A defect detection method driven by animation character model motion coding

Abstract

This invention relates to the field of animation engine technology and discloses a defect detection method driven by motion coding of anime character models. The method includes: parsing a motion coding sequence containing a bone transformation matrix, a model vertex topology index, and a skinning weight matrix; constructing a topological association network representing vertex connection weights based on the model vertex topology index and the skinning weight matrix; extracting the rotation quaternion increment between adjacent frames and locking candidate abnormal edges when the rotation quaternion increment exceeds the limit; calculating the Euclidean distance between the weight vectors of the two vertices at the ends of the candidate abnormal edge to obtain the weight distribution energy value; and determining the defect edge and outputting defect report data when the weight distribution energy value exceeds the judgment threshold. This invention achieves deformation evaluation at the data attribute level, avoids intersection operations in three-dimensional geometric space, and effectively improves the detection efficiency and model output stability of the animation generation system.

Description

Technical Field

[0001] This invention belongs to the field of animation engine technology, and in particular relates to a defect detection method for motion coding driven animation character models. Background Technology

[0002] Currently, the dynamic driving of 3D animation characters typically relies on the bone transformation matrix to control the hybrid skinning of mesh vertices. The animation engine parses motion-coded data and applies it to the character's skeleton, driving the associated mesh to produce corresponding skinning deformations. As the precision of digital assets continues to improve, the scale of mesh vertices in animation models often reaches millions, accompanied by high-frequency motion capture data input. Existing technologies generally follow a processing path of first generating geometric meshes and then performing defect detection. In this mode, the system first completes the coordinate transformation calculation of all vertices in each frame to generate a complete physical mesh before it can identify mesh interlacing or skin tearing through geometric algorithms such as bounding box intersection. This post-detection mechanism based on geometric coordinate generation makes the system's computational load linearly positively correlated with model accuracy and frame rate, often leading to saturation of the rendering pipeline's computational load when processing complex dynamic scenes.

[0003] Industry experts have attempted to reduce the detection range by introducing local sampling or multi-level detail techniques. However, the skinning weight distribution of anime characters exhibits highly nonlinear characteristics, and simple geometric sampling often fails to cover abnormal deformations in clothing accessories or subtle topological structures. Existing detection paths are constrained by the preliminary step of geometric coordinate recalculation, making it impossible to predict underlying conflicts between motion features and topological constraints at the data parsing level. For example, Chinese invention patent application CN112802161A discloses a three-dimensional virtual character intelligent skinning method that extracts geodesic distance features between vertices and bones using a deep learning model. Current automated weight prediction and binding solutions, based on deep learning, implicitly rely on offline inference of static features. They address how to quickly assess topological damage in dynamic motion rather than how to do so in real time. Due to a lack of awareness of dynamic kinematic parameters such as rotational quaternions in motion encoding, the solutions cannot predict potential topological deformation defects during data flow when faced with high-sampling-rate motion capture data or extreme joint folding conditions. The predicted weight distribution energy lacks an adaptive threshold correction mechanism when facing nonlinear artistic deformation requirements, resulting in the need for manual intervention in the actual animation production process to correct model interlacing or collapse.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a predictive detection mechanism based on motion coding sequences and the topological properties of anime character models, so as to achieve accurate localization of dynamic defects without generating vertex physical coordinates. Summary of the Invention

[0005] This invention provides a method for detecting defects driven by motion coding in anime character models, comprising the following steps: Step S101: Obtain the motion coding sequence, which includes the skeleton transformation matrix of consecutive frames, the topological index of model vertices, and the skin weight matrix. The skin weight matrix records the normalized weight distribution ratio of each model vertex affected by different bone nodes. Step S102: Based on the topological index of model vertices and the skinning weight matrix, a topological association network representing the inherent constraint characteristics of the animation character model in the data attribute layer is constructed by parsing the topological adjacency relationship of each model vertex. The topological association network contains attribute edges formed by adjacent vertices and their corresponding weight distribution vectors. Step S103: Extract the change in the quaternion of bone rotation of the bone transformation matrix between adjacent frames, and calculate the motion intensity component of the bone node based on the change in the quaternion of bone rotation. When the motion intensity component exceeds the preset dynamic discrimination threshold, retrieve the attribute edge associated with the affected bone node in the topological association network as a candidate abnormal edge. Step S104: Call the skinning weight matrix to analyze the difference in weight vectors between the two vertices of the candidate abnormal edge, and calculate the Euclidean distance between the weight vectors of the two vertices under the normalized weight distribution ratio to obtain the weight distribution energy value that characterizes the degree of deformation of the candidate abnormal edge in the data layer. Step S105: When the weight distribution energy value exceeds the preset deformation judgment threshold, the candidate abnormal edge is determined as a defect edge, and the local clustering density distribution characteristics of the defect edge in the topological network are statistically analyzed. Defect report data containing defect type, defect occurrence frame number and defect location index is output to drive the animation engine to perform rendering interception.

