An AI-based motion capture acquisition method and system

Abstract

The present invention provides a motion capture acquisition method and system based on artificial intelligence, including: multi-dimensionally collecting the physical sign representation information of the human body based on multi-dimensional sensors, and preprocessing the collected physical sign representation information; training and optimizing a deep learning model based on the preprocessed physical sign representation information, and performing multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model; fusing the real-time physical sign information collected by the multi-dimensional sensors, inputting the fusion result into the adapted deep learning model for analysis, and feeding back the analysis result to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result. The efficiency and reliability of motion capture are improved. At the same time, the operation of motion capture is simplified, and the effect of motion capture is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly relates to a motion capture acquisition method and system based on artificial intelligence. Background Art

[0002] At present, with the continuous development of technology, motion capture technology has become an indispensable part of industries such as film and television, games, etc. In recent years, motion capture technology based on artificial intelligence has gradually received attention due to its advantages such as high precision, real-time performance, and stability;

[0003] However, traditional motion capture methods mainly rely on electromagnetic induction, optical markers, inertial measurement, etc., but these methods have problems such as expensive equipment, complex operation, and strong limitations. At the same time, they cannot actively learn motion capture, resulting in a decline in motion capture efficiency and capture accuracy;

[0004] Therefore, in order to overcome the above defects, the present invention provides a motion capture acquisition method and system based on artificial intelligence. Summary of the Invention

[0005] The present invention provides a motion capture acquisition method and system based on artificial intelligence, which is used to multi-dimensionally collect the physical sign representation information of the human body through multi-dimensional sensors, ensuring the comprehensiveness and reliability of the collected physical sign representation information, so as to facilitate training and optimizing the deep learning model according to the collected physical sign representation information. Secondly, preprocess the collected physical sign representation information and train and optimize the deep learning model according to the preprocessing results, ensuring the accurate reliability of the deep learning. At the same time, perform multi-dimensional application scenario interaction adaptation on the deep learning model, facilitating to ensure that the deep learning model can be used in different application scenarios, thereby improving the reliability of capturing and collecting actions in different application scenarios. Finally, fuse the collected real-time physical sign information, facilitating the deep learning model to comprehensively and effectively capture the human body movements as a whole, and transmit the fusion result to the adapted deep learning model for analysis, ensuring the accurate reliability of the finally obtained action information, and feedback the obtained analysis result to the management terminal for management, improving the efficiency and reliability of motion capture, and at the same time, simplifying the operation of motion capture and ensuring the effect of motion capture.

[0006] The present invention provides a motion capture acquisition method based on artificial intelligence, including:

[0007] Step 1: Multi-dimensionally collect the physical sign representation information of the human body based on multi-dimensional sensors, and preprocess the collected physical sign representation information;

[0008] Step 2: Train and optimize the deep learning model based on the preprocessed physical sign representation information, and perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model;

[0009] Step 3: Fusion the real-time physical sign information collected by multi-dimensional sensors, input the fusion result into the adapted deep learning model for analysis, and feedback the analysis result to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result.

[0010] Preferably, for an AI-based motion capture acquisition method, in Step 1, multi-dimensionally collect the physical sign representation information of the human body based on multi-dimensional sensors, including:

[0011] Obtain the collection items of the physical sign representation information of the human body, and determine the corresponding collection parts on the human body for different collection items;

[0012] Extract the part features of the collection parts, mark specific tracking points on the multi-dimensional sensors based on the part features, and perform fixed-point tracking configuration on the multi-dimensional sensors based on the marking results;

[0013] Send a data collection instruction to the multi-dimensional sensors based on the fixed-point tracking configuration result, and control the multi-dimensional sensors to perform real-time tracking on the corresponding collection parts based on the data collection instruction to obtain the physical sign representation information of the human body.

[0014] Preferably, for an AI-based motion capture acquisition method, in Step 1, preprocess the collected physical sign representation information, including:

[0015] Obtain the collected physical sign representation information, perform dimension splitting on the physical sign representation information, and perform serialization processing on the dimension-split physical sign representation information to obtain a physical sign representation information sequence;

[0016] Compare the states of the physical sign representation information at adjacent moments in each dimension, and determine the change amount of the physical sign representation information in each dimension based on the state comparison;

[0017] Digitally encode different change amounts based on the encoding rule, perform logical association on the data encoding result based on the physical sign representation information sequence to obtain the digital signal corresponding to the physical sign representation information, and preprocess the physical sign representation information based on the digital signal.

[0018] Preferably, for an AI-based motion capture acquisition method, preprocess the physical sign representation information based on the digital signal, including:

[0019] Obtain the digital signal corresponding to the obtained physical sign representation information, determine the average value of the digital signals of the adjacent target points of each point, and replace the target value of the current point based on the average value;

[0020] Determine the filtering frequency band of the digital signal corresponding to the physical sign representation information based on the replacement result, configure the parameters of the filter based on the filtering frequency band, and perform filtering processing on the physical sign representation information based on the parameter configuration result;

[0021] Determine the value characteristics of the digital signal of the physical sign representation information in the time domain based on the filtering processing result, and use the value characteristics in the time domain as the first data feature;

[0022] At the same time, perform frequency domain conversion on the digital signal of the physical sign representation information based on a preset rule, determine the distribution and intensity characteristics of different frequency components of the digital signal of the physical sign representation information based on the frequency domain conversion result, and use the distribution and intensity characteristics of different frequency components as the second data feature;

[0023] Summarize the first data feature and the second data feature to obtain the target data feature of the physical sign representation information.

