An action-aware motion strategy optimization generation method and system

Abstract

The application relates to the technical field of action perception, and discloses a motion strategy optimization generation method and system of action perception, which comprises the following steps: extracting action time sequence feature vectors and physiological load feature vectors of multi-source action perception data; using an action perception evaluation model to calculate action deviation scores and action stability indexes between the action time sequence feature vectors and preset motion mode templates in each set of structured motion state data, as action perception semantic vectors; coupling and mapping the action perception semantic vectors and the physiological load feature vectors to obtain effective motion stimulation intensity vectors; combining individual health conditions, physical test results and individual fatigue attenuation coefficients to construct a multi-objective optimization function to generate a motion strategy. The application realizes collaborative quantitative modeling of action execution quality and physiological response intensity, performs multi-objective optimization solving under individual health constraints, and generates an individualized dynamic motion strategy which takes into account safety and training effect.

Description

Technical Field

[0001] This invention relates to the field of motion perception, and more particularly to a method and system for optimizing and generating motion strategies based on motion perception. Background Technology

[0002] With the rapid development of wearable devices, mobile health platforms, and artificial intelligence technologies, exercise prescription generation has gradually shifted from a traditional model relying on manual assessment and static planning to a personalized intervention system driven by multi-source data perception and intelligent algorithms. In areas such as chronic disease management, postoperative rehabilitation, elderly health promotion, and mass fitness, exercise prescriptions are considered a crucial technological carrier connecting medical and health guidance with daily life behaviors. However, existing research and applications are largely based on idealized assumptions: that users can fully execute the training plan according to the prescription, and that the data collected by wearable devices can continuously and accurately reflect the actual exercise load and physiological response. In real-world scenarios, user behavior is influenced by multiple factors such as time constraints, environmental conditions, psychological state, and physical fatigue, commonly resulting in training interruptions, distorted movements, reduced intensity, or improper sensor wearing, leading to significant discrepancies between the collected data and the actual training stimuli.

[0003] Existing research, such as CN114822763A, proposes a data-driven personalized exercise prescription recommendation method. This method preprocesses raw exercise data, performs feature selection and cluster analysis to construct a user profile containing basic attributes, physical test data, and exercise ability tags. Recommendations are then made based on an exercise prescription rule base, and the method incorporates subjective fatigue level (RPE) for effect evaluation and fine-tuning. While this method achieves personalized recommendations based on data tags and rule matching to some extent, improving the automation level of exercise prescription generation, its core remains primarily static attribute analysis and rule adjustment. This makes it difficult to achieve a globally optimal balance under individual differences, affecting the scientific validity and dynamic adaptability of the exercise prescription.

[0004] To address this issue, this invention proposes a motion-aware method for optimizing and generating motion strategies. By incorporating motion awareness during data acquisition, the method expands from relying solely on feedback from physiological and exercise data to understanding motion quality, execution integrity, and training semantics. This helps to build a more robust and secure adaptive adjustment mechanism for motion strategies. Furthermore, by combining individual user health information, the method enhances the system's ability to perceive the user's actual training state, providing a more reliable decision-making basis for personalized health interventions. Summary of the Invention

[0005] This invention provides a method and system for optimizing and generating motion strategies based on motion perception. Since existing motion strategies are mostly based on static physical examination data or human experience, they are difficult to dynamically reflect an individual's actual motion execution state. Steps S1-S2 utilize a motion perception assessment model to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each set of structured motion state data, forming a quality representation result at the motion perception semantic level. This enables a quantitative assessment of the accuracy, standardization, and stability of actual motion execution, transforming the previously difficult-to-quantify motion execution quality into a calculable and comparable numerical expression. Step S3 couples and maps the motion perception semantic vector with the physiological load feature vector, solving the problem of identifying motion that meets the standard but the load is insufficient, or load that meets the standard but the motion is incorrect, obtaining an effective motion stimulus intensity vector of effective standard motion intensity. Step S4 quantifies the individual fatigue decay coefficient based on individual differences, and then constructs a multi-objective optimization function to generate an individual's motion strategy. This solves the problem of traditional motion strategies lacking individual difference modeling and dynamic intensity optimization, achieving a motion strategy generation strategy that is individualized, safe, and synergistically optimized in terms of training effect.

[0006] To achieve the above objectives, the present invention provides a motion strategy optimization and generation method based on action perception, comprising the following steps: S1: Collect multi-source motion perception data based on a preset motion mode template, extract the motion timing feature vector and physiological load feature vector from the multi-source motion perception data, and use the preset motion mode template, motion timing feature vector and physiological load feature vector as a set of structured motion state data to form a structured motion state dataset. S2: Based on the structured motion state dataset, use the motion perception evaluation model to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each group of structured motion state data, and use the motion deviation score and motion stability index as the motion perception semantic vector of the structured motion state data. S3: Couple and map the motion perception semantic vector with the physiological load feature vector to obtain the effective motion stimulus intensity vector for each set of structured motion state data; S4: Based on the effective motion stimulus intensity vector of each set of structured motion state data, combined with individual health status, physical fitness test results and individual fatigue decay coefficient, construct a multi-objective optimization function, optimize the multi-objective optimization function, select multiple preset exercise mode templates suitable for individual health status, and use the selected preset exercise mode templates to form an exercise strategy.