[0006] Preferably, step S102 specifically includes: parsing the topological index of the model vertex to identify adjacent vertex pairs with shared edges; extracting the weight distribution coefficients of the corresponding adjacent vertex pairs in the skinning weight matrix; defining the adjacent vertex pairs as attribute edges in the topological association network, and using the weight distribution coefficients as inherent constraint attributes of the attribute edges to construct a parameterized topological association network.

[0007] Preferably, step S103 specifically includes: calculating the change in the bone rotation quaternion of the bone transformation matrix between adjacent time frames; calculating the relative angular velocity vector and displacement vector between bone nodes based on the change in the bone rotation quaternion; when the relative angular velocity vector or displacement vector exceeds the dynamic discrimination threshold, identifying the skinned vertices associated with the bone nodes, and marking the attribute edges adjacent to the skinned vertices in the topological association network as candidate abnormal edges.

[0008] Preferably, step S103 further includes: performing spherical linear interpolation deviation verification on the motion coding sequence to identify step abrupt change features caused by data acquisition breakpoints in the motion coding sequence; and performing signal truncation processing on sequence frames with step abrupt change features before calculating the weight distribution energy value to block the damaged motion coding from entering the skin deformation processing pipeline.

[0009] Preferably, in step S104, the weight distribution energy value is used to characterize the structural response measure of attribute edges tearing or volume collapse in the topological network driven by the skeleton transformation matrix; the weight distribution energy value increases monotonically with the increase of the weight distribution difference between the two vertices of the candidate abnormal edge, and is used to map the length expansion ratio of the attribute edge in the data attribute layer.

[0010] Preferably, the weight distribution energy value is obtained by calculating the difference in weight distribution between the two vertices of the candidate anomaly edge, and the calculation formula is as follows: Where E is the weighted distribution energy value, The skinning weight of the first vertex relative to the k-th bone. is the skinning weight of the second vertex relative to the k-th bone, and n is the total number of affected bones.

[0011] Preferably, the deformation determination threshold is dynamically corrected based on the global scaling parameters of the anime character model; when the motion coding sequence drives the anime character model to produce stretching deformation, a relaxation variable for the deformation determination threshold is added according to the non-realistic animation style characteristics of the anime character model to accommodate the non-linear artistic deformation requirements in the model representation.

[0012] Preferably, step S105 specifically includes: calculating the local clustering density of defective edges in the topological network; when the local clustering density exceeds a preset clustering level, extracting the vertex index corresponding to the defective edge and reconstructing the local topological data state to execute the interference verification logic, which is used to verify the normal vector conflict of adjacent vertices through the model vertex topological index.

[0013] Preferably, the output defect report data includes: generating a report file containing a defect type flag, the occurrence frame number, and a location index; mapping the location index to the skeletal hierarchy of the anime character model to locate the target bone node that caused the detected defect; and having a one-to-one correspondence between the defect location index and the vertex number in the model vertex topology index.

[0014] Preferably, when defective edges exist in the defect report data, the animation engine intercepts abnormal deformation frames before rendering output; the animation engine performs parameter correction based on the bone transformation matrix of the target bone node, and sends a sequence of control parameters for the target bone node to the animation production terminal to eliminate topological deformation defects of the anime character model in subsequent frames.

[0015] Compared with existing technologies, the defect detection method for animation character models driven by motion coding of the present invention has the following advantages: 1. In motion coding of anime character models, an undirected graph data structure is used to transform the static mesh attributes of the anime character model into topological adjacency relationships. By calculating the product of the absolute difference of skin weights between adjacent vertices and the relative rotation norm of the associated bones, a data attribute-driven edge distortion energy assessment mechanism is constructed. This transforms the dimensionality reduction of geometric interference recognition in traditional 3D space into a one-dimensional scalar retrieval based on topological edge attributes, making the detection process completely independent of the dependence on recalculating the physical coordinates of the model vertices. Since the computational complexity is only related to the local attributes of the topological edges and is no longer constrained by the number of polygons in the model, it effectively solves the problem of computational saturation caused by full deformation calculation when the real-time animation pipeline processes high-precision models with millions of polygons.