[0024] Preferably, in a motion capture acquisition method based on artificial intelligence, in step 2, train and optimize a deep learning model based on the preprocessed physical sign representation information, and perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model, including:

[0025] Obtain the preprocessed physical sign representation information, and perform data object division on the physical sign representation information based on a preset human body monitoring part to obtain a sub-action data set;

[0026] Extract the data features of the sub-action data set, and determine the action postures of the corresponding monitoring parts based on the data features;

[0027] Split the execution actions of the monitoring parts based on the action postures to obtain an action branch set and corresponding joint points, and quantify the action parameters of each action branch in the action branch set based on the splitting result according to the target values of the sub-action data set;

[0028] Obtain the execution parameter features of each action branch based on the quantization result. At the same time, obtain the position change features of the joint points under the corresponding execution parameter features, and associate the execution parameter features and the position change features of the joint points;

[0029] Statistically analyze the association results of the execution parameter features and the position change features of the joint points of the same monitoring part in different scenarios, and construct an action control group based on the statistical results;

[0030] Perform mapping analysis on the action control group to obtain the action feature representations and corresponding action feature parameters of each monitoring part in the time and space dimensions;

[0031] Iteratively train a deep learning model based on action feature representations and action feature parameters, and based on the model loss value of the deep learning model after each iterative training;

[0032] Determine the model optimization direction and parameters for the next iterative training based on the model loss value, and perform negative feedback iterative optimization on the deep learning model based on the model loss value, model optimization direction, and parameters to complete the training and optimization of the deep learning model.

[0033] Preferably, for an AI-based motion capture acquisition method, in step 2, perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model, including:

[0034] Obtain the trained and optimized deep learning model. At the same time, obtain the application scenario of the deep learning model, and determine the corresponding interaction requirements based on the execution items of the application scenario;

[0035] Determine the output and interaction methods of the deep learning model in different application scenarios based on the interaction requirements, and add an affiliated cooperation mechanism to the deep learning model based on the output and interaction methods;

[0036] Add an execution logic judgment mechanism to the affiliated cooperation mechanisms in different application scenarios based on the addition result, and perform parameter association between the execution logic judgment mechanism and the deep learning model;

[0037] Complete the multi-dimensional application scenario interaction adaptation of the deep learning model based on the parameter association result.

[0038] Preferably, for an AI-based motion capture acquisition method, in step 3, fuse the real-time physical sign information collected by multi-dimensional sensors, and input the fusion result into the adapted deep learning model for analysis, including:

[0039] Obtain the real-time physical sign information collected by multi-dimensional sensors, and extract the timestamp information corresponding to the real-time physical sign information;

[0040] Align the real-time physical sign information collected by multi-dimensional sensors in time based on the timestamp, and respectively extract the subject objects corresponding to the real-time physical sign information collected by each dimension sensor based on the time alignment result;

[0041] Determine the fusion logic for the real-time physical sign information based on the distribution position of the subject objects in the human body, determine the fusion data nodes between the real-time physical sign information based on the fusion logic, and at the same time, determine the weights of each subject object based on the action representation of each subject object;

[0042] Fuse the real-time vital sign information collected by the multi-dimensional sensors based on the fusion data nodes between the weights of each subject object and the real-time vital sign information, and input the fusion result into the adapted deep learning model for analysis to obtain the target action features corresponding to the real-time vital sign information.

[0043] Preferably, for an AI-based motion capture acquisition method, in step 3, feedback the analysis result to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result, including:

[0044] Obtain the target action features corresponding to the real-time vital sign information, and receive the interaction request sent by the management terminal;

[0045] Extract the scene identifier in the interaction request, and match it with the preset scene interaction mechanism based on the scene identifier to determine the target scene interaction mechanism corresponding to the current application scenario;

[0046] Feedback the target action features to the management terminal based on the target scene interaction mechanism, and create a temporary cache file in the management terminal;

[0047] Cache the received target action features in the temporary cache file.

[0048] The present invention provides an AI-based motion capture acquisition system, including:

[0049] A data acquisition and preprocessing module, used to perform multi-dimensional acquisition of the vital sign representation information of the human body based on multi-dimensional sensors, and preprocess the acquired vital sign representation information;

[0050] A model construction and adaptation module, used to train and optimize a deep learning model based on the preprocessed vital sign representation information, and perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model;

[0051] An analysis and interaction module, used to fuse the real-time vital sign information collected by the multi-dimensional sensors, input the fusion result into the adapted deep learning model for analysis, and feedback the analysis result to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result.

[0052] Preferably, for an AI-based motion capture acquisition system, the data acquisition and preprocessing module includes:

[0053] An acquisition part determination unit, used to obtain the acquisition items of the vital sign representation information of the human body, and determine the acquisition parts corresponding to different acquisition items on the human body;

[0054] A sensor configuration unit, used to extract the part features of the acquisition part, perform specific tracking point marking on the multi-dimensional sensors based on the part features, and perform fixed-point tracking configuration on the multi-dimensional sensors based on the marking result;

