AI analysis-based motion capture actor intelligent screening method and system

Abstract

The invention provides a motion capture actor intelligent screening method and system based on AI analysis, and the method comprises the steps: generating a structured motion task template through a semantic understanding unit based on an input natural language motion description text, and the motion task template comprises a motion sequence, an emotion vector and a key part list; based on the action task template, guiding the actor to execute a standardized action through a data acquisition and processing unit, and acquiring skeleton action data of the actor; calculating a control capability score and a system suitability score through a capability evaluation unit based on the skeleton action data and the key part list; and calculating an expression consistency score through a semantic alignment unit, and finally, fusing the expression consistency score, the control ability score and the system suitability score to generate a comprehensive score for actor screening.

Description

Technical Field

[0001] This invention belongs to the field of digital film and television, and in particular relates to a method and system for intelligent selection of motion capture actors based on AI analysis. Background Technology

[0002] With the increasing reliance on motion capture in film, games, virtual digital humans, and virtual reality, the rapid and efficient selection of suitable motion capture actors has become a critical issue for the industry. Existing selection methods typically rely on subjective human observation to determine whether an actor can perform the required movements or possesses certain physical qualities. This method lacks standardized criteria, is prone to subjective bias, and lacks quantifiable evaluation criteria, easily leading to a mismatch between candidates and the actual motion capture task, resulting in capture failures or repeated re-recording, increasing time and costs. Some research has attempted to introduce computer vision and action recognition methods to assist in the selection process, but these often remain at the level of superficial action recognition or simple pose comparison, failing to systematically consider the multi-dimensional performance of actors in actual motion capture environments. For example, common methods neglect whether the actor's movements can be stably captured by the equipment, lack adaptability modeling in noisy and occluded environments, and cannot measure the consistency between the movements and the director's semantic intent. Therefore, how to introduce a unified semantic task description during the selection stage, stability and amplitude constraints during the acquisition process, data adaptability evaluation in complex scenes, and quantification of the consistency between semantics and movements remain core problems that current technologies have not yet solved. Summary of the Invention

[0003] The purpose of this invention is to design an intelligent screening method and system for motion capture actors based on AI analysis. This system can generate structured task templates through natural language motion description parsing, ensuring that each candidate is tested under the same conditions. It integrates control capabilities, system adaptability, and expression consistency, and outputs a comprehensive score that can be directly used for screening and recommendation.

[0004] To achieve the above objectives, a first aspect of the present invention provides an intelligent screening method for motion capture actors based on AI analysis, the method comprising:

[0005] Based on the input natural language action description text, a structured action task template is generated through a semantic understanding unit. The action task template includes an action sequence, an emotion vector, and a list of key parts. The action sequence is obtained by matching the semantic embedding of the text with a standard action template library. The emotion vector is obtained by extracting emotion adverbs from the text and mapping them to a three-dimensional emotion space. The list of key parts is obtained by identifying body part entities in the text and mapping them to standard joint identifiers.

[0006] Based on the action task template, the data acquisition and processing unit guides the actor to perform standardized actions and collects their skeletal action data; wherein, the execution rhythm is adjusted by the rhythm dimension in the emotion vector; during the acquisition process, image data is acquired through a multi-camera system, three-dimensional joint coordinates are extracted by the skeleton estimation model, and the joint movement amplitude and stability residual index are calculated.

[0007] Based on the skeleton motion data and the list of key parts, the control capability score and system adaptability score are calculated by the capability assessment unit. The control capability score is obtained by analyzing the sliding consistency error and trajectory offset within the motion segment. The system adaptability score is obtained by simulating occlusion and reconstructing the skeleton data, and evaluating the reconstruction error and its volatility.

[0008] An expression consistency score is calculated using a semantic alignment unit. This score is based on the similarity between the semantic embedding of the task action sequence and the feature embedding of the actor's actual actions, and is weighted and penalized by combining emotion vectors and spatial offsets of key parts. Finally, the expression consistency score, control ability score, and system adaptability score are fused to generate a comprehensive score for actor selection.