[0016] 2. By establishing a nonlinear mapping relationship between the bone scaling factor and the topological elastic tolerance attribute constant, the detection system is equipped with the ability to adaptively recognize exaggerated deformation features in non-realistic animation styles. When the motion sequence drives the model to produce artistic stretching or compression, the system dynamically adjusts the interference judgment threshold of the local mesh according to the global scaling parameter. While maintaining the logical integrity of the topological structure, it is compatible with the nonlinear deformation requirements in animation art expression. This mechanism avoids the false interception of exaggerated deformation that conforms to the artistic setting by the traditional rigid physical collision judgment, and improves the practical value of the detection results in complex animation creation environments.

[0017] 3. In the pre-deformation calculation stage, deviation rate verification of skeletal quaternion spherical linear interpolation is introduced. Motion coding temporal coherence analysis is used to construct a defense path against underlying data pollution. When the input motion data sequence causes a step change due to sensor noise or acquisition breakpoint, the system completes the targeted localization and signal truncation of abnormal features before the vertex deformation analysis logic is started, preventing damaged data encoding from entering the subsequent skin deformation pipeline. This technical logic and the back-end distortion energy assessment form a closed loop protection to ensure that the animation generation system maintains the stability of the output screen when facing unstable motion capture data sources. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of the motion coding defect detection method for anime character models according to the present invention; Figure 2 This is the logic diagram for calculating the weight energy of candidate abnormal edges and determining defects in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, lateral, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated.

[0021] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0022] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0023] A defect detection method driven by motion coding for anime character models includes the following steps: Step S101: Obtain the motion coding sequence, which includes the skeleton transformation matrix of consecutive frames, the topological index of model vertices, and the skin weight matrix. The skin weight matrix records the normalized weight distribution ratio of each model vertex affected by different bone nodes. Step S102: Based on the topological index of model vertices and the skinning weight matrix, a topological association network representing the inherent constraint characteristics of the animation character model in the data attribute layer is constructed by parsing the topological adjacency relationship of each model vertex. The topological association network contains attribute edges formed by adjacent vertices and their corresponding weight distribution vectors. Step S103: Extract the change in the quaternion of bone rotation of the bone transformation matrix between adjacent frames, and calculate the motion intensity component of the bone node based on the change in the quaternion of bone rotation. When the motion intensity component exceeds the preset dynamic discrimination threshold, retrieve the attribute edge associated with the affected bone node in the topological association network as a candidate abnormal edge. Step S104: Call the skinning weight matrix to analyze the difference in weight vectors between the two vertices of the candidate abnormal edge, and calculate the Euclidean distance between the weight vectors of the two vertices under the normalized weight distribution ratio to obtain the weight distribution energy value that characterizes the degree of deformation of the candidate abnormal edge in the data layer. Step S105: When the weight distribution energy value exceeds the preset deformation judgment threshold, the candidate abnormal edge is determined as a defect edge, and the local clustering density distribution characteristics of the defect edge in the topological network are statistically analyzed. Defect report data containing defect type, defect occurrence frame number and defect location index is output to drive the animation engine to perform rendering interception.

[0024] Preferably, step S102 specifically includes: parsing the topological index of the model vertex to identify adjacent vertex pairs with shared edges; extracting the weight distribution coefficients of the corresponding adjacent vertex pairs in the skinning weight matrix; defining the adjacent vertex pairs as attribute edges in the topological association network, and using the weight distribution coefficients as inherent constraint attributes of the attribute edges to construct a parameterized topological association network.

[0025] Preferably, step S103 specifically includes: calculating the change in the bone rotation quaternion of the bone transformation matrix between adjacent time frames; calculating the relative angular velocity vector and displacement vector between bone nodes based on the change in the bone rotation quaternion; when the relative angular velocity vector or displacement vector exceeds the dynamic discrimination threshold, identifying the skinned vertices associated with the bone nodes, and marking the attribute edges adjacent to the skinned vertices in the topological association network as candidate abnormal edges.

[0026] Preferably, step S103 further includes: performing spherical linear interpolation deviation verification on the motion coding sequence to identify step abrupt change features caused by data acquisition breakpoints in the motion coding sequence; and performing signal truncation processing on sequence frames with step abrupt change features before calculating the weight distribution energy value to block the damaged motion coding from entering the skin deformation processing pipeline.

[0027] Preferably, in step S104, the weight distribution energy value is used to characterize the structural response measure of attribute edges tearing or volume collapse in the topological network driven by the skeleton transformation matrix; the weight distribution energy value increases monotonically with the increase of the weight distribution difference between the two vertices of the candidate abnormal edge, and is used to map the length expansion ratio of the attribute edge in the data attribute layer.

