vertex, η h Represents the set weighting coefficient, satisfying ∑η h =1.

[0037] Furthermore, the specific process of determining whether the repair is successful is as follows:

[0038] The repair trajectory of the abnormal vertex is re-embedded into the overall driving trajectory of the skeleton node to which it belongs, and the time domain deviation index value and frequency domain deviation index value of the skeleton node under the action instruction are recalculated. The new action deviation comprehensive value is determined based on this. If the new action deviation comprehensive value is less than the preset action deviation comprehensive threshold, it is judged that the repair is successful. Otherwise, it is judged that the repair fails.

[0039] Furthermore, a display terminal is included for distinguishing and marking vertices that have been repaired successfully and vertices that have failed to be repaired using color coding.

[0040] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0041] 1. The present invention constructs a space-skeleton dual indexing system based on three-dimensional space division and skeletal hierarchical relationships, enabling bidirectional mapping of vertices in structural space and motion control. After abnormal skeletal motion is identified, it can quickly and accurately trace back to the corresponding abnormal vertex set, thereby effectively improving the positioning efficiency and spatial resolution of abnormal tracing.

[0042] 2. This invention, by fusing time-domain and frequency-domain feature extraction methods, performs a fast Fourier transform on skeletal motion trajectories. Combining indicators such as the dominant frequency amplitude and trajectory mean, it constructs a multidimensional fusion feature vector to assess the degree of skeletal motion deviation. This vector can effectively identify various types of trajectory anomalies, including minor jitter, nonlinear drift, and excessive swing, thereby significantly improving the accuracy and robustness of anomaly detection and reducing both false alarm and missed detection rates.

[0043] 3. This invention proposes a structure-dependent neighborhood repair strategy that combines spatial neighborhood search with skeletal hierarchical constraints. After identifying abnormal vertices, it automatically selects a set of reference vertices with similar skeletal structures and trajectories, and performs weighted linear interpolation repair based on their three-dimensional trajectories. This strategy balances trajectory fitting accuracy with model structure rationality, avoiding local mutations and secondary distortion, thereby ensuring that the repaired motion trajectory is smooth and natural, with consistent structure, significantly improving animation continuity and visual realism.

[0044] 4. Based on the tension and compression of the bone nodes, combined with the motion deviation value and coupling degree, the motion deviation comprehensive threshold is updated and tested to better reflect the coordination between bones and the mechanical characteristics of complex movements, thereby optimizing threshold judgment and avoiding misjudgments and missed detections. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 It is the overall module block diagram of the present invention. DETAILED DESCRIPTION

[0047] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] like Figure 1 As shown, an intelligent error detection and repair system in 3D animation production includes: a pre-processing module, an anomaly detection module, a repair processing module and a display terminal;

[0049] The pre-processing module is used to perform space-bone dual index marking on vertices. The specific process is as follows: a three-dimensional rectangular space area containing the entire character model is constructed with the center point of the character model as the origin, and it is recorded as the bounding box;

[0050] Divide the bounding box into N equal parts along the x, y, and z axes. x 、N y 、N z Parts, totaling N x ×N y ×N z A small cube, and record it as a voxel unit;

[0051] Each voxel unit is labeled with a triplet index (i, j, k), where i∈[1, N x ], represents the row number, j∈[1, N y ], represents the column number, k∈[1, N z ], indicating the layer number;

[0052] Then extract the three-dimensional coordinates of each vertex in the character model, determine the voxel unit it falls into, and use the triplet index corresponding to the voxel unit as the spatial index of the vertex;

[0053] It should be noted that if the vertex happens to fall on the interface, dividing line, or dividing point of a voxel unit, the following attribution rules are used: for the x-axis ("row") direction, the side with the smaller index on both sides of the dividing line is taken; for the y-axis ("column") direction, the side with the higher index on both sides of the dividing line is taken; for the z-axis ("layer") direction, the side with the lower index on both sides of the dividing line is taken. Then, the obtained row, column, and layer index values ​​(i, j, k) are combined as the triplet index of the voxel unit to which the vertex belongs, thereby obtaining the spatial index of the vertex;