[0009] Furthermore, the semantic understanding unit adopts a semantic encoding model based on the Transformer architecture, and performs action sequence matching by calculating the similarity between the semantic embedding vector of the input text and the embedding vector of each template in the standard action template library.

[0010] Furthermore, the generation of emotion vectors includes: identifying emotion adverbs from text semantic embeddings and obtaining their attention scores, and projecting the emotion adverbs onto a three-dimensional emotion space containing emotion intensity, emotion polarity, and rhythm urgency through a preset mapping matrix.

[0011] Furthermore, the generation of the key parts list includes: extracting body part entities from the text using entity recognition technology, and querying a predefined mapping table to convert the body part entities into standard joint identifiers.

[0012] Furthermore, the data acquisition and processing unit guides actors to perform actions by displaying standard demonstration animations and graphic descriptions on a prompt screen, where the playback rate of the demonstration animations is linearly adjusted by the rhythm urgency dimension value in the emotion vector.

[0013] Furthermore, the joint motion amplitude is calculated as the average Euclidean distance of the joint relative to its initial position during the execution of the action; the stability residual index is obtained by calculating the normalized variance of the joint's position within the sliding time window and the mean of the window.

[0014] Furthermore, the control capability score is obtained through sliding consistency error analysis, which specifically includes: for each key joint in each action segment, calculating the difference between the current frame joint position and the mean of the sliding window, and introducing a trajectory offset suppression term to evaluate the degree of trajectory drift.

[0015] Furthermore, the system adaptability score is obtained through the following steps: adding simulated occlusion perturbations to the original skeleton data, reconstructing the occluded data using a time series recovery network, calculating the reconstruction error, and evaluating the error volatility by combining the occlusion-sensitive regularization term.

[0016] Furthermore, the semantic alignment unit uses a multi-layer Transformer encoder to convert action sequences into semantic vectors, and uses temporal convolution and bidirectional gated recurrent units to convert skeleton action data into feature vectors. The expression consistency score is obtained by calculating the similarity between the two and combining the sentiment modulation factor and spatial offset penalty term.

[0017] A second aspect of the invention provides an AI-based intelligent screening system for motion capture actors, the system comprising:

[0018] The instruction processing module is used to generate a structured action task template based on the input natural language action description text through a semantic understanding unit. The action task template includes an action sequence, an emotion vector, and a list of key body parts. The action sequence is obtained by matching the semantic embedding of the text with a standard action template library. The emotion vector is obtained by extracting emotion adverbs from the text and mapping them to a three-dimensional emotion space. The list of key body parts is obtained by identifying body part entities in the text and mapping them to standard joint identifiers.

[0019] The skeleton data acquisition module is used to guide actors to perform standardized movements based on the action task template through the data acquisition and processing unit, and to acquire their skeleton movement data; wherein, the execution rhythm is adjusted by the rhythm dimension in the emotion vector; during the acquisition process, image data is acquired through a multi-camera system, three-dimensional joint coordinates are extracted through the skeleton estimation model, and joint movement amplitude and stability residual index are calculated;

[0020] The capability assessment module is used to calculate the control capability score and system adaptability score based on the skeleton motion data and the list of key parts through the capability assessment unit. The control capability score is obtained by analyzing the sliding consistency error and trajectory offset within the motion segment. The system adaptability score is obtained by simulating occlusion and reconstructing the skeleton data, and evaluating the reconstruction error and its volatility.

[0021] The results generation module is used to calculate the expression consistency score through the semantic alignment unit. The score is based on the similarity between the semantic embedding of the task action sequence and the feature embedding of the actor's actual action, and is weighted and penalized by combining the emotion vector and the spatial offset of key parts. Finally, the expression consistency score, control ability score and system adaptability score are merged to generate a comprehensive score for actor selection.