[0055] A data acquisition unit, configured to send a data acquisition instruction to a multi-dimensional sensor based on a fixed-point tracking configuration result, and control the multi-dimensional sensor to perform real-time tracking on a corresponding acquisition part based on the data acquisition instruction, so as to obtain the physical sign representation information of the human body.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] 1. The multi-dimensional acquisition of the physical sign representation information of the human body by the multi-dimensional sensor ensures the comprehensiveness and reliability of the acquired physical sign representation information, thereby facilitating the training and optimization of the deep learning model according to the acquired physical sign representation information. Secondly, the acquired physical sign representation information is preprocessed, and the deep learning model is trained and optimized according to the preprocessing result, ensuring the accurate reliability of the deep learning. At the same time, the multi-dimensional application scenario interaction adaptation of the deep learning model is carried out, which is convenient to ensure that the deep learning model can be applied in different application scenarios, thereby improving the reliability of capturing and collecting actions in different application scenarios. Finally, the acquired real-time physical sign information is fused, which is convenient for the deep learning model to comprehensively and effectively capture the human body movements as a whole, and the fusion result is transmitted to the adapted deep learning model for analysis, ensuring the accurate reliability of the finally obtained action information, and feeding back the obtained analysis result to the management terminal for management, improving the efficiency and reliability of action capture, and at the same time, simplifying the operation of action capture and ensuring the effect of action capture.

[0058] 2. By determining the acquisition items of the physical sign representation information of the human body, it is possible to accurately and effectively determine the acquisition part according to the acquisition items, thereby facilitating the fixed-point tracking configuration of the multi-dimensional sensor according to the determined acquisition part. Secondly, according to the configuration result, the data acquisition instruction sent in real time is received, and when the data acquisition instruction is received, the multi-dimensional sensor is timely controlled to perform real-time tracking on the corresponding acquisition part, so as to comprehensively and effectively acquire the physical sign representation information of the human body, providing data support for the comprehensive training of the deep learning model.

[0059] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structure specifically pointed out in this application document.

[0060] The technical solution of the present invention will be further described in detail below with reference to the drawings and embodiments. Description of the Drawings

[0061] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. They are used in conjunction with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the accompanying drawings:

[0062] Figure 1 It is a method for motion capture and acquisition based on artificial intelligence in an embodiment of the present invention;

[0063] Figure 2 It is a flowchart of step 1 in a method for motion capture and acquisition based on artificial intelligence in an embodiment of the present invention;

[0064] Figure 3 It is a structural diagram of a motion capture and acquisition system based on artificial intelligence in an embodiment of the present invention. Detailed implementation manners

[0065] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to explain and illustrate the present invention and are not used to limit the present invention.

[0066] Embodiment 1:

[0067] This embodiment provides a method for motion capture and acquisition based on artificial intelligence, as Figure 1 shown, including:

[0068] Step 1: Multidimensionally collect the physical sign representation information of the human body based on multi-dimensional sensors, and preprocess the collected physical sign representation information;

[0069] Step 2: Train and optimize a deep learning model based on the preprocessed physical sign representation information, and perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model;

[0070] Step 3: Fuse the real-time physical sign information collected by the multi-dimensional sensors, input the fusion result into the adapted deep learning model for analysis, and feedback the analysis result to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result.

[0071] In this embodiment, the multi-dimensional sensors refer to multiple different types of sensors used to collect different physical sign representation information of the human body.

[0072] In this embodiment, the physical sign representation information refers to action information, posture information, expression information, etc. that appear during the movement of the human body.

[0073] In this embodiment, multi-dimensional collection refers to the process of collecting the physical sign representation information of the human body through multi-dimensional sensors.

[0074] In this embodiment, preprocessing refers to processing such as denoising, filtering, and feature extraction on the collected physical sign information.

[0075] In this embodiment, multi-dimensional application scenario interaction adaptation refers to ensuring that the constructed deep learning model can be used in different application scenarios. For example, in the field of human-computer interaction, functions such as gesture recognition, facial expression recognition, and body posture recognition are realized. In the field of virtual reality, functions such as action simulation and animation production of virtual characters are realized.

[0076] In this embodiment, fusing the real-time physical sign information collected by multi-dimensional sensors refers to correlating the collected real-time physical sign information, aiming to conduct a more comprehensive analysis of the human motion trajectory.

[0077] The working principle and beneficial effects of the above technical solution are as follows: Through multi-dimensional sensors, multi-dimensional collection of the physical sign information of the human body is carried out, ensuring the comprehensiveness and reliability of the collected physical sign information, thus facilitating the training and optimization of the deep learning model based on the collected physical sign information. Secondly, preprocessing the collected physical sign information and training and optimizing the deep learning model according to the preprocessing results ensure the accuracy and reliability of deep learning. At the same time, multi-dimensional application scenario interaction adaptation is carried out on the deep learning model, facilitating ensuring that the deep learning model can be applied in different application scenarios, thereby improving the reliability of capturing and collecting actions in different application scenarios. Finally, fusing the collected real-time physical sign information facilitates the deep learning model to comprehensively and effectively capture the human body movements as a whole, and transmitting the fusion result to the adapted deep learning model for analysis, ensuring the accuracy and reliability of the finally obtained action information, and feeding back the obtained analysis result to the management terminal for management, improving the efficiency and reliability of action capture. At the same time, it also simplifies the operation of action capture and guarantees the effect of action capture.

[0078] Embodiment 2:

[0079] Based on Embodiment 1, this embodiment provides an action capture and collection method based on artificial intelligence. As Figure 2 shown, in step 1, multi-dimensional collection of the physical sign information of the human body is carried out based on multi-dimensional sensors, including:

[0080] Step 101: Obtain the collection items of the physical sign information of the human body and determine the corresponding collection parts on the human body for different collection items;

[0081] Step 102: Extract the part features of the collection part, mark specific tracking points on the multi-dimensional sensors based on the part features, and configure fixed-point tracking for the multi-dimensional sensors based on the marking results;

[0082] Step 103: Send a data acquisition instruction to the multi-dimensional sensor based on the fixed-point tracking configuration result, and control the multi-dimensional sensor to perform real-time tracking on the corresponding acquisition part according to the data acquisition instruction to obtain the physical sign representation information of the human body.