[0054] Based on the skeletal structure of the character model, the center of mass coordinates of each bone node in three-dimensional space (i.e., the position of the bone joint) are extracted, and a hierarchical number LID is assigned to each bone node. The top-level root bone has a hierarchical number LID set to 0. The hierarchical number LID of each child bone is increased by 1 compared to its parent bone, and the number increases layer by layer. This can clearly mark the parent-child relationship and hierarchical structure between bones;

[0055] Determine the Euclidean distance from each vertex in the character model to the centroid coordinates of all bone nodes, and use the name and distance value corresponding to the bone with the minimum Euclidean distance as the bone index of the vertex;

[0056] It should be pointed out that if multiple candidate bones need to be retained, they can be sorted from small to large according to the distance, and the first Q bone nodes with the smallest distance and their corresponding distances are selected to form a candidate list as the bone extension index of the vertex;

[0057] The spatial index and bone index of the vertex form a spatial-bone dual index;

[0058] In a specific embodiment, the present invention converts the originally disordered vertex data into ordered data with clear spatial hierarchy and skeletal hierarchy through dual labeling of spatial index and skeletal index, which is beneficial to making subsequent key links such as abnormality detection and repair faster, more accurately positioned, and more efficient in processing.

[0059] The anomaly detection module is used to predefine a set of action instructions and their corresponding standard action templates, such as arm raising, forward stepping, waving, turning, etc., and input the action instruction sequence in sequence as needed;

[0060] The action execution engine drives the character model to perform actions. The action execution engine is driven according to the parameter information attached to each action instruction (such as the target bone node, rotation angle, duration and execution frequency). During the process, the three-dimensional coordinate information of each bone node in each frame is collected to obtain a collection of execution action atlases sorted in chronological order.

[0061] Based on the execution action atlas, the fusion feature vector of the action trajectory of each skeleton node under the current action instruction is constructed. The specific process is as follows:

[0062] Extract the three-dimensional coordinates (x(t a ), y(t a ), z(t a )), where t a represents the a-th frame moment, a=1, 2, 3…A, A represents the total number of frames of the action, and a three-dimensional sequence is constructed based on this, that is, the time domain trajectory of each skeleton node in the action;

[0063] Perform fast Fourier transform (i.e. convert into frequency domain representation) on the time domain trajectory of each skeleton node in the action to obtain spectrum information, which includes the f-th direction of the skeleton node in the x, y, and z directions. r The amplitude at the frequency;

[0064] Among them, the bone node is the fth in the x direction r The amplitude at the frequency |x(f r ) | The calculation formula is:

[0065]

[0066] Among them, x(f r ) indicates that the bone node is the fth in the x direction r The energy value at a frequency, represents the rth frequency component (from low frequency to high frequency), Δt represents the time difference, Represents the complex exponential (i.e., the basis of the sine and cosine functions), Re represents the real part, and Im represents the imaginary part;

[0067] Calculate the fth direction of the bone node in the y and z directions r The amplitude at the frequency can be obtained by the same logic;

[0068] Extract the amplitudes of the first r (e.g. r=10) main frequencies from the spectrum information and calculate their average amplitude value as the representative feature of the frequency domain in, Represents the average amplitude value of the bone node in the x, y, and z directions respectively;

[0069] The average coordinates of each direction are extracted from the time domain trajectory as the representative features of the time domain (u x ,u y ,u y );

[0070] Fuse the representative features of time domain and frequency domain to construct the fusion feature vector of the execution trajectory

[0071] Extract the reference execution action graph set from the database and perform the same Fourier trajectory processing on it to construct the fusion feature vector of the reference execution action trajectory

[0072] The difference between the execution action trajectory and the reference execution action trajectory is calculated in the time domain and frequency domain respectively, and the time domain deviation index value Ds and the frequency domain deviation index value Dp are obtained respectively. The calculation formula is:

[0073] The time domain deviation index value Ds and the frequency domain deviation index value are weighted to obtain the motion deviation comprehensive value, which is calculated as follows: D sp =β1×D s +β2×D p , where β1 and β2 represent the preset weight coefficients respectively;

[0074] Compare and analyze the motion deviation comprehensive value of each skeletal node under the motion instruction with the preset motion deviation comprehensive threshold. If the motion deviation comprehensive value of a skeletal node under the motion instruction is greater than or equal to the preset motion deviation comprehensive threshold, the skeletal node under the motion instruction is judged to be abnormal. Conversely, if the motion deviation comprehensive value of a skeletal node under the motion instruction is less than the preset motion deviation comprehensive threshold, the skeletal node under the motion instruction is judged to be normal.