[0022] The beneficial technical effects of the present invention are at least as follows:

[0023] To address the aforementioned issues, this invention provides an AI-based intelligent selection method and system for motion capture actors. It generates structured task templates through natural language motion description parsing, ensuring each candidate undergoes testing under identical conditions. By employing high-precision skeleton data acquisition and motion stability measurement mechanisms, it introduces a quantitative evaluation of candidate control during the acquisition phase. Through sliding consistency error analysis and occlusion recovery residual evaluation, it establishes quantitative indicators reflecting execution stability and system adaptability. By combining a semantic alignment embedding model with emotion modulation and key part spatial offset penalties, it proposes a method for calculating expression consistency in motion capture scenarios. Finally, through a unified fusion scoring function, it integrates control ability, system adaptability, and expression consistency, outputting a comprehensive score that can be directly used for selection and recommendation. This invention not only solves the problems of existing selection processes relying on subjective experience, lacking systematicity and comparability, but also introduces indicators highly correlated with actual motion capture scenarios, making the selection results closely related to subsequent acquisition quality, thus improving the scientific rigor and reliability of the selection process. Attached Figure Description

[0024] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0025] Figure 1 This is a flowchart of the AI-based motion capture actor intelligent selection method of the present invention.

[0026] Figure 2 This is a framework diagram of the AI-based motion capture actor intelligent selection system of the present invention. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0028] In one or more embodiments, such as Figure 1 As shown, an intelligent selection method for motion capture actors based on AI analysis is disclosed, the method including the following:

[0029] S1: Based on the input natural language action description text, a structured action task template is generated through a semantic understanding unit. The action task template includes an action sequence, an emotion vector, and a list of key body parts. The action sequence is obtained by matching the semantic embedding of the text with a standard action template library. The emotion vector is obtained by extracting emotion adverbs from the text and mapping them to a three-dimensional emotion space. The list of key body parts is obtained by identifying body part entities in the text and mapping them to standard joint identifiers.

[0030] Specifically, directors or animation directors typically describe action requirements using natural language instructions, such as "quickly turn around and glare forward, extending your right arm straight down towards the ground." To enable the system to subsequently guide actors' performances in a standardized manner and compare execution results, this free-flowing form of expression must be transformed into structured action task templates.

[0031] First, enter the action text. (Strings no longer than 128 characters) will enter the embedding encoding module. The encoding model used is BERT-Base, with a 12-layer Transformer structure and an embedding dimension of 768. The model has been fine-tuned on an internally built action description corpus for action semantic understanding. After word vector encoding and context modeling, the output representation is as follows:

[0032] ;

[0033] in, It is action text The contextual semantic embedding result is a form like tensor, The length of the token sequence. This embedding vector is then used to construct the action task template. , containing action sequences Emotion Vector List of key parts .

[0034] Action sequence The calculation is performed as follows: The system maintains a standard action template library, pre-storing semantic embedding representations of several standard actions. Calculate the cosine similarity with each template embedding vector, and select the one with the highest similarity. Each action is considered as a sequence of task actions. For example, if... To "quickly turn around and glare ahead", then Possible matches are ["turn around", "gaze", "point"].

[0035] Emotion Vector The generation is based on a weighted embedding of sentiment adverbs in the text. The system... Extract all adverbs and their attention scores (e.g., "quickly," "angrily"), and project them onto a three-dimensional emotion space using a mapping matrix, corresponding to: emotion intensity, emotion polarity, and rhythmic urgency, respectively. If "slowly squat down" appears, then... Possibly This indicates low intensity, neutrality, and low urgency.

[0036] List of key parts The generation employs a dictionary-based entity recognition matching mechanism. The system searches for phrases containing anatomical entities in the text, such as "right arm," "left knee," and "both hands," and then maps them to the internally used motion capture joint IDs through a lookup table; for example, "right wrist" is mapped to ID_07. If the text does not explicitly mention body parts, then... It includes the main joints predefined in the motion template by default.

[0037] Finally, the system constructs the output structure:

[0038] Action sequence, type is a list of strings, source is Matching results with the action template embedding vector library;

[0039] : Emotion expression vector, which is a three-dimensional floating-point array and is derived from the weighted projection of the embedding of emotion adverbs;

[0040] : A list of key action part IDs, of type integer, sourced from text entity recognition and standard joint mapping.