[0007] As a further improvement of the present invention: Further, in step S1, multi-source motion perception data is collected based on a preset motion pattern template, including: S11: Construct multiple sets of preset exercise mode templates based on the standard training action library. Each preset exercise mode template consists of description text for four exercise stages, namely, preparation stage, squatting stage, lowest point stage, and recovery stage. S12: Deploy multimodal sensing devices on standard demonstration personnel, including inertial measurement units, plantar pressure sensors, heart rate sensors, blood oxygen sensors, and surface electromyography sensors. S13: The standard demonstrator completes the specified action under the requirements of the preset movement mode template, and uses the multimodal sensing device to collect multi-source motion perception data to obtain the multi-source motion perception data corresponding to the preset movement mode template. The multi-source motion perception data is then segmented according to the movement stage of the preset movement mode template to obtain multi-source motion sub-perception data for each movement stage. The multi-source motion sub-perception data includes joint angle sequence, plantar pressure sequence, heart rate sequence, blood oxygen saturation sequence, and electromyographic signal.

[0008] Furthermore, step S1, which extracts the action temporal features and physiological load features from the multi-source action sub-sensing data, also includes: S14: Perform first-order difference on the joint angle sequence to obtain a first-order difference sequence of joint angles, and calculate the sequence mean of the first-order difference sequence of joint angles as a joint angular velocity feature. S15: Perform second-order difference on the joint angle sequence to obtain a second-order difference sequence of joint angles, and calculate the sequence mean of the second-order difference sequence of joint angles as a feature of joint angular acceleration. S16: Based on the joint angle sequence, calculate the fluctuation degree of joint angle and the motion coordination coupling index; normalize the plantar pressure sequence and calculate the standard deviation of the normalized plantar pressure sequence as the plantar pressure fluctuation coefficient; splice the joint angular velocity feature, joint angular acceleration feature, fluctuation degree of joint angle, motion coordination coupling index and plantar pressure fluctuation coefficient to obtain the motion temporal feature of the multi-source motion sub-sensing data; S17: Obtain the resting heart rate and maximum heart rate of the standard demonstration personnel, and calculate the heart rate load index of the heart rate sequence based on the resting heart rate and maximum heart rate; S18: Normalize the blood oxygen saturation sequence and electromyography signal, and calculate the standard deviation of the blood oxygen saturation sequence and the root mean square value of the electromyography signal after normalization, which are respectively used as the blood oxygen fluctuation coefficient and the electromyography activation intensity. S19: The heart rate load index, blood oxygen fluctuation coefficient and electromyographic activation intensity are spliced ​​together as the physiological load characteristics of the multi-source sub-motion perception data.

[0009] Furthermore, based on the action temporal features and physiological load features of the multi-source action sub-sensing data, a structured motion state dataset is constructed. The structured motion state dataset consists of several sets of structured motion state data, wherein each set of structured motion state data includes a preset motion mode template, an action temporal feature vector, and a physiological load feature vector. The action temporal feature vector is obtained by concatenating the action temporal features of the multi-source action sub-sensing data for each motion stage, and the physiological load feature vector is obtained by concatenating the physiological load features of the multi-source action sub-sensing data for each motion stage.