[0083] In this embodiment, the acquisition item refers to the type of physical sign representation information of the human body that needs to be acquired. For example, it can be the movement conditions of various tissues, the postures presented at different moments during operation, and the user's expressions, etc.

[0084] In this embodiment, the part feature refers to information such as the position corresponding to the acquisition part on the human body.

[0085] In this embodiment, the specific tracking point marking refers to limiting the tracking object of the corresponding sensor according to the part feature to ensure that each sensor can comprehensively and effectively monitor and collect the motion data of the corresponding acquisition part.

[0086] In this embodiment, the fixed-point tracking configuration refers to configuring the specific operating parameters corresponding to each sensor during the tracking process.

[0087] The working principle and beneficial effects of the above technical solution are: By determining the acquisition items of the physical sign representation information of the human body, it is possible to accurately and effectively determine the acquisition part according to the acquisition items, thereby facilitating the fixed-point tracking configuration of the multi-dimensional sensor according to the determined acquisition part. Secondly, according to the configuration result, the data acquisition instruction sent is received in real time, and when the data acquisition instruction is received, the multi-dimensional sensor is timely controlled to perform real-time tracking on the corresponding acquisition part, so as to comprehensively and effectively collect the physical sign representation information of the human body, providing data support for the comprehensive training of the deep learning model.

[0088] Embodiment 3:

[0089] Based on Embodiment 1, this embodiment provides an AI-based motion capture acquisition method. In Step 1, preprocess the acquired physical sign representation information, including:

[0090] Obtain the acquired physical sign representation information, perform dimension splitting on the physical sign representation information, and serialize the dimension-split physical sign representation information to obtain a physical sign representation information sequence;

[0091] Compare the states of the physical sign representation information at adjacent moments in each dimension, and determine the change amount of the physical sign representation information in each dimension based on the state comparison;

[0092] Digitally encode different variation amounts based on an encoding rule, and logically associate the data encoding result with a sequence of physical sign representation information to obtain a digital signal corresponding to the physical sign representation information, and preprocess the physical sign representation information based on the digital signal.

[0093] In this embodiment, dimension splitting refers to splitting the physical sign representation information according to the data type. For example, it can be split into action information, posture information, expression information, etc.

[0094] In this embodiment, serialization processing refers to arranging the split physical sign representation information, that is, determining the specific data content corresponding to different moments. Among them, the sequence of physical sign representation information is the result of serialization processing.

[0095] In this embodiment, the encoding rule is set in advance. For example, it can be binary encoding, etc.

[0096] The working principle and beneficial effects of the above technical solution are: by splitting and analyzing the physical sign representation information, accurately and effectively determine the variation amount of the physical sign representation information in each dimension, and digitally encode different variation amounts through the encoding rule to realize format conversion of the physical sign representation information and obtain the corresponding digital signal, thereby facilitating preprocessing of the physical sign representation information and providing convenience for training and optimizing the deep learning model.

[0097] Embodiment 4:

[0098] Based on Embodiment 3, this embodiment provides an AI-based motion capture acquisition method, which preprocesses the physical sign representation information based on the digital signal, including:

[0099] Obtain the digital signal corresponding to the physical sign representation information, and determine the average value of the digital signals of the adjacent target points of each point, and replace the target value of the current point based on the average value;

[0100] Determine the filtering frequency band of the digital signal corresponding to the physical sign representation information based on the replacement result, configure the parameters of the filter based on the filtering frequency band, and perform filtering processing on the physical sign representation information based on the parameter configuration result;

[0101] Determine the value-taking characteristics of the digital signal of the physical sign representation information in the time domain based on the filtering processing result, and use the value-taking characteristics in the time domain as the first data feature;

[0102] At the same time, perform frequency domain conversion on the digital signal of the physical sign representation information based on a preset rule, determine the distribution and intensity characteristics of different frequency components of the digital signal of the physical sign representation information based on the frequency domain conversion result, and use the distribution and intensity characteristics of different frequency components as the second data feature;

[0103] Summarize the first data feature and the second data feature to obtain the target data feature of the physical sign representation information.

[0104] In this embodiment, the target value refers to the specific value corresponding to the data at the current position point.

[0105] In this embodiment, the filtering frequency band is a specific frequency range corresponding to the data that needs to be filtered out in the digital signal, that is, the data within this frequency range needs to be filtered out.

[0106] In this embodiment, the first data feature refers to the value distribution of the physical sign representation information in the time domain.

[0107] In this embodiment, the preset rule is set in advance and is used to perform format conversion on the physical sign representation information, that is, to convert the physical sign representation information into a frequency-domain signal.

[0108] In this embodiment, the distribution and intensity features refer to the distribution of data signals with different frequencies in the physical sign representation information in the frequency domain and the corresponding value change conditions.

[0109] The working principle and beneficial effects of the above technical solution are as follows: By cleaning the digital signal corresponding to the physical sign representation information, the accuracy and reliability of the obtained physical sign representation information are ensured. Secondly, by processing and feature extraction of the cleaned physical sign representation information, effective extraction of the target data feature of the physical sign representation information is achieved.