[0041] The system marks this structure as This serves as input for subsequent actor performance instructions, used to standardize action execution. For example, if the input text is "Slowly squat down and look straight ahead," the system might output: , , .

[0042] S2: Based on the action task template, the data acquisition and processing unit guides the actor to perform standardized actions and collects their skeletal action data; wherein, the execution rhythm is adjusted by the rhythm dimension in the emotion vector; during the acquisition process, image data is acquired through a multi-camera system, three-dimensional joint coordinates are extracted by the skeleton estimation model, and the joint movement amplitude and stability residual index are calculated;

[0043] Specifically, this step aims to transform the standardized action task structure generated in step one. It is applied to real candidate auditions to complete standardized actions under system guidance, and generates structured motion data through high-precision skeleton data acquisition equipment. This provides a foundation for subsequent control capability analysis and expression consistency modeling. This step is specifically designed for the unique application scenario of motion capture actor selection, introducing a data acquisition stability assessment mechanism, a dynamic rhythm control mechanism, and a skeleton sampling constraint mechanism based on scene sparsity, which are of significance for the selection of candidates. This ensures that data acquisition not only provides a physical basis for subsequent modeling, but also introduces the judgment of the differences in the candidates' abilities in the acquisition itself.

[0044] The system receives the output structure from the previous step. Then, first read the action sequence. Each action node in the system calls upon standard demonstration animations and graphic descriptions for each task, which are then displayed on the prompt screen to guide the actors in completing all task instructions in the target sequence. The playback rhythm of the prompt screen is synchronized by emotional vectors. The urgency dimension is adjusted, and this value is in Linear mapping within the interval to the cue rhythm modulation ratio This is used to control the presentation speed and the rhythm of the voice prompts. For example, ,like Then the animation playback rate is 1 / 3 of the original rate. This mechanism ensures that the candidates' performance pace is consistent with the expected pace of the task, effectively preventing "pace misalignment" from interfering with subsequent evaluations.

[0045] During the execution of the action, the system acquires skeleton data through four synchronous infrared cameras deployed in the 3D scene, with a frame rate of [frame rate per second]. The system utilizes a skeleton fitting module built using the OpenPose motion estimation framework to perform real-time analysis of image data and extract the 3D spatial positions of key joints. This module performs 2D joint localization on each frame of the image, and then combines binocular depth projection and time-series smoothing to obtain the 3D skeleton points. ,in Indicates the first frame, Indicates the first A key joint, and All output joint numbers must strictly match the structure. The defined joint ID sequence ensures that the variable structure is aligned.

[0046] To address common issues like insufficient range of motion and blurry movements in motion capture actors, the system introduces a motion expression range normalization term during data acquisition. This term measures the ratio between the range of motion of a joint and the length of its static structure. For any joint... ,exist In an intraframe action sequence, the average expression amplitude is denoted as:

[0047] ;

[0048] in, Indicates the first The first frame The three-dimensional coordinates of each joint This represents the initial position of the joint at the start frame of the action. This expressed amplitude is used to evaluate whether the candidate has performed sufficient physical movement amplitude as required by the instruction. If the range of motion is significantly lower than the historical reference range of the preset motion template (established by the standard library of qualified motion capture actors), it may indicate that the range of motion is insufficient and is not suitable for certain roles that require a high degree of clear expression of motion.

[0049] To further consider the skeleton localization jitter caused by factors such as occlusion and low light in real-world scenarios, and to improve data quality and introduce an early metric for candidate "body control stability," the system introduces a residual index based on spatial filtering difference for skeleton stability. This indicator comprehensively measures the performance within the sliding time window. Does the joint stably oscillate around a certain central trajectory? Record the joint... In the The position of the frame is ,but:

[0050] ;

[0051] in, This is the size of the sliding window, typically set to a value of [value to be filled in]. ; To prevent division by zero for the minimum value, take The smaller the value of this indicator, the closer the joint is to a stable trajectory during execution; the larger the value, the more divergent or fluctuating the movement, reflecting differences in the body's control capabilities. The system can provide real-time alerts for unstable joints during data collection or use it as a basis for stability scoring in subsequent analysis.