[0028] Preferably, the weight distribution energy value is obtained by calculating the difference in weight distribution between the two vertices of the candidate anomaly edge, and the calculation formula is as follows: Where E is the weighted distribution energy value, The skinning weight of the first vertex relative to the k-th bone. is the skinning weight of the second vertex relative to the k-th bone, and n is the total number of affected bones.

[0029] Preferably, the deformation determination threshold is dynamically corrected based on the global scaling parameters of the anime character model; when the motion coding sequence drives the anime character model to produce stretching deformation, a relaxation variable for the deformation determination threshold is added according to the non-realistic animation style characteristics of the anime character model to accommodate the non-linear artistic deformation requirements in the model representation.

[0030] Preferably, step S105 specifically includes: calculating the local clustering density of defective edges in the topological network; when the local clustering density exceeds a preset clustering level, extracting the vertex index corresponding to the defective edge and reconstructing the local topological data state to execute the interference verification logic, which is used to verify the normal vector conflict of adjacent vertices through the model vertex topological index.

[0031] Preferably, the output defect report data includes: generating a report file containing a defect type flag, the occurrence frame number, and a location index; mapping the location index to the skeletal hierarchy of the anime character model to locate the target bone node that caused the detected defect; and having a one-to-one correspondence between the defect location index and the vertex number in the model vertex topology index.

[0032] Preferably, when defective edges exist in the defect report data, the animation engine intercepts abnormal deformation frames before rendering output; the animation engine performs parameter correction based on the bone transformation matrix of the target bone node, and sends a sequence of control parameters for the target bone node to the animation production terminal to eliminate topological deformation defects of the anime character model in subsequent frames.

[0033] Example 1: Current animation engines process anime character models containing a large number of polygonal mesh vertices and receive high-frequency motion capture data with a sampling frequency of 120Hz. Topological tearing caused by extreme joint folding or displacement can easily lead to excessive computational load for full mesh coordinate recalculation and geometric intersection. The system acquires the motion encoding sequence in the input stream, including the bone transformation matrix of consecutive frames, the topological index of model vertices, and the skinning weight matrix, and records the normalized weight distribution ratio of each model vertex affected by different bone nodes. The system parses the topological index of model vertices to identify adjacent vertex pairs with shared edges, and extracts the weight distribution coefficients of the corresponding adjacent vertex pairs in the skinning weight matrix. By defining adjacent vertex pairs as attribute edges in the topological association network and using the weight distribution coefficients as inherent constraint attributes of attribute edges, a parameterized topological association network representing the inherent constraint characteristics of the anime character model at the data attribute layer is constructed. This transforms the model's 3D spatial geometric interference recognition requirements into feature retrieval operations for the underlying relational graph data.

[0034] Before calculating deformation, the motion-coded sequence is subjected to spherical linear interpolation deviation verification. The interpolation deviation rate of the skeleton quaternion between adjacent frames is calculated. If a jump of π / 4 radians occurs in a single frame, the system identifies the step-change feature caused by the data acquisition breakpoint in the motion-coded sequence and performs signal truncation processing on the sequence frame with the step-change feature to block the damaged motion code from entering the skin deformation processing pipeline. For compliant motion-coded sequences, the change of the skeleton rotation quaternion of the skeleton transformation matrix between adjacent time frames is calculated. Based on the change of the skeleton rotation quaternion, the relative angular velocity vector and displacement vector between skeleton nodes are calculated. If the relative angular velocity vector or displacement vector exceeds the preset dynamic discrimination threshold, the system identifies the skin vertex associated with the skeleton node and marks the attribute edge adjacent to the skin vertex in the topological association network as a candidate abnormal edge.

[0035] The system calls the skinning weight matrix to analyze the difference in weight vectors between the two vertices of the candidate anomaly edge, and calculates the Euclidean distance between the weight vectors of the two vertices under the normalized weight distribution ratio. This yields the weight distribution energy value, which characterizes the degree of deformation of the candidate anomaly edge in the data layer. The calculation formula is as follows: Where E is the weighted distribution energy value, The skinning weight of the first vertex relative to the k-th bone. The skinning weight of the second vertex relative to the k-th bone, where n is the total number of affected bones; this weight distribution energy value is used to map the length expansion ratio of attribute edges in the data attribute layer. The system dynamically corrects the preset deformation judgment threshold based on the global scaling parameters of the anime character model. When the motion coding sequence drives the anime character model to produce stretching deformation, a relaxation variable for the deformation judgment threshold is added based on the non-realistic animation style characteristics of the anime character model. If the calculated weight distribution energy value exceeds the corrected deformation judgment threshold, the candidate abnormal edge is determined as a defective edge. The local clustering density of the defective edge in the topological network is calculated. If the local clustering density exceeds the preset clustering level, the system extracts the vertex index corresponding to the defective edge and reconstructs the local topological data state to execute. The interference verification logic verifies the normal vector conflict of adjacent vertices through the topological index of the model vertex. After verification, a report file containing defect type flag, occurrence frame number and position index is generated as defect report data. The position index is mapped to the skeletal hierarchy of the anime character model to locate the target bone node that caused the detected defect. There is a one-to-one correspondence between the defect position index and the vertex number in the topological index of the model vertex. The animation engine uses the defect report data to intercept abnormal deformation frames containing the defective edge before rendering output, and performs parameter correction based on the bone transformation matrix of the target bone node. It sends the control parameter sequence for the target bone node to the animation production terminal to eliminate the topological deformation defect of the anime character model in subsequent frames.