[0075] Based on the space-bone dual index of the vertex, the vertex corresponding to the bone node judged to be abnormal is marked as an abnormal vertex;

[0076] In a specific embodiment, the present invention defines a set of action instructions and their standard action templates, and uses an action execution engine to drive the character model to collect three-dimensional coordinate data of bone nodes on a frame-by-frame basis, constructs an execution action atlas, and then extracts representative features of each bone node in the time domain and frequency domain, and generates a fusion feature vector, which can accurately capture the action changes of the bone nodes in the time domain and frequency domain to avoid misjudgment or missed detection, and then compares the deviation with the fusion feature vector of the reference action template, calculates the comprehensive value of the action deviation, and compares it with a preset threshold to realize automatic identification of abnormal bone nodes, and further marks the vertices associated with the abnormal bone nodes as abnormal vertices based on the spatial position of the vertex and the bone hierarchy relationship, thereby achieving high precision and automation capabilities in action recognition and anomaly detection, providing a reliable foundation for subsequent repair processing, and effectively improving the efficiency and quality control level of animation production;

[0077] The repair processing module is used to determine the spatial index of the abnormal vertex in the preset bounding box, determine the voxel unit position of the vertex, and establish a spatial neighborhood search range with the voxel unit as the center. Common neighborhood construction methods include: using six neighborhoods or second-order twenty-six neighborhood voxel units as reference areas to construct a candidate spatial neighborhood set of the abnormal vertex;

[0078] It should be pointed out that the six-neighborhood voxel unit refers to the six directly adjacent units of up and down, left and right, and front and back, and the second-order twenty-six-neighborhood voxel unit refers to the second-order cubic adjacent area extended to all directions around it, with a total of 26 adjacent units;

[0079] Based on the candidate spatial neighborhood set of the abnormal vertex, the vertices associated with the same bone node as the current abnormal vertex or its adjacent bone nodes (such as parent bones, child bones) are screened out to form a structurally related neighborhood set. This helps to make trajectory references based on similar spatial positions and similar bone structures, thereby improving the accuracy of the repair results;

[0080] For each vertex in the structure-related neighborhood set, its three-dimensional trajectory under the current action instruction is extracted, and similarity analysis is performed with the three-dimensional trajectory of the abnormal vertex to obtain the similarity score value of each vertex. The specific analysis is as follows:

[0081] Extract the three-dimensional coordinates (x and y) of each vertex in the structure-related neighborhood set corresponding to the skeleton node at each frame moment in the entire action process under the current action instruction. θ (t a ), y θ (t a ), z θ (t a )), based on which the three-dimensional trajectory of each vertex in the structure-related neighborhood set is constructed;

[0082] Synchronously extract the three-dimensional coordinates (x φ (t a ), y φ (t a ), z φ (t a )), and thus construct the three-dimensional trajectory of the abnormal vertex;

[0083] According to the formula:

[0084] Get the similarity score Hsd of each vertex, where Csd represents the difference value;

[0085] Sort the vertices by similarity score from high to low, and select the first 1 / 3 of the vertices as the reference vertex set for repair interpolation;

[0086] Based on the reference vertex set, the trajectory is weighted linear interpolated to obtain the repair trajectory of the abnormal vertex. The repair function formula is: Among them, h represents the number of the reference vertex, H represents the total number of reference vertex numbers, Gj represents the three-dimensional trajectory of the repaired abnormal vertex, Gs h represents the three-dimensional trajectory of the h-th reference vertex, η h Represents the set weighting coefficient, satisfying Ση h =1;

[0087] The repair trajectory of the abnormal vertex is re-embedded into the overall driving trajectory of the skeleton node to which it belongs, and the time domain deviation index value and frequency domain deviation index value of the skeleton node under the action instruction are recalculated to determine the new action deviation comprehensive value. If the new action deviation comprehensive value is less than the preset action deviation comprehensive threshold, it is judged that the repair is successful, otherwise it is judged as a repair failure;

[0088] For successfully repaired vertices, their repaired trajectories will be input into the character model as the final result, and the database will be updated synchronously. This ensures that similar anomalies will not recur during subsequent animation rendering or action execution.