[0052] S3: Based on the skeleton motion data and the list of key parts, the control capability score and system adaptability score are calculated by the capability assessment unit; the control capability score is obtained by analyzing the sliding consistency error and trajectory offset within the motion segment; the system adaptability score is obtained by simulating occlusion and reconstructing skeleton data, and evaluating the reconstruction error and its volatility.

[0053] Specifically, in the first two steps, the system has completed the processing of the action task structure. The generation of the data guides candidates to complete the standardized execution of the task, and the acquisition of 3D skeleton motion data tensors is achieved through the motion capture system. The core objective of this step is to... Based on this, and combined with the action sequence in the task requirements and list of key joints We will construct a reproducible and targeted modeling process to evaluate whether candidates have the ability to complete motion capture tasks with high quality from two dimensions: motion control capability and system adaptability.

[0054] Unlike common posture quality assessment methods, this step is specifically designed for motion capture actor selection, emphasizing whether candidates can "stablely and systematically perform standardized tasks." This means not only must the motion trajectory be stable, but the motion data must also be recoverable even in the event of system sampling anomalies (such as occlusion or frame drops) under actual acquisition conditions. This "collectibility" assessment is an issue not considered in traditional actor training and human posture analysis tasks.

[0055] Firstly, for the evaluation of motion control capability, we propose a method based on sliding consistency error analysis. This is used to measure whether a candidate can maintain stable joint control during a movement. Each standard movement segment... Corresponding skeleton data sub-segment This segment is extracted using task-defined frame annotations. For each key joint... We calculate its sliding control consistency index in this segment:

[0056] ;

[0057] in, Indicates the first Frame number The three-dimensional coordinates of each joint Indicates from frame arrive The mean of the sliding window, This represents the total number of frames in this action segment. To avoid division by zero for extremely small constants, It is a regularization weight term.

[0058] Here, in the second regularization term, we introduce a trajectory drift suppression term in the control consistency evaluation, which determines whether the sliding average deviates from the attitude center at the beginning of the motion. This design is specifically for motion capture applications because during long-duration motions, candidates may experience spatial misalignment, leading to a decrease in system capture quality, and this "trajectory drift" cannot be identified by traditional smoothness metrics.

[0059] By averaging over all joints and summing over all task segments, we obtain the overall control score:

[0060] ;

[0061] The lower the score, the stronger the candidate's control, the more stable and consistent their action execution, and the more suitable they are for working in demanding motion capture scenarios.

[0062] To evaluate system adaptability, we consider recovery capabilities under occlusion, noise, and missing data conditions. Specifically, we analyze the original data tensor... Add simulated occlusion perturbation and randomly block Construct a perturbation tensor from partial joint data in the frame. The reconstructed data is then performed using a lightweight time-series recovery network, outputting a reconstructed tensor. .

[0063] The network structure consists of three parts:

[0064] Three 1D convolutional layers (kernel=3, channel=64 / 64 / 128) extract local temporal features;

[0065] A bidirectional GRU module is used for time series modeling.

[0066] The attention module of residual connections is used to fuse context to recover occluded data.

[0067] The system defines the reconstruction residual as the average normalized error for each joint and adds an occlusion-sensitive regularization term to consider whether the reconstruction error increases significantly at joints with a high probability of occlusion.

[0068] ;

[0069] in, For the first A set of frames in which joints are occluded. Indicates joint The variance of the error in the occluded segment. is the regularization coefficient. This regularization term is designed to prevent the model from ignoring error fluctuations only when the average error is low, ensuring that the system's adaptability is not only excellent in static accuracy but also robust in stability.

[0070] Ultimately, this step outputs two metrics:

[0071] Control ability score The lower the better, as it indicates stable movement;

[0072] System compatibility score The lower the better, as it indicates stronger resistance to shading.

[0073] S4: Calculate the expression consistency score through the semantic alignment unit. The score is based on the similarity between the semantic embedding of the task action sequence and the feature embedding of the actor's actual action, and is weighted and penalized by combining the emotion vector and the spatial offset of key parts. Finally, the expression consistency score, control ability score and system adaptability score are merged to generate a comprehensive score for actor selection.