[0110] Embodiment 5:

[0111] Based on Embodiment 1, this embodiment provides an AI-based motion capture acquisition method. In step 2, based on the preprocessed physical sign representation information, train and optimize the deep learning model, and perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model, including:

[0112] Obtain the preprocessed physical sign representation information, and based on the preset human body monitoring parts, divide the physical sign representation information into data objects to obtain a sub-action data set;

[0113] Extract the data features of the sub-action data set, and determine the action postures of the corresponding monitoring parts based on the data features;

[0114] Based on the action postures, split the executed actions of the monitoring parts to obtain an action branch set and the corresponding joint points, and quantify the action parameters of each action branch in the action branch set according to the target values of the sub-action data set based on the splitting results;

[0115] Based on the quantization results, obtain the execution parameter characteristics of each action branch. At the same time, acquire the position change characteristics of the joint points under the corresponding execution parameter characteristics, and associate the execution parameter characteristics with the position change characteristics of the joint points;

[0116] Statistically analyze the association results of the execution parameter characteristics and the position change characteristics of the joint points for the same monitored body part in different scenarios, and construct an action control group based on the statistical results;

[0117] Conduct mapping analysis on the action control group to obtain the action feature representations and corresponding action feature parameters of each monitored body part in the time and space dimensions;

[0118] Iteratively train the deep learning model based on the action feature representations and action feature parameters, and based on the model loss value of the deep learning model after each iterative training;

[0119] Determine the model optimization direction and parameters for the next iterative training based on the model loss value, and perform negative feedback iterative optimization on the deep learning model based on the model loss value, model optimization direction, and parameters to complete the training and optimization of the deep learning model.

[0120] In this embodiment, the preset human body monitoring parts are known in advance. For example, they can be the legs, head, arms, etc., with the purpose of effectively splitting the physical sign representation information.

[0121] In this embodiment, the sub-action dataset refers to splitting the physical sign representation information according to the preset human body monitoring parts (that is, splitting the physical sign representation information into specific data corresponding to each human body monitoring part according to the preset human body monitoring parts).

[0122] In this embodiment, the data characteristics refer to the value range and corresponding data structure of the sub-action dataset.

[0123] In this embodiment, splitting the execution action of the monitoring part means splitting the execution action of the monitoring part into corresponding components. For example, when the leg is walking, the execution action of the leg can be split into a lifting part, a stretching part, a falling part, etc. Among them, the action branch set is the splitting result.

[0124] In this embodiment, the joint point refers to the associated position between different actions after splitting the execution action of the monitoring part.

[0125] In this embodiment, the execution parameter characteristics refer to the specific action movement conditions corresponding to each action branch. For example, it can be the amplitude size of the action, etc.

[0126] In this embodiment, the position change characteristics refer to the movement trajectory change of the joint point during the movement process.

[0127] In this embodiment, the action feature representation refers to all the motion conditions presented by each monitored part in different scenarios.

[0128] In this embodiment, the action feature parameter refers to the specific action magnitude corresponding to each monitored part during the motion process.

[0129] In this embodiment, the model loss value refers to the performance loss of the deep learning model after each iterative training. The smaller the value of the model loss value, the better the performance of the model loss value.

[0130] In this embodiment, the model optimization direction and parameters for the next iterative training refer to the parameters that need to optimize the training direction and the training parameter values in this training direction for the next training based on the previous training of the deep learning model.

[0131] The working principle and beneficial effects of the above technical solution are as follows: By splitting the sign feature information, an accurate and effective determination of the sub-action data set corresponding to each monitored part is achieved. Secondly, by analyzing the sub-action data set of each monitored part, an effective determination of the execution parameter characteristics and the position change characteristics of the joint points of each monitored part in different scenarios is realized, and then the locking of the action feature representation and the corresponding action feature parameters is achieved, thereby providing a training basis for training the deep learning model. Finally, the deep learning model is iteratively trained with the obtained action feature representation and action feature parameters, ensuring the training effect of the deep learning model, and thus the reliability of action capture and collection in different application scenarios.

[0132] Embodiment 6:

[0133] Based on Embodiment 1, this embodiment provides an action capture and collection method based on artificial intelligence. In step 2, a multi-dimensional application scenario interaction adaptation is performed on the trained and optimized deep learning model, including:

[0134] Obtain the trained and optimized deep learning model. At the same time, obtain the application scenario of the deep learning model, and determine the corresponding interaction requirements based on the execution items of the application scenario;

[0135] Determine the output and interaction methods of the deep learning model in different application scenarios based on the interaction requirements, and add an affiliated cooperation mechanism to the deep learning model based on the output and interaction methods;

[0136] Add an execution logic judgment mechanism to the affiliated cooperation mechanism in different application scenarios based on the addition result, and perform parameter association between the execution logic judgment mechanism and the deep learning model;

[0137] Complete the multi-dimensional application scenario interaction adaptation of the deep learning model based on the parameter association result.

[0138] In this embodiment, the execution item refers to the type of motion capture that needs to be performed by the deep learning model.

[0139] In this embodiment, the interaction requirement refers to the ultimate purpose to be achieved.

[0140] In this embodiment, the affiliated cooperation mechanism refers to the cooperation scheme configured to achieve the purpose of scene interaction during the application process of the deep learning model. The purpose is to ensure the effective operation of the deep learning model and facilitate the coordination of the deep learning model in a timely manner according to the requirements of different application scenarios.

[0141] In this embodiment, the execution logic judgment mechanism is used to judge whether the deep learning model meets the corresponding execution conditions, so as to facilitate the adaptation of scene interaction through the deep learning model in a timely manner when the conditions are met.