[0036] Example 2: The current animation engine testing platform is deployed on a graphics workstation with floating-point computing capabilities. It uses the CMU motion capture database as the test data source. The input data includes a high-frequency human motion encoding sequence with a sampling frequency of 120Hz, used to drive an anime character model with 1,500,000 polygonal mesh vertices. To simulate high-frequency spatial jitter noise caused by reflective point occlusion in real optical capture equipment, Gaussian white noise with a signal-to-noise ratio of 20dB is superimposed on the original signal source in the test environment. The system sets a dynamic discrimination threshold for the relative angular velocity vector. The engineering constraint for setting the parameter is to balance the detection rate of minor topological tears with the false alarm rate of high-dynamic movements. This threshold is calculated based on the local mesh vertex density around the affected bone nodes. When the local mesh vertex density increases, the probability of mesh interpenetration caused by minor displacements increases. The system reduces the dynamic discrimination threshold based on an inverse proportional function of vertex density. Taking a complex shoulder joint with a density of 450 vertices per square centimeter as an example, the calculated output dynamic discrimination threshold is 8.5 rad / s. This value is used as the benchmark condition for screening candidate abnormal edges.

[0037] The experiment extracted the step change intensity from motion capture data as the test variable, categorizing it into three levels: low-intensity, medium-intensity, and high-intensity change. Control group 1 used the traditional global bounding box geometric intersection algorithm, while control group 2 used the proposed solution's processing flow, omitting the spherical linear interpolation deviation verification step. An out-of-range control group had its dynamic discrimination threshold fixed at 25.0 rad / s. The experimental group used the aforementioned complete parameters and logical benchmarks, inputting a motion-coded sequence with a high-intensity change level. The original input data contained skeletal quaternion jump signals with instantaneous deviations reaching 0.85 rad within a single frame interval. Control group 2 included noisy anomalies... Directly injecting data into the topological network causes distortion in the calculation of the weight distribution energy values ​​of attribute edges within the network. The experimental group calculates the interpolation deviation rate of the quaternions of bones between adjacent frames. Since the calculated deviation rate of 0.85 rad is greater than the preset extreme value limit of π / 4 radians, the system truncates the corresponding abnormal signal. For the compliant sequence after noise removal, the experimental group extracts the change in the quaternion of bone rotation between adjacent frames and calculates that the relative angular velocity vector of the current bone node is 18.2 rad / s. Since this value is greater than the dynamic discrimination threshold of 8.5 rad / s, the system identifies and locks the associated skinned vertices, calls the skinned weight matrix to analyze the difference in weight vectors between adjacent vertices, and uses the formula... The calculated result is 1.45; where E is the weight distribution energy value. Let be the skinning weight of one vertex of the attribute edge relative to the k-th bone. is the skinning weight of the other vertex of the attribute edge relative to the k-th bone, and n is the total number of bones affected.

[0038] The system corrects the preset deformation judgment threshold based on the global scaling parameters of the anime character model, determining the corrected deformation judgment threshold to be 1.20. Since the weight distribution energy value of 1.45 is greater than the corrected deformation judgment threshold, the system identifies the corresponding attribute edge as a defect edge causing topological tearing, and calculates the local cluster density of such defect edges to be 65 edges per square centimeter. This density value is greater than the preset clustering level of 50 edges per square centimeter. The system outputs a defect location index containing specific target bone nodes. Experimental data tests reveal multi-dimensional parameter correlation attributes. When processing high-intensity abrupt change level noisy data, the average single-frame computation time of control group 1 increased to 850.5ms, while control group 2 was affected by step abrupt changes... The false alarm rate for defect detection reached 42.3%. After filtering out high-frequency jitter disturbances, the average calculation time per frame in the experimental group stabilized at 12.5ms, and the defect detection accuracy reached 98.2%. For the out-of-range control group, when the dynamic discrimination threshold was set to be greater than the calibration upper limit of 25.0 rad / s, the missed detection of edge topological tear features caused the defect false alarm rate to rise to 35.8%. When the dynamic discrimination threshold was set to be less than the calibration lower limit of 5.0 rad / s, the false alarm rate increased to 28.4%. The test values ​​confirmed that the dynamic discrimination threshold of the calculated output constitutes the parameter range for suppressing high-frequency noise disturbances and locating local topological distortions, maintaining the accuracy of animation topological deformation detection while reducing the load of three-dimensional geometric intersection calculation.