[0074] Specifically, this step, building upon the completion of the first three steps, is the crucial final stage for intelligent screening and decision-making. Its task is to analyze the skeletal data of candidates after they have completed specified actions. Task template structure and action execution ability score System compatibility score It determines whether the candidate's actions are consistent with the original semantic task and generates the final screening and recommendation score. This step is the "output point" of the entire process, determining whether the screening system possesses the essential function of judging "whether the actor is suitable." Therefore, it not only needs to output accurate scoring results, but also needs to reflect an expressive modeling mechanism and system fusion mechanism specifically designed for motion capture actor screening scenarios in its modeling design.

[0075] Traditional motion analysis or human pose recognition systems often focus on "motion classification" or "motion completion rate," but this is far from sufficient in motion capture screening scenarios. The core elements determining the success of a motion capture task are whether the candidate accurately "understands the task semantics," whether they "express actions in a way that the system can capture," and whether they can "stablely complete their expression under high-paced and high-pressure conditions." Therefore, this step introduces a "task expression consistency modeling mechanism" customized for these needs, integrating the controllability and system adaptability of the previous stage to form an expression-driven scoring process with a clear fusion logic.

[0076] The first task in this step is to... Action sequences and The system assesses the consistency of expression in the skeleton actions. To this end, a consistency scoring module based on action semantic alignment is constructed. Specifically, the task action sequence... Converted into semantic vectors through a semantic action embedding model The model employs a three-layer stacked Transformer encoder structure. The input is a standard action sequence identifier (e.g., "turn around," "point"), and the output is an embedding vector for each action. Finally, average pooling is used to form the overall action semantic representation vector. Meanwhile, motion data Then, a sequence feature vector is extracted using a three-layer temporal convolutional encoding structure combined with a bidirectional GRU. , used to characterize the actions actually performed by the candidate.

[0077] Traditional semantic alignment often uses cosine similarity, but in this approach, we consider the emotional tension in the task. This significantly affects the rhythm and style of movements. Therefore, we introduced an emotion modulation term and an error weighting term, allowing the system to assign higher weights to high-intensity expressive tasks and penalize movements with significant structural deviations. The final expression consistency score... The definition is as follows:

[0078] ;

[0079] in: These are action semantic embedding vectors, derived from standard task templates. It is generated by a three-layer Transformer encoder, and its structure is as follows: Real number vectors; The actual action feature vector of the candidate is formed by the action tensor. The structure is generated after processing with three layers of 1D convolution (kernel=3, stride=1, channel=64 / 128 / 256) and one layer of bidirectional GRU (hidden=128). ; It is a three-dimensional emotion vector, derived from the emotion tags in the task template, representing the intensity of expression in the current task; It is a mood regulator (such as) The higher the intensity of the task expression, the stricter the system's requirements for the accuracy of the expression. It is the spatial offset difference vector of key parts, defined as the mean difference between the actual trajectory center of each key joint during execution and the reference trajectory center in the standard motion library, representing the spatial offset of the motion expression at the key joint; It is a spatial offset penalty factor (such as) ), used to control the penalty for spatial drift on the score; To prevent division by zero of small constants (such as...) ).

[0080] This formula integrates semantic similarity, task emotion requirements, and action space accuracy, creating a scoring model tailored for motion capture expression scenarios. It incorporates spatial error (…). The semantic scoring function is incorporated to emphasize that the expression must not only be "semantically consistent" but also "positionally matched" in key parts, so as to avoid situations where "the rhythm of the action is correct but the hand is raised crookedly" and is misjudged as accurate expression.

[0081] Ultimately, the system will express a consistency score. Controlled scoring in step three System compatibility score The data is then integrated, and a filtering and recommendation score is calculated. This score is used to generate the final candidate screening results. In the next stage, this score will be output to the screening list system, where it will be sorted and recommended based on client-defined weights.

[0082] The output of this step is:

[0083] Expression consistency score This measures the degree of consistency between the candidate's actual actions and the semantic expression of the task.

[0084] Comprehensive screening score This information is then used to rank the candidates and generate recommendation results.