[0010] Further, in step S2, the motion perception evaluation model is used to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each group of structured motion state data, including: S21: The motion perception evaluation model includes an input layer, a template semantic embedding layer, a motion deviation scoring calculation layer, and a motion stability calculation layer. The input layer receives the motion temporal feature vector from the structured motion state data and a preset motion mode template. It also performs word segmentation and word vector conversion on the descriptive text in the preset motion mode template to obtain the word vector sequence of the preset motion mode template in the preparation stage, squatting stage, lowest point stage, and recovery stage. S22: The template semantic embedding layer receives word vector sequences of a preset motion mode template in the preparation stage, squatting stage, lowest point stage, and recovery stage, and extracts semantic encoding features of the word vector sequences in the preparation stage, squatting stage, lowest point stage, and recovery stage respectively. S23: The motion deviation scoring calculation layer calculates the stage deviation between the semantic coding features and the motion temporal features of any motion stage, and performs weighted and normalized processing on the stage deviation to obtain the motion deviation score between the motion temporal feature vector in the structured motion state data and the preset motion mode template. Specifically, the formula for calculating the motion deviation score between the motion timing feature vector and the preset motion pattern template is as follows: ; in, This represents the motion deviation score between the motion temporal feature vector in the m-th group of structured motion state data and the preset motion pattern template, where M represents the number of groups of structured motion state data. This represents the scaling factor. This represents the weighting coefficient for the e-th motion phase. The four phases of the exercise are, in order, preparation, squatting, lowest point, and recovery. This represents the temporal characteristics of the action in the e-th motion phase within the m-th group of structured motion state data. This represents the semantic encoding features of the e-th motion stage in the m-th group of structured motion state data. Represents the L2 norm. Represents an exponential function with the natural constant as its base; S24: The motion stability calculation layer calculates the degree of fluctuation of the motion temporal features in the motion phase, extracts the motion coordination coupling index from the motion temporal features, and calculates the standard deviation of the extracted motion coordination coupling index as the coordination stability term of the motion phase. Specifically, the formulas for calculating the degree of fluctuation in the temporal characteristics of the motion phase are as follows: ; in, This indicates the degree of fluctuation in the temporal characteristics of the motion phase in the m-th group of structured motion state data; S25: Based on the degree of fluctuation of the action timing characteristics and the cooperative stability term of the motion phase, the nonlinear function is used to calculate the motion stability index between the action timing feature vector and the preset motion mode template.

[0011] Specifically, the formula for calculating the motion stability index between the motion temporal feature vector and the preset motion mode template in the structured motion state data is as follows: ; in, This represents the motion stability index between the motion temporal feature vector in the m-th group of structured motion state data and the preset motion pattern template. Control parameters that indicate the degree of fluctuation in the timing characteristics of actions. The control parameters represent the cooperative stability terms. This represents the cooperative stability term for the motion phase in the m-th group of structured motion state data.

[0012] Further, in step S3, the action-aware semantic vector is coupled and mapped with the physiological load feature vector, including: Based on the motion perception semantic vector and physiological load feature vector of the structured motion state data, a coupling mapping method is used to map the physiological load features of different motion stages to obtain the effective motion stimulus intensity of different motion stages. The effective motion stimulus intensities of different motion stages are concatenated as the effective motion stimulus intensity vector of the structured motion state data. The mapping formula for the effective motion stimulus intensity of different motion stages is as follows: ; ; in, The effective motion stimulus intensity of the m-th group of structured motion state data in the e-th motion phase is represented by M, where M represents the number of groups of structured motion state data. The first to fourth motion phases are the preparation phase, squatting phase, lowest point phase, and recovery phase, respectively. The numbers represent the heart rate load index, blood oxygen variability coefficient, and electromyographic activation intensity of the m-th group of structured exercise data in the e-th exercise phase, respectively. This represents the motion deviation score for the m-th group of structured motion state data. The motion stability index represents the m-th group of structured motion state data. This indicates the weighting of the contributions from the heart, lungs, and muscles. Indicates the stability modulation coefficient; This represents the motion deviation term of the m-th group of structured motion state data in the e-th motion stage. This represents the penalty coefficient for motion deviation. Indicates the blood oxygen fluctuation suppression coefficient. This represents an exponential function with the natural constant as its base.

[0013] Furthermore, in step S4, a multi-objective optimization function is constructed by combining individual health status, physical fitness test results, and individual fatigue decay coefficient, including: S41: Collect the user's individual health status and physical fitness test results, wherein the individual health status includes the normalized representation of maximum oxygen uptake, skeletal muscle mass and heart rate variability, and the physical fitness test results are the normalized representation of the distance walked by the user within a specified time. S42: Based on the effective motion stimulus intensity vector of the structured motion state data, the user's individual health status, and the physical fitness test results, construct the structured motion state data relative to the user's individual fatigue decay coefficient; S43: To maximize the efficiency of motion stimulation, a multi-objective optimization function is constructed. The multi-objective optimization function selects several sets of structured motion state data from the structured motion state dataset and optimizes the multi-objective optimization function with the goal of maximizing the motion efficiency of the preset motion mode template corresponding to the selected structured motion state data. The selected and optimized preset motion mode template is then used to form a motion strategy.

[0014] Furthermore, the optimization of the multi-objective optimization function in step S4 further includes: S431: Binary number the structured motion state data in the structured motion state dataset and randomly generate multiple populations, wherein each population contains multiple different sets of structured motion state data that satisfy the preset number of motion mode templates included in the motion strategy. S432: Use the preset movement pattern template in the population as the input of the multi-objective optimization function, and use the output of the multi-objective optimization function as the fitness of the population; S433: With the goal of maximizing the fitness of the population, the population is crossovered and mutated based on the genetic algorithm to iteratively generate a new population until the preset maximum number of iterations is reached, and the preset movement pattern template in the population with the highest fitness is selected to form the movement strategy.