[0089] In a specific embodiment, the present invention proposes a structure-related neighborhood repair strategy that combines spatial neighborhood search with bone hierarchical constraints. After determining the abnormal vertex, it automatically selects a set of reference vertices with similar bone structures and trajectories, and performs weighted linear interpolation repair based on their three-dimensional trajectories. This strategy takes into account both trajectory fitting accuracy and model structure rationality, avoids local mutations and secondary distortion, and thus ensures that the repaired motion trajectory is smooth and natural, with consistent structure, greatly improving animation continuity and visual realism.

[0090] The display terminal is used to visually mark the vertices that have been repaired successfully and those that have failed. The specific process is as follows:

[0091] Color coding is used to distinguish and mark vertices that have been repaired successfully and those that have failed to be repaired. For example, vertices that have been repaired successfully are marked in green, and vertices that have failed to be repaired are marked in red. Through this visualization method, animators can intuitively identify abnormal vertices and their repair effects, and can further perform manual weight adjustment or motion compensation optimization based on the visualization feedback results, thereby greatly improving the interaction efficiency and quality control capabilities during the 3D animation production process.

[0092] Finally, the present invention also provides an optimization method for the motion deviation comprehensive threshold. Considering that in many actions, bones not only move independently, but also work closely together. For example, when bending, the movements of the waist and legs need to be precisely coordinated. If the coordination is not good, obvious errors will occur. The complexity of this coordinated action is very critical for evaluating the motion deviation comprehensive threshold. Secondly, therefore, in complex actions, the stress conditions of skeletal joints are also crucial to the complexity of the action. The stress conditions of the joints include tension and compression, which usually affect the smoothness and naturalness of the action. Therefore, the coupling analysis between bones and the direction of joint tension and compression are introduced to optimize the preset motion deviation comprehensive threshold:

[0093] During an action, the rotation changes of two bones may affect each other. The core idea of ​​this process is to evaluate whether two bones have coordinated motion in the same action by measuring the covariance of the rotation angles between the bones, as follows:

[0094] Calculate the bone rotation change, starting from the rotation angle change of each bone, and define the bone node r d and r l The bone rotation changes can be calculated by the position changes between each frame, then:

[0095]

[0096] Where Δθ d Represents the skeleton node r d The bone rotation change, q d (t) represents the skeleton node r d The quaternion representation at time t, arccos(·) represents the inverse cosine function, and the skeleton node r is obtained in the same way. l The bone rotation change Δθ l ;

[0097] For the bone node r d and r l Calculate their covariance Cov(Δθ d ,Δθl ), reflecting the degree of their coordinated movement in the action;

[0098] Using covariance to determine the skeleton node r d and r l The coupling degree measures the coordination of two skeleton nodes in the action. d ,r l ) is calculated as:

[0099]

[0100] Where σΔθ d and σΔθ l They represent the standard deviation of the bone rotation changes of the corresponding bone nodes;

[0101] If Coupling(r d ,r l )≈1, it means that the two skeleton nodes are almost completely coordinated;

[0102] Coupling d ,r l )≈0, it means that the bone rotation changes of the two bone nodes are irrelevant;

[0103] In complex movements, the stress on the skeletal joints is also crucial to the complexity of the movement. The stress on the joints includes tension and compression, which usually affects the smoothness and naturalness of the movement. Therefore:

[0104] Calculate the moment M of the skeletal joint d (t) = r d (t)×F d (t)

[0105] Among them, F d (t) represents the force applied to the skeletal node;

[0106] Tension: If the direction of force is stretching, the tension d (t) is equal to the skeleton node r d The absolute value of the moment. If the direction of the force is compression, the tension is 0;

[0107] Compression: When the force is in compression, the compression force d (t) is equal to the skeleton node r d The absolute value of the moment. If the direction of the force is tensile, the compressive force is 0;

[0108] Therefore, the sum of the tension and compression forces of all skeletal nodes is calculated, combined with the motion deviation value and coupling degree, and different weighted coefficients are assigned to the three to obtain an updated motion deviation comprehensive threshold. The above-mentioned detection is then performed to better reflect the coordination between bones and the mechanical properties in complex movements, thereby optimizing threshold judgment and avoiding misjudgment and missed detection.