[0142] The working principle and beneficial effects of the above technical solution are as follows: By determining the application scenario of the deep learning model, the interaction requirements of different application scenarios are locked, so as to facilitate the adaptation of scene interaction of the deep learning model according to the interaction requirements, and ensure that the deep learning model can be applied in different application scenarios.

[0143] Embodiment 7:

[0144] Based on Embodiment 1, this embodiment provides a motion capture acquisition method based on artificial intelligence. In step 3, the real-time physical sign information collected by the multi-dimensional sensor is fused, and the fusion result is input into the adapted deep learning model for analysis, including:

[0145] Obtain the real-time physical sign information collected by the multi-dimensional sensor, and extract the timestamp information corresponding to the real-time physical sign information;

[0146] Align the real-time physical sign information collected by the multi-dimensional sensor based on the timestamp, and extract the subject objects corresponding to the real-time physical sign information collected by each dimension sensor based on the time alignment result;

[0147] Determine the fusion logic of the real-time physical sign information based on the distribution position of the subject object in the human body, determine the fusion data nodes between the real-time physical sign information based on the fusion logic, and at the same time, determine the weight of each subject object based on the action representation of each subject object;

[0148] Fuse the real-time physical sign information collected by the multi-dimensional sensor based on the weights of each subject object and the fusion data nodes between the real-time physical sign information, and input the fusion result into the adapted deep learning model for analysis to obtain the target action features corresponding to the real-time physical sign information.

[0149] In this embodiment, the main object refers to the specific monitoring part corresponding to the real-time vital sign information.

[0150] In this embodiment, the fusion logic is used to represent the logic of fusing real-time vital sign information between different main objects when performing real-time vital sign information fusion.

[0151] In this embodiment, the fusion data node refers to the data fusion position point between different real-time vital sign data when performing data fusion on different real-time vital sign information.

[0152] The working principle and beneficial effects of the above technical solution are as follows: By analyzing the collected real-time vital sign information, accurate and effective fusion of different real-time vital sign information for different monitors is achieved, and the fusion result is input into the deep learning model for training, which provides guarantee for determining the target action characteristics and ensures the reliability and accuracy of action capture.

[0153] Embodiment 8:

[0154] Based on Embodiment 1, this embodiment provides an action capture acquisition method based on artificial intelligence. In step 3, the analysis result is fed back to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result, including:

[0155] Obtain the target action characteristics corresponding to the real-time vital sign information and receive the interaction request sent by the management terminal;

[0156] Extract the scene identifier in the interaction request and match it with the preset scene interaction mechanism to determine the target scene interaction mechanism corresponding to the current application scene;

[0157] Feed back the target action characteristics to the management terminal based on the target scene interaction mechanism and create a temporary cache file in the management terminal;

[0158] Cache the received target action characteristics in the temporary cache file.

[0159] In this embodiment, the scene identifier is a marking symbol used to distinguish different scenes.

[0160] In this embodiment, the target scene interaction mechanism refers to the scene interaction mechanism applicable to the current application scene and is one of the preset scene interaction mechanisms.

[0161] The working principle and beneficial effects of the above technical solution are as follows: By analyzing the interaction request, the scene identifier corresponding to the interaction request is locked, so as to determine the corresponding target scene interaction mechanism according to the scene identifier, feed back the obtained target action characteristics to the management terminal for caching according to the target scene interaction mechanism, and effectively manage the captured target action characteristics.

[0162] Example 9:

[0163] This embodiment provides an AI-based motion capture acquisition system, as Figure 3 shown, including:

[0164] A data acquisition and preprocessing module, configured to perform multi-dimensional acquisition of the physical sign information of the human body based on multi-dimensional sensors, and preprocess the acquired physical sign information;

[0165] A model construction and adaptation module, configured to train and optimize a deep learning model based on the preprocessed physical sign information, and perform multi-dimensional application scenario interaction adaptation on the trained and optimized deep learning model;

[0166] An analysis and interaction module, configured to fuse the real-time physical sign information collected by the multi-dimensional sensors, input the fusion result into the adapted deep learning model for analysis, and feedback the analysis result to the corresponding management terminal based on the multi-dimensional application scenario interaction adaptation result.

[0167] The working principle and beneficial effects of the above technical solution are as follows: The multi-dimensional acquisition of the physical sign information of the human body by the multi-dimensional sensors ensures the comprehensiveness and reliability of the acquired physical sign information, thus facilitating the training and optimization of the deep learning model based on the acquired physical sign information. Secondly, the acquired physical sign information is preprocessed, and the deep learning model is trained and optimized according to the preprocessing result, ensuring the accurate reliability of deep learning. At the same time, multi-dimensional application scenario interaction adaptation is performed on the deep learning model, facilitating the ensuring that the deep learning model can be applied in different application scenarios, thereby improving the reliability of motion capture acquisition in different application scenarios. Finally, the collected real-time physical sign information is fused, facilitating the deep learning model to comprehensively and effectively capture the human body movements as a whole, and transmitting the fusion result to the adapted deep learning model for analysis, ensuring the accurate reliability of the finally obtained motion information, and feeding back the obtained analysis result to the management terminal for management, improving the efficiency and reliability of motion capture, and at the same time, simplifying the operation of motion capture and ensuring the effect of motion capture.