[0085] In one or more embodiments, such as Figure 2 As shown, an AI-based intelligent selection system for motion capture actors is disclosed, the system comprising:

[0086] The instruction processing module is used to generate a structured action task template based on the input natural language action description text through a semantic understanding unit. The action task template includes an action sequence, an emotion vector, and a list of key body parts. The action sequence is obtained by matching the semantic embedding of the text with a standard action template library. The emotion vector is obtained by extracting emotion adverbs from the text and mapping them to a three-dimensional emotion space. The list of key body parts is obtained by identifying body part entities in the text and mapping them to standard joint identifiers.

[0087] The skeleton data acquisition module is used to guide actors to perform standardized movements based on the action task template through the data acquisition and processing unit, and to acquire their skeleton movement data; wherein, the execution rhythm is adjusted by the rhythm dimension in the emotion vector; during the acquisition process, image data is acquired through a multi-camera system, three-dimensional joint coordinates are extracted through the skeleton estimation model, and joint movement amplitude and stability residual index are calculated;

[0088] The capability assessment module is used to calculate the control capability score and system adaptability score based on the skeleton motion data and the list of key parts through the capability assessment unit. The control capability score is obtained by analyzing the sliding consistency error and trajectory offset within the motion segment. The system adaptability score is obtained by simulating occlusion and reconstructing the skeleton data, and evaluating the reconstruction error and its volatility.

[0089] The results generation module is used to calculate the expression consistency score through the semantic alignment unit. The score is based on the similarity between the semantic embedding of the task action sequence and the feature embedding of the actor's actual action, and is weighted and penalized by combining the emotion vector and the spatial offset of key parts. Finally, the expression consistency score, control ability score and system adaptability score are merged to generate a comprehensive score for actor selection.

[0090] It is worth noting that the specific workflow of the AI-based motion capture actor intelligent screening system provided in this embodiment of the invention is the same as that of the AI-based motion capture actor intelligent screening method described in the above embodiments, and will not be repeated here.

[0091] This invention also provides an AI-based intelligent screening device for motion capture actors, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiments of the AI-based intelligent screening method for motion capture actors, for example... Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above system embodiments.

[0092] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the AI-based motion capture actor intelligent screening device.

[0093] The AI-based intelligent selection device for motion capture actors can be a desktop computer, laptop, handheld computer, or cloud server, among other computing devices. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the AI-based intelligent selection device for motion capture actors may also include input / output devices, network access devices, and buses.

[0094] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASACs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the AI-based motion capture actor intelligent screening device, connecting all parts of the device via various interfaces and lines.