[0109] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent error detection and repair system for 3D animation production, characterized in that: include: A pre-processing module for labeling vertices in the character model with space-bone dual indexes; The anomaly detection module is used to drive the character model to perform actions in sequence according to the action instructions and their corresponding standard action templates, and collect the three-dimensional coordinate information of each skeleton node in each frame to construct an execution action atlas; Based on the execution action graph, a fusion feature vector of the execution trajectory of each skeleton node under the current action instruction is constructed, wherein the fusion feature vector of the execution action trajectory is formed by fusing the representative features of the time domain and the frequency domain; Compare and analyze the fused feature vector of the executed motion trajectory with the fused feature vector of the reference executed motion trajectory to obtain the comprehensive value of motion deviation, and use it to determine whether the skeleton node is abnormal; Based on the space-bone dual index of the vertex, the vertex corresponding to the bone node judged to be abnormal is marked as an abnormal vertex; The repair processing module is used to determine the voxel unit where the abnormal vertex is located, construct a structure-related neighborhood set, extract the three-dimensional trajectory of each vertex in the structure-related neighborhood set, and perform similarity analysis with the three-dimensional trajectory of the abnormal vertex to determine the reference vertex set. Based on this, the repair trajectory of the abnormal vertex is obtained, and then the new action deviation comprehensive value is re-determined to determine whether the repair is successful. If the repair is successful, the repair trajectory is input into the character model as the final result.

2. The intelligent error detection and repair system for 3D animation production according to claim 1, characterized in that: Perform space-bone dual index marking on the vertices in the character model. The specific marking process is as follows: Construct a three-dimensional rectangular space area containing the entire character model with the center point of the character model as the origin, and record it as the bounding box; Divide the bounding box into N equal parts along the x, y, and z axes. x 、N y 、N z Parts, totaling N x ×N y ×N z A small cube, and record it as a voxel unit; Label each voxel unit with a triplet index, then extract the 3D coordinates of each vertex in the character model, determine the voxel unit it falls into, and use the triplet index corresponding to the voxel unit as the spatial index of the vertex; Based on the bone structure of the character model, extract the center of mass coordinates of each bone node in 3D space, determine the Euclidean distance from each vertex in the character model to the center of mass coordinates of all bones, and use the name and distance value corresponding to the bone with the minimum Euclidean distance as the bone index of the vertex; The spatial index and bone index of the vertex form a spatial-bone dual index.

3. The intelligent error detection and repair system for 3D animation production according to claim 1, characterized in that: Construct the fusion feature vector of the execution action trajectory. The specific construction process is as follows: Extract the three-dimensional coordinates (x(t a ), y(t a ), z(t a )), where t a represents the a-th frame moment, a=1, 2, 3…A, A represents the total number of frames of the action, and a three-dimensional sequence is constructed based on this, that is, the time domain trajectory of each skeleton node in the action; Perform fast Fourier transform on the time domain trajectory of each skeleton node in the action to obtain spectrum information, which includes the f-th direction of the skeleton node in the x, y, and z directions. r The amplitude at the frequency; Extract the amplitudes of the first r main frequencies from the spectrum information and calculate their average amplitude value as the representative feature of the frequency domain in, Represents the average amplitude value of the bone node in the x, y, and z directions respectively; The average coordinates of each direction are extracted from the time domain trajectory as the representative features of the time domain (u x ,u y ,u y ); Fuse the representative features of time domain and frequency domain to construct the fusion feature vector of the execution trajectory 4. The intelligent error detection and repair system for 3D animation production according to claim 1, characterized in that: The process of solving the comprehensive value of action deviation is: The difference between the execution action trajectory and the reference execution action trajectory is calculated in the time domain and frequency domain respectively, and the time domain deviation index value and the frequency domain deviation index value are obtained respectively; The time domain deviation index value and the frequency domain deviation index value are weighted and calculated to obtain the motion deviation comprehensive value.