[0168] Example 10:

[0169] Based on Example 9, this embodiment provides an AI-based motion capture acquisition system. The data acquisition and preprocessing module includes:

[0170] An acquisition part determination unit, configured to obtain the acquisition items of the physical sign information of the human body, and determine the corresponding acquisition parts on the human body for different acquisition items;

[0171] A sensor configuration unit, configured to extract the feature of the acquisition site, mark specific tracking points on the multi-dimensional sensors based on the site feature, and perform fixed-point tracking configuration on the multi-dimensional sensors based on the marking result;

[0172] A data acquisition unit, configured to issue a data acquisition instruction to the multi-dimensional sensors based on the fixed-point tracking configuration result, and control the multi-dimensional sensors to perform real-time tracking on the corresponding acquisition site based on the data acquisition instruction, so as to obtain the physical sign representation information of the human body.

[0173] The working principle and beneficial effects of the above technical solution are as follows: By determining the acquisition items of the physical sign representation information of the human body, accurate and effective determination of the acquisition site can be achieved according to the acquisition items, so as to facilitate the fixed-point tracking configuration of the multi-dimensional sensors according to the determined acquisition site. Secondly, according to the configuration result, the issued data acquisition instruction is received in real time, so that when the data acquisition instruction is received, the multi-dimensional sensors are timely controlled to perform real-time tracking on the corresponding acquisition site, realizing comprehensive and effective acquisition of the physical sign representation information of the human body, and providing data support for the comprehensive training of the deep learning model.

[0174] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A motion capture acquisition method based on artificial intelligence, characterized in that: include: Step 1: Collect the human body's vital sign information in multiple dimensions based on the multi-dimensional sensor, and pre-process the collected vital sign information; Step 2: Train and optimize the deep learning model based on the preprocessed vital sign representation information, and perform multi-dimensional application scenario interactive adaptation on the trained and optimized deep learning model; Step 3: Fuse the real-time vital sign information collected by multi-dimensional sensors, input the fusion results into the adapted deep learning model for analysis, and feed back the analysis results to the corresponding management terminal based on the interactive adaptation results of the multi-dimensional application scenarios; Among them, in step 2, the deep learning model is trained and optimized based on the preprocessed vital sign representation information, and the trained and optimized deep learning model is interactively adapted to multi-dimensional application scenarios, including: Obtaining preprocessed vital sign representation information, and dividing the vital sign representation information into data objects based on preset human body monitoring parts to obtain a sub-action data set; Extracting data features of the sub-action data set, and determining the action posture of the corresponding monitoring part based on the data features; The execution action of the monitoring part is split based on the action posture to obtain the action branch set and the corresponding joint points, and the action parameters of each action branch in the action branch set are quantified based on the split result and the target value of the sub-action data set; Based on the quantization results, the execution parameter features of each action branch are obtained. At the same time, the position change features of the joint points under the corresponding execution parameter features are obtained, and the execution parameter features and the position change features of the joint points are associated; The correlation results of the execution parameter characteristics and the position change characteristics of the joint points of the same monitoring part in different scenarios are statistically analyzed, and an action control group is constructed based on the statistical results; Mapping analysis is performed on the action control group to obtain the action feature representation of each monitored part in time and space dimensions and the corresponding action feature parameters; Iteratively train the deep learning model based on the action feature representation and action feature parameters, and based on the model loss value of the deep learning model after each iterative training; Based on the model loss value, the model optimization direction and parameters for the next iterative training are determined, and based on the model loss value, model optimization direction and parameters, the deep learning model is negatively feedback iteratively optimized to complete the training and optimization of the deep learning model.

2. The method for motion capture based on artificial intelligence according to claim 1, characterized in that: In step 1, multi-dimensional collection of human body vital sign information is performed based on multi-dimensional sensors, including: Acquire the collection items of information representing physical signs of the human body, and determine the collection parts on the human body corresponding to different collection items; Extracting the part features of the collected part, marking specific tracking points of the multi-dimensional sensor based on the part features, and performing fixed-point tracking configuration of the multi-dimensional sensor based on the marking results; Based on the fixed-point tracking configuration result, a data collection instruction is sent to the multi-dimensional sensor, and based on the data collection instruction, the multi-dimensional sensor is controlled to track the corresponding collection part in real time to obtain the vital sign representation information of the human body.

3. The method for motion capture based on artificial intelligence according to claim 1, characterized in that: In step 1, the collected vital sign representation information is preprocessed, including: Acquire the collected physical sign representation information, perform dimension splitting on the physical sign representation information, and perform serialization processing on the physical sign representation information after dimension splitting to obtain a physical sign representation information sequence; Compare the states of the vital sign representation information at adjacent moments in each dimension, and determine the change amount of the vital sign representation information in each dimension based on the state comparison; Different changes are digitally encoded based on encoding rules, and the data encoding results are logically associated based on the vital sign characterization information sequence to obtain a digital signal corresponding to the vital sign characterization information, and the vital sign characterization information is preprocessed based on the digital signal.

4. The method for motion capture based on artificial intelligence according to claim 3, characterized in that: Preprocessing of vital sign representation information based on digital signals includes: Obtaining the digital signal corresponding to the obtained vital sign representation information, and determining the average value of the digital signals of the adjacent target points of each point, and replacing the target value of the current point based on the average value; Determine the filtering frequency band of the digital signal corresponding to the vital sign characterization information based on the replacement result, configure the filter parameters based on the filtering frequency band, and perform filtering processing on the vital sign characterization information based on the parameter configuration result; Determine the value characteristics of the digital signal of the vital sign representation information in the time domain based on the filtering processing result, and use the value characteristics of the time domain as the first data feature; At the same time, the digital signal of the vital sign representation information is converted into a frequency domain based on a preset rule, and the distribution and intensity characteristics of different frequency components of the digital signal of the vital sign representation information are determined based on the frequency domain conversion result, and the distribution and intensity characteristics of the different frequency components are used as the second data features; The first data feature and the second data feature are aggregated to obtain the target data feature of the vital sign representation information.