[0039] Example 3: Current animation engines handle anime character models with non-uniform scaling characteristics. A fixed deformation threshold can easily cause false alarms during the overall scaling-up phase or missed alarms during the scaling-down phase. The system executes a dynamic calibration program for the deformation threshold, reads the bone transformation matrix of the root bone node of the anime character model, extracts the main diagonal elements of the matrix to calculate the model's global scaling parameters, and reads the preset basic deformation tolerance constant 0.55. The system calculates the product of the global scaling parameter and the basic deformation tolerance constant to generate the baseline deformation threshold. The system extracts the local stretching ratio of the current bone node at a specific time frame. Since the local stretching ratio is greater than 1.0, the system calculates the product of the baseline deformation threshold and the square root of this local stretching ratio, outputs the corrected deformation threshold, constructs a comparison value to determine whether the weight distribution energy value exceeds the limit, selects the arithmetic mean of the main diagonal elements of the model's root bone transformation matrix as the global scaling parameter S, and combines the basic deformation tolerance constant C with the global scaling parameter S and the non-uniform scaling parameter S. The modified deformation judgment threshold H is obtained by multiplying the preset style gain coefficient G of the realistic animation style. The style gain coefficient G is determined based on the offline annotation value of the extreme stretch sampling frame under the animation style. The value is selected from 1.2 to 1.5 for cartoon style and 1.0 for realistic style. A quantitative mapping relationship between the data attribute layer and artistic expression requirements is established without recalculating vertex coordinates. The system determines the candidate abnormal edge as the defect edge based on the modified deformation judgment threshold. It searches for second-order adjacent attribute edges with a topological distance of 2 in the topological association network. The total number of defect edges and the total number of attribute edges within the search range are counted. The ratio of the total number of defect edges to the total number of attribute edges is calculated as the local cluster density. Since the local cluster density is greater than the preset clustering level of 0.15, the vertex index corresponding to the defect edge is extracted. The three-dimensional space coordinates of the model vertex in the current frame are read according to the vertex index. The current face normal vector of the triangle facet shared by the defect edge is calculated using the coordinates of adjacent vertices. The initial face normal vector of the model under the resting bound posture is extracted.

[0040] For candidate anomaly regions with a local clustering density exceeding 0.15, extract the set of vertex indices associated with defect edges. The graphics processing unit performs skinning coordinate transformation operations on the local vertices within the set to generate the local spatial coordinates of the current frame, and calculates the unit normal vector of the triangular facet formed by adjacent vertices in the current frame. With the unit normal vector at the initial attitude The dot product is used to verify the interference. If the dot product result is less than 0, it indicates that a local area has experienced normal vector flipping, leading to mesh interpenetration defects. To avoid full geometric intersection operations, the topology status of high-risk areas is verified. The dot product eigenvalue of the current face normal vector and the initial face normal vector is calculated. Since the dot product eigenvalue is less than 0, the system determines that the model mesh has experienced normal vector flipping and interpenetration. The system records the defect edge data of the normal vector flipped defect and generates defect report data containing defect type flag, occurrence frame number, and location index. The defect report data is then transmitted to the rendering pipeline. The animation engine locates the target bone node based on the defect location index image, intercepts abnormal deformation frames containing normal conflicts, calculates the coordinate backtracking compensation component to restore the normal vector to a state greater than 0 based on the bone transformation matrix of the target bone node, and sends a control parameter sequence covering the coordinate backtracking compensation component to the animation production terminal to reset the topological coordinates of the model vertices that caused the interpenetration.