[0015] This invention also proposes a motion strategy optimization generation system with motion perception, which includes a data processing module, a feature extraction module, a feature calculation module, and a motion strategy generation module to realize the motion strategy optimization generation method with motion perception as described above.

[0016] Compared with existing technologies, this invention proposes a motion strategy optimization generation method and system based on motion perception, which has the following beneficial effects: First, this invention utilizes word segmentation and word vector conversion to map phased action descriptions into semantic encoding features. Then, a multi-layer Transformer structure is used to align the semantic space to the motion feature space, making the descriptive text computable. The action deviation score is calculated using phased weighted L2 distance combined with an exponential mapping method, which can highlight the quality differences in key stages (such as the squatting stage and the lowest point stage) and enhance the sensitivity to the core aspects of the action. The action stability index achieves a joint evaluation of local trajectory stability and overall kinetic chain coordination by nonlinearly fusing the degree of action temporal fluctuations and the cooperative stability term.

[0017] Meanwhile, this invention uses a weighted combination of heart rate load index and electromyographic activation intensity as the basic physiological stimulation item. These indicators respectively characterize cardiopulmonary metabolic stress and local muscle mechanical output, allowing the effective stimulus intensity to simultaneously reflect both systemic and local load levels; secondly, a motion stability index is introduced. As a multiplicative modulation factor, it improves stimulus conversion efficiency under good motor control; furthermore, it utilizes an exponential penalty term that includes motor deviation scores and blood oxygen fluctuation coefficients. It provides dual suppression of deviations in movement structure and unstable oxygen supply, preventing high-load but low-quality or high-risk movements from being misjudged as highly effective stimuli. Attached Figure Description

[0018] Figure 1This is a flowchart illustrating a motion strategy optimization generation method for motion perception provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the deployment of a multimodal sensing device according to an embodiment of the present invention; Figure 3 as well as Figure 4 This is a comparison chart of motion strategies under different indicators provided in an embodiment of the present invention.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The realization of the objectives, functional characteristics, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] This invention provides a motion-aware motion strategy optimization generation method and system. The execution entity of the motion-aware motion strategy optimization generation method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the motion-aware motion strategy optimization generation method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0022] Reference Figure 1 as well as Figure 2 As shown, Embodiment 1 of the present invention is as follows: A motion-aware motion strategy optimization generation method, the method comprising: S1: Collect multi-source motion perception data based on a preset motion mode template, extract the motion timing feature vector and physiological load feature vector from the multi-source motion perception data, and use the preset motion mode template, motion timing feature vector and physiological load feature vector as a set of structured motion state data to form a structured motion state dataset.

[0023] Specifically, multi-source motion perception data is collected based on a preset motion pattern template, including: S11: Construct multiple sets of preset exercise mode templates based on the standard training action library. Each preset exercise mode template consists of description text for four exercise stages, namely, preparation stage, squatting stage, lowest point stage, and recovery stage. Specifically, the standard training movement library contains a variety of training movements, including squats, lunges, push-ups, etc. For any training movement in the standard training movement library, the training movement is divided into four movement stages, namely the preparation stage, squatting stage, lowest point stage, and recovery stage, and a description text for each movement stage is generated. The description text of the four movement stages is used as the preset movement mode template corresponding to the training movement. S12: Deploy multimodal sensing devices on standard demonstration personnel, including inertial measurement units, plantar pressure sensors, heart rate sensors, blood oxygen sensors, and surface electromyography sensors. Specifically, refer to, for example Figure 2 The diagram shows the deployment of a multimodal sensing device. The inertial measurement unit (IMU) is fixed to the upper arms, forearms, thighs, and lower legs of a standard demonstrator using elastic straps or motion clips. It collects the three-axis acceleration, three-axis angular velocity, and three-axis magnetic field of the standard demonstrator at different acquisition times at these positions. An attitude fusion algorithm is used to calculate the attitude quaternions for each position, and these quaternions are converted into attitude matrices based on a global coordinate system. The global coordinate system is represented as follows: X-axis represents the east-west direction, Y-axis represents the north-south direction, and Z-axis represents the up-down direction. Furthermore, the attitude fusion algorithm is an extended Kalman filter algorithm. Based on the posture matrices of the standard demonstrator at different acquisition times for the positions of the upper arms, forearms, thighs, and lower legs, the relative rotation matrices between the upper arms and forearms and between the thighs and lower legs are calculated respectively. The relative rotation matrices are then decomposed using Euler angle decomposition to obtain the pitch angles. The pitch angles are used as the joint angles corresponding to the relative rotation matrices, thereby constructing the joint angle vectors of the standard demonstrator at different acquisition times. The joint angle vectors sequentially include the joint angles between the left upper arm and forearm, the right upper arm and forearm, the left thigh and lower leg, and the right thigh and lower leg of the standard demonstrator at different acquisition times. Flexible array-type plantar pressure sensors are embedded in the insoles of the standard demonstrator's feet, and the maximum plantar pressure values ​​of the standard demonstrator's feet at different collection times are calculated to form a plantar pressure sequence at continuous collection times. Heart rate and blood oxygen sensors are worn on the wrist or inner upper arm of standard demonstration personnel to simultaneously collect heart rate and blood oxygen saturation at different collection times, forming heart rate and blood oxygen saturation sequences respectively. Surface electromyography (EMG) sensors were attached to the surface of the main muscle groups of the standard demonstrator to collect EMG signals, with the erector spinae muscles being the main muscle group. It should be noted that the standard demonstrators mentioned are demonstrators who are in the recovery period after exercise. S13: The standard demonstrator completes the specified action under the requirements of the preset movement mode template, and uses the multimodal sensing device to collect multi-source motion perception data to obtain the multi-source motion perception data corresponding to the preset movement mode template. The multi-source motion perception data is then segmented according to the movement stage of the preset movement mode template to obtain multi-source motion sub-perception data for each movement stage. The multi-source motion sub-perception data includes joint angle sequence, plantar pressure sequence, heart rate sequence, blood oxygen saturation sequence, and electromyographic signal.