5. The method for motion capture based on artificial intelligence according to claim 1, characterized in that: In step 2, the trained and optimized deep learning model is interactively adapted to multi-dimensional application scenarios, including: Obtain the trained and optimized deep learning model, and at the same time, obtain the application scenario of the deep learning model, and determine the corresponding interaction requirements based on the execution items of the application scenario; Determine the output and interaction methods of deep learning models in different application scenarios based on interaction requirements, and add auxiliary coordination mechanisms to the deep learning models based on the output and interaction methods; Based on the added results, add an execution logic judgment mechanism to the subsidiary coordination mechanisms of different application scenarios, and associate the execution logic judgment mechanism with the deep learning model through parameters; Based on the parameter association results, interactive adaptation of the deep learning model to multi-dimensional application scenarios is completed.

6. The method for motion capture based on artificial intelligence according to claim 1, characterized in that: In step 3, the real-time vital sign information collected by the multi-dimensional sensors is fused, and the fusion results are input into the adapted deep learning model for analysis, including: Obtain real-time vital sign information collected by multi-dimensional sensors, and extract timestamp information corresponding to the real-time vital sign information; The real-time vital sign information collected by the multi-dimensional sensors is time-aligned based on the timestamp, and the subject objects corresponding to the real-time vital sign information collected by the sensors of each dimension are respectively extracted based on the time alignment result; Determine the fusion logic of the real-time vital sign information based on the distribution position of the subject object in the human body, and determine the fusion data nodes between the real-time vital sign information based on the fusion logic. At the same time, determine the weight of each subject object based on the action representation of each subject object; The real-time vital sign information collected by multi-dimensional sensors is fused based on the weights of each subject object and the fusion data nodes between the real-time vital sign information, and the fusion results are input into the adapted deep learning model for analysis to obtain the target action features corresponding to the real-time vital sign information.

7. The method for motion capture based on artificial intelligence according to claim 1, characterized in that: In step 3, the analysis results are fed back to the corresponding management terminal based on the multi-dimensional application scenario interactive adaptation results, including: Obtain target action features corresponding to real-time vital sign information, and receive interaction requests sent by the management terminal; Extract the scene identifier in the interaction request, and match the scene identifier with the preset scene interaction mechanism to determine the target scene interaction mechanism corresponding to the current application scene; Feedback the target action features to the management terminal based on the target scene interaction mechanism, and create a temporary cache file in the management terminal; The received target action features are cached in a temporary cache file.

8. A motion capture system based on artificial intelligence, characterized in that: include: A data acquisition and preprocessing module is used to collect multi-dimensional information on the body's vital signs based on multi-dimensional sensors, and to preprocess the collected vital signs; Model building and adaptation module, used to train and optimize the deep learning model based on the preprocessed vital sign representation information, and to interactively adapt the trained and optimized deep learning model to multi-dimensional application scenarios; The analysis and interaction module is used to fuse the real-time vital sign information collected by multi-dimensional sensors, input the fusion results into the adapted deep learning model for analysis, and feed back the analysis results to the corresponding management terminal based on the multi-dimensional application scenario interactive adaptation results; Among them, the model construction and adaptation module includes: Obtaining preprocessed vital sign representation information, and dividing the vital sign representation information into data objects based on preset human body monitoring parts to obtain a sub-action data set; Extracting data features of the sub-action data set, and determining the action posture of the corresponding monitoring part based on the data features; The execution action of the monitoring part is split based on the action posture to obtain the action branch set and the corresponding joint points, and the action parameters of each action branch in the action branch set are quantified based on the split result and the target value of the sub-action data set; Based on the quantization results, the execution parameter features of each action branch are obtained. At the same time, the position change features of the joint points under the corresponding execution parameter features are obtained, and the execution parameter features and the position change features of the joint points are associated; The correlation results of the execution parameter characteristics and the position change characteristics of the joint points of the same monitoring part in different scenarios are statistically analyzed, and an action control group is constructed based on the statistical results; Mapping analysis is performed on the action control group to obtain the action feature representation of each monitored part in time and space dimensions and the corresponding action feature parameters; Iteratively train the deep learning model based on the action feature representation and action feature parameters, and based on the model loss value of the deep learning model after each iterative training; Based on the model loss value, the model optimization direction and parameters for the next iterative training are determined, and based on the model loss value, model optimization direction and parameters, the deep learning model is negatively feedback iteratively optimized to complete the training and optimization of the deep learning model.

9. The artificial intelligence-based motion capture system according to claim 8, characterized in that: Data acquisition and preprocessing modules, including: A collection part determination unit, used to obtain collection items of information representing physical signs of a human body, and determine collection parts corresponding to different collection items on the human body; A sensor configuration unit, used for extracting the part features of the collection part, marking specific tracking points of the multi-dimensional sensor based on the part features, and performing fixed-point tracking configuration of the multi-dimensional sensor based on the marking results; The data acquisition unit is used to send data acquisition instructions to the multi-dimensional sensor based on the fixed-point tracking configuration result, and control the multi-dimensional sensor to track the corresponding acquisition part in real time based on the data acquisition instruction to obtain the vital sign representation information of the human body.

Images