[0041] Example 4: When the system faces the deployment of anime character models with a completely new hierarchical topology, it executes an offline calibration procedure. The system imports a benchmark motion sequence, which includes continuous data frames transitioning from the model's standard bound posture to extreme joint folding movements. Driven by the benchmark motion sequence, the system extracts skin deformation data from the model surface frame by frame and simultaneously calculates the weight distribution differences of all attribute edges in the topological network. When the critical time frame at which the first physical interpenetration phenomenon occurs in the 3D mesh patch is detected, the system records the weight distribution energy value corresponding to the associated candidate abnormal edge as critical sample data. This data acquisition process is repeated across multiple benchmark models with different polygon mesh resolutions to construct a critical sample dataset, and the arithmetic mean and standard deviation of the critical sample dataset are calculated. The system calculates the basic deformation tolerance constant according to the formula C=μ-σ, where C is the basic deformation tolerance constant, μ is the arithmetic mean of the critical sample dataset, and σ is the standard deviation of the critical sample dataset. The system writes the calculated basic deformation tolerance constant as a fixed value into the data dictionary of the underlying verification logic to establish a numerical comparison benchmark based on the physical interpenetration boundary characteristics.

[0042] Before the animation production pipeline officially connects the anime character model to be tested into the real-time driving system, a pre-deployment debugging procedure is executed. The system traverses the topological index of the model vertex of the anime character model to be tested, and divides the corresponding local joint regions according to the bone hierarchy structure. The number of vertices per square centimeter in each local joint region is counted to generate the local mesh vertex density of that region, and the preset safe angle variable envelope feature parameters are extracted. The system uses the least squares method to fit the numerical constraint coefficient between the local mesh vertex density and the safe angle variable envelope feature parameters, and constructs an inverse proportional mapping function with the local mesh vertex density as the independent variable and the dynamic discrimination threshold as the dependent variable. Based on the inverse proportional mapping function, the threshold data of the anime character model to be tested at each bone node position is calculated, and the identifier of each bone node is bound to the corresponding threshold data and stored in the address space of the hardware memory to generate an association lookup table. When motion capture data flows into the driving pipeline, the system retrieves the corresponding value in the association lookup table based on the identifier of the target bone node, forming the parameter monitoring boundary for adapting to high-frequency motion coding driving.

[0043] Example 5: When the system encounters an abnormal situation where physical occlusion causes a loss of continuity in the motion capture signal source, it executes an interpolation deviation rate quantization calculation procedure and a fault-tolerant replacement mechanism for the motion coding sequence. It extracts the skeletal rotation quaternions from adjacent time frames, obtains the first unit quaternion from the previous time frame and the second unit quaternion from the current time frame, and calculates the absolute value of the inner product of the first and second unit quaternions. The system then applies the formula... Calculate the interpolation bias rate of the skeletal quaternion, where θ is the interpolation bias rate. For the first unit of quaternions, The second unit is a quaternion. The interpolation deviation rate corresponds to the angular displacement of the skeleton rotation between two adjacent frames. The system compares the interpolation deviation rate with the preset extreme limit π / 4 radians to form a verification benchmark for the validity of the data frame state.

[0044] If the calculated interpolation deviation rate exceeds the preset extreme value limit, the system determines that the motion coding sequence of the current time frame has a step-change characteristic. While truncating the current abnormal signal, it activates the historical data compensation logic. It reads the historical compliant sequence stored in the hardware memory, extracts the five consecutive valid skeletal rotation quaternions of the nearest current time frame to construct a sliding time window, and calculates the average angular velocity vector of adjacent quaternions within the sliding time window. Using the skeletal rotation quaternion of the nearest time frame as the reference starting point, the system performs spherical linear interpolation to generate a predicted quaternion, which is then injected into the skin deformation processing pipeline as compensation replacement data for the current time frame. This maintains the temporal continuity of the driving data and eliminates the stagnation in topological association network feature retrieval caused by breakpoints in the underlying acquisition. The system also calculates the skeletal rotation quaternions of adjacent time frames. and The absolute value of the inner product, according to The interpolation deviation rate θ represents the angular displacement of the skeleton between two adjacent frames. When the interpolation deviation rate exceeds a preset extreme limit of π / 4 radians, it is determined that the data frame has a step-like abrupt change characteristic caused by sensor occlusion. The average angular velocity vector of five consecutive valid frames in the historical sliding window in the hardware memory is extracted and used to perform spherical linear interpolation prediction on the current frame to maintain the continuity of the temporal processing of the topological network under data loss conditions. The embodiments of this application have been described above with reference to the accompanying drawings. In the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many forms under the guidance of this application without departing from the spirit and scope of protection of this application, and all of them are within the protection of this application.