[0024] Step S1, which extracts the action temporal features and physiological load features from the multi-source action sub-sensing data, also includes: S14: Perform first-order difference on the joint angle sequence to obtain a first-order difference sequence of joint angles, and calculate the sequence mean of the first-order difference sequence of joint angles as a joint angular velocity feature. S15: Perform second-order difference on the joint angle sequence to obtain a second-order difference sequence of joint angles, and calculate the sequence mean of the second-order difference sequence of joint angles as a feature of joint angular acceleration. It should be noted that joint angular velocity characteristics represent the rate of change of joint angle per unit time, reflecting the rhythm and explosiveness of movement. By characterizing the speed of movement through angular velocity, the risk of movement sluggishness or excessive speed can be identified; joint angular acceleration characteristics reflect the acceleration or deceleration trend of movement, and can identify sudden movement changes or impact risks. S16: Based on the joint angle sequence, calculate the fluctuation degree of joint angle and the motion coordination coupling index; normalize the plantar pressure sequence and calculate the standard deviation of the normalized plantar pressure sequence as the plantar pressure fluctuation coefficient; splice the joint angular velocity feature, joint angular acceleration feature, fluctuation degree of joint angle, motion coordination coupling index and plantar pressure fluctuation coefficient to obtain the motion temporal feature of the multi-source motion sub-sensing data; Specifically, the degree of fluctuation of the joint angle and motion coordination coupling index The calculation formula is: ; ; ; ; in, This represents the data value of the joint angle sequence at the nth acquisition time. The j-th joint angle, where The first to fourth joint angles are, in order, the joint angle between the left upper arm and forearm, the joint angle between the right upper arm and forearm, the joint angle between the left thigh and calf, and the joint angle between the right thigh and calf. N represents the length of the sequence data. Let represent the first-order difference value of the j-th joint angle at the nth acquisition time. This represents the first-order difference value of the k-th joint angle at the nth acquisition time. Let represent the root mean square value of the j-th joint angle. This represents the root mean square value of the k-th joint angle. This represents the joint structure coupling coefficient between the j-th joint angle and the k-th joint angle; As an embodiment of the present invention, for example, the joint angle between the left upper arm and forearm and the joint angle between the right upper arm and forearm belong to the contralateral kinetic chain, while the joint angle between the left upper arm and forearm and the joint angle between the left thigh and calf belong to the ipsilateral kinetic chain. It should be noted that this invention extracts joint angular velocity and joint angular acceleration features by performing first-order and second-order difference processing on the joint angle sequence, respectively, to achieve a layered characterization of the rhythmic characteristics and impact trends of movements. By combining the degree of joint angle fluctuation and the motion coordination coupling index, the structural coordination level and control stability between multiple joints can be quantified, thus expanding from single-joint amplitude features to multi-joint dynamic coordination features. Specifically, the degree of joint angle fluctuation characterizes the dispersion of a single joint's motion trajectory relative to its average motion posture, while the motion coordination coupling index characterizes the structurally weighted correlation between the angular velocities of multiple joints. If the angular velocities of the same-side kinetic chain joints change synchronously in the same direction, the coupling value increases, indicating smooth kinetic chain transmission. Simultaneously, this invention introduces a plantar pressure fluctuation coefficient into the movement timing features, enhancing the ability to identify lower limb support stability and center of gravity control status. This gives the movement timing features a multi-dimensional expressive ability of rhythm, stability, and coordination, improving the accuracy of identifying movement sluggishness, imbalance, and explosive abnormalities.