5. The intelligent error detection and repair system in 3D animation production according to claim 1, characterized in that: Construct a structure-related neighborhood set. The specific construction process is as follows: For the spatial index of the vertex that has been determined to be abnormal in the preset bounding box, the voxel unit position is determined, and the spatial neighborhood search range is established with the voxel unit as the center. The neighborhood construction method includes: using six neighborhoods or second-order twenty-six neighborhood voxel units as reference areas to construct a candidate spatial neighborhood set of the abnormal vertex; Based on the candidate spatial neighborhood set of the abnormal vertex, vertices that are associated with the same skeleton node as the current abnormal vertex or its adjacent skeleton node are screened out to form a structurally related neighborhood set.

6. The intelligent error detection and repair system in 3D animation production according to claim 1, characterized in that: The solution process for determining the reference vertex set is: According to the three-dimensional trajectory of each vertex in the structurally relevant neighborhood set and the three-dimensional trajectory of the abnormal vertex, the difference value is determined to obtain the similarity score value; Sort the vertices by similarity score from high to low, and select the first 1 / 3 of the vertices as the reference vertex set.

7. The intelligent error detection and repair system for 3D animation production according to claim 1, characterized in that: The solution process for the repair trajectory of abnormal vertices is: Based on the reference vertex set, the trajectory is weighted linear interpolated to obtain the repair trajectory of the abnormal vertex. The repair function formula is: Among them, h represents the number of the reference vertex, H represents the total number of reference vertex numbers, Gj represents the three-dimensional trajectory of the repaired abnormal vertex, Gs h represents the three-dimensional trajectory of the h-th reference vertex, η h Represents the set weighting coefficient, satisfying ∑η h =1.

8. The intelligent error detection and repair system for 3D animation production according to claim 1, characterized in that: The specific process of judging whether the repair is successful is as follows: The repair trajectory of the abnormal vertex is re-embedded into the overall driving trajectory of the skeleton node to which it belongs, and the time domain deviation index value and frequency domain deviation index value of the skeleton node under the action instruction are recalculated. The new action deviation comprehensive value is determined based on this. If the new action deviation comprehensive value is less than the preset action deviation comprehensive threshold, it is judged that the repair is successful. Otherwise, it is judged that the repair fails.

9. The intelligent error detection and repair system for 3D animation production according to claim 8, characterized in that: Introducing the coupling analysis between bones and the direction of joint tension and compression to optimize the preset motion deviation comprehensive threshold: Calculate the bone rotation change and define the bone node r d and r l The bone rotation changes are: Where Δθ d Represents the skeleton node r d The bone rotation change, q d (t) represents the skeleton node r d The quaternion representation at time t, arccos(·) represents the inverse cosine function, and the skeleton node r is obtained in the same way. l The bone rotation change Δθ l ; For the bone node r d and r l Calculate their covariance Cov(Δθ d ,Δθ l ); Using covariance to determine the skeleton node r d and r l Coupling(r d ,r l ), the calculation formula is: Among them, σΔθd and σΔθl represent the standard deviation of the bone rotation change of the corresponding bone node; If Coupling(r d ,r l )≈1, it means that the two skeleton nodes are coordinated; Coupling d ,r l )≈0, it means that the bone rotation changes of the two bone nodes are independent; Calculate the moment M of the skeletal joint d (t) = r d (t)×F d (t); Among them, F d (t) represents the force applied to the skeletal node; Tension: If the force direction is stretching, the tension is equal to the bone node r d The absolute value of the moment. If the direction of the force is compression, the tension is 0; Compression: When the force direction is compression, the compression force is equal to the bone node r d The absolute value of the moment. If the direction of the force is tensile, the compressive force is 0; The sum of the tension and compression of all skeletal nodes is calculated, combined with the motion deviation value and coupling degree, and different weighted coefficients are assigned to the three to obtain the updated motion deviation comprehensive threshold.

10. The intelligent error detection and repair system in 3D animation production according to claim 1, characterized in that: The system also includes a display terminal for distinguishing and marking vertices that have been successfully repaired and vertices that have failed to be repaired using color coding.

Images