Claims

1. A defect detection method driven by motion coding of anime character models, characterized in that, Includes the following steps: Step S101: Obtain the motion coding sequence, which includes the skeleton transformation matrix of consecutive frames, the topological index of model vertices, and the skin weight matrix. The skin weight matrix records the normalized weight distribution ratio of each model vertex affected by different bone nodes. Step S102: Based on the topological index of model vertices and the skinning weight matrix, a topological association network representing the inherent constraint characteristics of the animation character model in the data attribute layer is constructed by parsing the topological adjacency relationship of each model vertex. The topological association network contains attribute edges formed by adjacent vertices and their corresponding weight distribution vectors. Step S103: Extract the change in the quaternion of bone rotation of the bone transformation matrix between adjacent frames, and calculate the motion intensity component of the bone node based on the change in the quaternion of bone rotation. When the motion intensity component exceeds the preset dynamic discrimination threshold, retrieve the attribute edge associated with the affected bone node in the topological association network as a candidate abnormal edge. Step S104: Call the skinning weight matrix to analyze the difference in weight vectors between the two vertices of the candidate abnormal edge, and calculate the Euclidean distance between the weight vectors of the two vertices under the normalized weight distribution ratio to obtain the weight distribution energy value that characterizes the degree of deformation of the candidate abnormal edge in the data layer. Step S105: When the weight distribution energy value exceeds the preset deformation judgment threshold, the candidate abnormal edge is determined as a defect edge, and the local clustering density distribution characteristics of the defect edge in the topological network are statistically analyzed. Defect report data containing defect type, defect occurrence frame number and defect location index is output to drive the animation engine to perform rendering interception.

2. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, Step S102 specifically includes: parsing the topological index of the model vertices to identify adjacent vertex pairs with shared edges; extracting the weight distribution coefficients of the corresponding adjacent vertex pairs in the skinning weight matrix; defining adjacent vertex pairs as attribute edges in the topological association network, and using the weight distribution coefficients as inherent constraint attributes of the attribute edges to construct a parameterized topological association network.

3. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, Step S103 specifically includes: calculating the change in the bone rotation quaternion of the bone transformation matrix between adjacent time frames; calculating the relative angular velocity vector and displacement vector between bone nodes based on the change in the bone rotation quaternion; when the relative angular velocity vector or displacement vector exceeds the dynamic discrimination threshold, identifying the skinned vertices associated with the bone nodes, and marking the attribute edges adjacent to the skinned vertices in the topological association network as candidate abnormal edges.

4. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, Step S103 further includes: performing spherical linear interpolation deviation verification on the motion coding sequence to identify step abrupt change features caused by data acquisition breakpoints in the motion coding sequence; and performing signal truncation processing on sequence frames with step abrupt change features before calculating the weight distribution energy value to block the damaged motion coding from entering the skin deformation processing pipeline.

5. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, In step S104, the weight distribution energy value is used to characterize the structural response measure of attribute edges in the topological network being torn or collapsed under the drive of the skeleton transformation matrix; the weight distribution energy value increases monotonically with the increase of the weight distribution difference between the two vertices of the candidate abnormal edge, and is used to map the length expansion ratio of the attribute edge in the data attribute layer.

6. The defect detection method for motion coding driven animation character models according to claim 5, characterized in that, The weight distribution energy value is obtained by calculating the difference in weight distribution between the two vertices of the candidate anomaly edge. The calculation formula is as follows: Where E is the weighted distribution energy value, The skinning weight of the first vertex relative to the k-th bone. is the skinning weight of the second vertex relative to the k-th bone, and n is the total number of affected bones.

7. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, The deformation judgment threshold is dynamically adjusted based on the global scaling parameters of the anime character model. When the motion coding sequence drives the anime character model to produce stretching deformation, a relaxation variable for the deformation judgment threshold is added according to the non-realistic animation style characteristics of the anime character model to accommodate the non-linear artistic deformation requirements in the model representation.

8. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, Step S105 specifically includes: calculating the local clustering density of defective edges in the topological network; when the local clustering density exceeds the preset clustering level, extracting the vertex index corresponding to the defective edge and reconstructing the local topological data state to execute the interference verification logic, which is used to verify the normal vector conflict of adjacent vertices through the model vertex topological index.

9. The defect detection method for motion coding driven animation character models according to claim 1, characterized in that, The output defect report data includes: generating a report file containing a defect type flag, the occurrence frame number, and a location index; mapping the location index to the skeletal hierarchy of the anime character model to locate the target bone node that caused the detected defect; and ensuring a one-to-one correspondence between the defect location index and the vertex number in the model's vertex topology index.

10. A defect detection method for motion coding driven animation character models according to claim 9, characterized in that, When defective edges are present in the defect report data, the animation engine intercepts abnormal deformation frames before rendering output. The animation engine performs parameter corrections based on the bone transformation matrix of the target bone node and sends a sequence of control parameters for the target bone node to the animation production terminal to eliminate topological deformation defects of the anime character model in subsequent frames.

Images