[0025] S17: Obtain the resting heart rate and maximum heart rate of the standard demonstration personnel, and calculate the heart rate load index of the heart rate sequence based on the resting heart rate and maximum heart rate; Specifically, the formula for calculating the heart rate load index of the heart rate sequence is as follows: ; in, Indicates the heart rate load index. This indicates the resting heart rate of the standard demonstration personnel. This indicates the maximum heart rate of the standard demonstration personnel. This represents the data value of the heart rate sequence at the nth acquisition time. S18: Normalize the blood oxygen saturation sequence and electromyography signal, and calculate the standard deviation of the blood oxygen saturation sequence and the root mean square value of the electromyography signal after normalization, which are respectively used as the blood oxygen fluctuation coefficient and the electromyography activation intensity. S19: The heart rate load index, blood oxygen fluctuation coefficient and electromyographic activation intensity are spliced ​​together as the physiological load characteristics of the multi-source sub-motion perception data.

[0026] It should be noted that this invention characterizes the relative load intensity of the cardiovascular system through the heart rate load index, reflects the oxygen supply stability through the blood oxygen fluctuation coefficient, and describes the local muscle recruitment level through electromyographic activation intensity, thereby achieving synchronous modeling of metabolic stress and neuromuscular output.

[0027] Based on the action temporal features and physiological load features of the multi-source action sub-sensing data, a structured motion state dataset is constructed. The structured motion state dataset consists of several sets of structured motion state data, wherein each set of structured motion state data includes a preset motion mode template, an action temporal feature vector, and a physiological load feature vector. The action temporal feature vector is obtained by concatenating the action temporal features of the multi-source action sub-sensing data for each motion stage, and the physiological load feature vector is obtained by concatenating the physiological load features of the multi-source action sub-sensing data for each motion stage.

[0028] As an embodiment of the present invention, the structured motion state dataset is represented in the following form: ; ; ; ; in, Represents a structured motion state dataset. Representing a structured motion state dataset The m-th group of structured motion state data in the dataset, where M represents the number of groups of structured motion state data. The following are structured motion state data. The preset exercise mode template, action timing feature vector, and physiological load feature vector are included. These are, in order, preset motion mode templates. The descriptive text for the preparation phase, the descent phase, the lowest point phase, and the recovery phase. These are, in order, action temporal feature vectors. The timing characteristics of the movements corresponding to the preparation phase, squatting phase, lowest point phase, and recovery phase. These are, in order, the physiological load feature vectors. Physiological load characteristics corresponding to the preparation phase, squatting phase, lowest point phase, and recovery phase.

[0029] S2: Based on the structured motion state dataset, use the motion perception evaluation model to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each group of structured motion state data, and use the motion deviation score and motion stability index as the motion perception semantic vector of the structured motion state data.

[0030] Specifically, the motion perception evaluation model is used to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each set of structured motion state data, including: S21: The motion perception evaluation model includes an input layer, a template semantic embedding layer, a motion deviation scoring calculation layer, and a motion stability calculation layer. The input layer receives the motion temporal feature vector from the structured motion state data and a preset motion mode template. It also performs word segmentation and word vector conversion on the descriptive text in the preset motion mode template to obtain the word vector sequence of the preset motion mode template in the preparation stage, squatting stage, lowest point stage, and recovery stage. Optionally, the jieba word segmentation tool can be used to segment the descriptive text, and the Bert model or Word2Vec model can be used for word vector conversion. S22: The template semantic embedding layer receives word vector sequences of a preset motion mode template in the preparation stage, squatting stage, lowest point stage, and recovery stage, and extracts semantic encoding features of the word vector sequences in the preparation stage, squatting stage, lowest point stage, and recovery stage respectively. Specifically, the template semantic embedding layer adopts a pre-trained embedding network structure to map the word vector sequence corresponding to the descriptive text such as "the squatting phase should move down smoothly" into semantic encoding features consistent with the motion temporal feature distribution. The pre-trained embedding network structure adopts a multi-layer Transformer model structure. S23: The motion deviation scoring calculation layer calculates the stage deviation between the semantic coding features and the motion temporal features of any motion stage, and performs weighted and normalized processing on the stage deviation to obtain the motion deviation score between the motion temporal feature vector in the structured motion state data and the preset motion mode template. Specifically, the formula for calculating the motion deviation score between the motion timing feature vector and the preset motion pattern template is as follows: ; in, This represents the motion deviation score between the motion temporal feature vector in the m-th group of structured motion state data and the preset motion pattern template, where M represents the number of groups of structured motion state data. Indicates the scaling factor, set It is 0.2. This represents the weighting coefficient for the e-th motion phase. The four phases of the exercise are, in order, preparation, squatting, lowest point, and recovery. In order, they are respectively , This represents the temporal characteristics of the action in the e-th motion phase within the m-th group of structured motion state data. This represents the semantic encoding features of the e-th motion stage in the m-th group of structured motion state data. Represents the L2 norm. Represents an exponential function with the natural constant as its base; S24: The motion stability calculation layer calculates the degree of fluctuation of the motion temporal features in the motion phase, extracts the motion coordination coupling index from the motion temporal features, and calculates the standard deviation of the extracted motion coordination coupling index as the coordination stability term of the motion phase. Specifically, the formulas for calculating the degree of fluctuation in the temporal characteristics of the motion phase are as follows: ; in, This indicates the degree of fluctuation in the temporal characteristics of the motion phase in the m-th group of structured motion state data; S25: Based on the degree of fluctuation of the action timing characteristics and the cooperative stability term of the motion phase, the nonlinear function is used to calculate the motion stability index between the action timing feature vector and the preset motion mode template.