[0095] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the AI-based motion capture actor intelligent screening device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the operation of the air conditioner controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0096] The module integrated into the AI-based motion capture actor intelligent screening device, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0097] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0098] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An AI analysis-based motion capture actor intelligent screening method, characterized in that, The method comprises: Based on the input natural language action description text, a structured action task template is generated by a semantic understanding unit, the action task template comprising an action sequence, an emotion vector and a key part list; wherein the action sequence is obtained by matching the text semantic embedding with a standard action template library; the emotion vector is obtained by extracting emotion adverbs in the text and mapping to a three-dimensional emotion space; the key part list is obtained by identifying body part entities in the text and mapping to standard joint identifiers. Based on the action task template, a data acquisition and processing unit guides an actor to perform standardized actions and collects their skeletal action data; wherein the execution rhythm is adjusted by the rhythm dimension in the emotion vector; during the collection process, image data is obtained through a multi-camera system, three-dimensional joint coordinates are extracted through a skeletal estimation model, and joint motion amplitude and stability residual indicators are calculated. Based on the skeletal action data and the key part list, a capability evaluation unit calculates a control capability score and a system adaptability score; the control capability score is obtained by analyzing the sliding consistency error and trajectory deviation in the action segment; the system adaptability score is obtained by simulating occlusion and reconstructing skeletal data to evaluate reconstruction error and its volatility. An expression consistency score is calculated by a semantic alignment unit, the score being based on the similarity between the semantic embedding of the task action sequence and the feature embedding of the actor's actual action, and being weighted and penalized in combination with the emotion vector and the key part space offset; finally, the expression consistency score, the control capability score and the system adaptability score are fused to generate a comprehensive score for actor selection. 2.The method of claim 1, wherein, The semantic understanding unit uses a semantic encoding model based on the Transformer architecture to calculate the similarity between the semantic embedding vector of the input text and the embedding vector of each template in the standard action template library to match the action sequence. 3.The method of claim 1, wherein, The generation of the emotion vector includes: identifying emotion adverbs from the text semantic embedding and obtaining their attention scores, and projecting the emotion adverbs to a three-dimensional emotion space containing emotion intensity, emotion polarity and rhythm urgency through a pre-set mapping matrix. 4.The method of claim 1, wherein, The generation of the key part list includes: extracting body part entities from the text through entity recognition technology, and converting the body part entities to standard joint identifiers by querying a pre-defined mapping table. 5.The method of claim 1, wherein, The data acquisition and processing unit guides the actor to perform actions by displaying standard demonstration animations and text descriptions on a prompt screen, wherein the playback rate of the demonstration animation is linearly adjusted by the rhythm urgency dimension value in the emotion vector. 6.The method of claim 1, wherein, The joint motion amplitude is calculated as the average Euclidean distance of the joint relative to the initial position during action execution; The stability residual indicator is obtained by calculating the normalized variance of the joint position within the sliding time window and the window mean. 7.The method of claim 1, wherein the AI-based analysis is based on a machine learning algorithm. The control capability score is obtained by sliding consistency error analysis, specifically including: for each key joint in each action segment, the difference between the current frame joint position and the sliding window mean is calculated, and a trajectory deviation suppression term is introduced to evaluate the degree of trajectory drift. 8.The method of claim 1, wherein the AI-based analysis is based on a machine learning algorithm. The system adaptability score is obtained by adding simulated occlusion disturbance to the original skeleton data, reconstructing the occluded data using a time series recovery network, calculating reconstruction error and evaluating error volatility combined with an occlusion-sensitive regularization term. 9.The method of claim 1, wherein the method further comprises: The semantic alignment unit converts the action sequence into a semantic vector using a multi-layer Transformer encoder, converts the skeleton action data into a feature vector using a temporal convolution and a bidirectional gated recurrent unit, and obtains an expression consistency score by calculating the similarity between the two and combining an emotion adjustment factor and a spatial offset penalty term.

10. An AI analysis-based motion capture actor intelligent screening system, characterized in that, The system comprises: An instruction processing module for generating a structured action task template based on the input natural language action description text through a semantic understanding unit, the action task template including an action sequence, an emotion vector, and a key part list; wherein the action sequence is obtained by matching the text semantic embedding with a standard action template library; the emotion vector is obtained by extracting the emotional adverbs in the text and mapping them to a three-dimensional emotion space; and the key part list is obtained by identifying the body part entities in the text and mapping them to standard joint identifiers; A skeleton data acquisition module for guiding an actor to perform standardized actions based on the action task template through a data acquisition and processing unit and acquiring their skeleton action data; wherein the execution rhythm is adjusted by the rhythm dimension in the emotion vector; during the acquisition process, image data is obtained through a multi-camera system, three-dimensional joint coordinates are extracted through a skeleton estimation model, and joint motion amplitude and stability residual indicators are calculated; An ability evaluation module for calculating a control ability score and a system adaptability score based on the skeleton action data and the key part list through an ability evaluation unit; the control ability score is obtained by analyzing the sliding consistency error and trajectory offset within the action segment; and the system adaptability score is obtained by simulating occlusion and reconstructing the skeleton data to evaluate the reconstruction error and its volatility; A result generation module for calculating an expression consistency score through a semantic alignment unit, the score being based on the similarity between the semantic embedding of the task action sequence and the feature embedding of the actor's actual action, and being weighted and penalized in combination with the emotion vector and the key part spatial offset; finally, the expression consistency score, the control ability score, and the system adaptability score are fused to generate a comprehensive score for actor selection.

Images