[0031] Specifically, the formula for calculating the motion stability index between the motion temporal feature vector and the preset motion mode template in the structured motion state data is as follows: ; in, This represents the motion stability index between the motion temporal feature vector in the m-th group of structured motion state data and the preset motion pattern template. Control parameters that indicate the degree of fluctuation in the timing characteristics of actions. The control parameters representing the cooperative stability term are set. Set to 0.4 It is 0.6. This represents the cooperative stability term for the motion phase in the m-th group of structured motion state data.

[0032] As an embodiment of the present invention, multiple sets of structured motion state data are acquired, and the actual motion deviation score and actual motion stability index of each set of structured motion state data are manually determined. The training loss function is constructed with the goal of minimizing the deviation between the actual motion deviation score in each set of structured motion state data and the motion deviation score output by the model, and the deviation between the actual motion stability index in each set of structured motion state data and the motion stability index output by the model. Based on the training loss function, the trainable parameters in the template semantic embedding layer are optimized and trained. Optionally, the optimization training method adopts the gradient descent algorithm or the Adam optimizer.

[0033] Specifically, the movement deviation score reflects the degree of structural difference between the actual movement timing characteristics and the template. Since the standard demonstrator is in the recovery period, their actual movement timing characteristics represent safe, controllable, and moderately loaded reference characteristics. Therefore, the lower the movement deviation score, the closer the actual movement is to the safe movement pattern during the recovery period. The lower the movement deviation score, the more structural deviation there is from the preset movement pattern template, which may be too fast, too deep, or have unbalanced force exertion, and is not suitable as a movement strategy movement.

[0034] The motion stability index characterizes the level of fluctuation of motion over time and the consistency of multi-joint coordination. The higher the motion stability index, the smoother the rhythm of the preset motion pattern template, the more coordinated and continuous the motion, and the more precise the control. The lower the motion stability index, the more likely there is a risk of rhythm disorder or coordination fluctuation when performing the preset motion pattern template.

[0035] S3: Couple and map the motion perception semantic vector with the physiological load feature vector to obtain the effective motion stimulus intensity vector for each set of structured motion state data.

[0036] Specifically, coupling and mapping the action-aware semantic vector with the physiological load feature vector includes: Based on the motion perception semantic vector and physiological load feature vector of the structured motion state data, a coupling mapping method is used to map the physiological load features of different motion stages to obtain the effective motion stimulus intensity of different motion stages. The effective motion stimulus intensities of different motion stages are concatenated as the effective motion stimulus intensity vector of the structured motion state data. The mapping formula for the effective motion stimulus intensity of different motion stages is as follows: ; ; in, The effective motion stimulus intensity of the m-th group of structured motion state data in the e-th motion phase is represented by M, where M represents the number of groups of structured motion state data. The first to fourth motion phases are the preparation phase, squatting phase, lowest point phase, and recovery phase, respectively. The numbers represent the heart rate load index, blood oxygen variability coefficient, and electromyographic activation intensity of the m-th group of structured exercise data in the e-th exercise phase, respectively. This represents the motion deviation score for the m-th group of structured motion state data. The motion stability index represents the m-th group of structured motion state data. This indicates the weighting of the contributions from the heart, lungs, and muscles. Indicates the stability modulation coefficient, set The values ​​were set to 0.6 and 0.4 respectively. =1; This represents the motion deviation term of the m-th group of structured motion state data in the e-th motion stage. This represents the penalty coefficient for motion deviation. Indicates the blood oxygen fluctuation suppression coefficient. This represents an exponential function with the natural constant as its base, and is set to... Set to 0.4 It is 0.3.

[0037] Furthermore, the effective exercise stimulus intensity refers to the effective amount of training stimulus that is truly absorbed by the body and can produce adaptive changes within a unit exercise phase, provided that the movement quality is compliant, the neural control is stable, and the physiological safety is guaranteed.

[0038] S4: Based on the effective motion stimulus intensity vector of each set of structured motion state data, combined with individual health status, physical fitness test results and individual fatigue decay coefficient, construct a multi-objective optimization function, optimize the multi-objective optimization function, select multiple preset exercise mode templates suitable for individual health status, and use the selected preset exercise mode templates to form an exercise strategy.

[0039] In step S4, a multi-objective optimization function is constructed by combining individual health status, physical fitness test results, and individual fatigue decay coefficient, including: S41: Collect the user's individual health status and physical fitness test results, wherein the individual health status includes the normalized representation of maximum oxygen uptake, skeletal muscle mass and heart rate variability, and the physical fitness test results are the normalized representation of the distance walked by the user within a specified time; optionally, the specified time is set to 8 minutes or 10 minutes; S42: Based on the effective motion stimulus intensity vector of the structured motion state data, the user's individual health status, and the physical fitness test results, construct the structured motion state data relative to the user's individual fatigue decay coefficient; Specifically, the formula for calculating the structured motion state data relative to the user's individual fatigue attenuation coefficient is as follows: ; in, This represents the individual fatigue decay coefficient of the m-th group of structured motion state data relative to the user. The normalized representations of maximum oxygen uptake, skeletal muscle mass, and heart rate variability are shown in that order, respectively. This indicates the results of the physical fitness test; The higher the individual fatigue decay coefficient, the heavier the physical burden on the current user's body from the preset exercise mode template corresponding to the structured motion state data. S43: To maximize the efficiency of motion stimulation, a multi-objective optimization function is constructed. The multi-objective optimization function selects several sets of structured motion state data from the structured motion state dataset and optimizes the multi-objective optimization function with the goal of maximizing the motion efficiency of the preset motion mode template corresponding to the selected structured motion state data. The selected and optimized preset motion mode template is then used to form a motion strategy.

[0040] Specifically, the functional expression of the multi-objective optimization function is as follows: ; in, Represents a multi-objective optimization function. This indicates the motion strategy to be optimized. , Representing movement strategies The q-th preset motion mode template in the text, where Q represents the number of preset motion mode templates included in the motion strategy, is set to 4. Preset motion mode template The corresponding structured motion state data relative to the user's individual fatigue decay coefficient. This represents the preset minimum individual fatigue decay coefficient (e.g., set to 1). Indicates selection The minimum value in, Indicates the attenuation control coefficient, set It is 1.2. Preset motion mode template The corresponding structured motion state data represents the effective motion stimulus intensity in the e-th motion phase. Preset motion mode template The motion efficiency of the preset motion mode template corresponding to the structured motion state data.

[0041] It should be noted that this invention constructs an individual fatigue decay coefficient by introducing individual health status and physical fitness test results, thereby realizing a proportional modeling between exercise stimulus intensity and individual carrying capacity. It uses normalized indicators such as maximum oxygen uptake, skeletal muscle mass, heart rate variability, and timed walking distance to form a comprehensive capacity denominator, enabling the same structured exercise state data to generate differentiated burden assessments among different individuals.

[0042] Based on this, the present invention embeds the individual fatigue decay coefficient into the exponential inhibition function and multiplies it with the phased effective exercise stimulus intensity to form an exercise efficiency term. This allows the high-stimulation but high-burden template to be automatically subject to decay constraints, thereby constructing a multi-objective optimization function. This function ensures the benefits of exercise stimulus while suppressing the risk of individual overload, achieving a movement strategy generation mechanism that maximizes benefits, controls fatigue, and matches abilities. This improves the safety, individual adaptability, and training efficiency of the movement strategy.

[0043] The optimization of the multi-objective optimization function in step S4 further includes: S431: Binary number the structured motion state data in the structured motion state dataset and randomly generate multiple populations, wherein each population contains multiple different sets of structured motion state data that satisfy the preset number of motion mode templates included in the motion strategy. S432: Use the preset movement pattern template in the population as the input of the multi-objective optimization function, and use the output of the multi-objective optimization function as the fitness of the population; S433: With the goal of maximizing the fitness of the population, the population is crossovered and mutated based on the genetic algorithm to iteratively generate a new population until the preset maximum number of iterations is reached, and the preset movement pattern template in the population with the highest fitness is selected to form the movement strategy.

[0044] Specifically, the maximum number of iterations is set to 50.

[0045] Example 2: A motion-aware motion strategy optimization generation system includes a data processing module, a feature extraction module, a feature calculation module, and a motion strategy generation module, to implement the motion-aware motion strategy optimization generation method as described in Example 1: The data processing module is used to collect multi-source motion perception data based on a preset motion mode template, extract the motion timing feature vector and physiological load feature vector of the multi-source motion perception data, and use the preset motion mode template, motion timing feature vector and physiological load feature vector as a set of structured motion state data. The feature extraction module is used to calculate the action deviation score and action stability index between the action temporal feature vector and the preset motion mode template in each group of structured motion state data using the action perception evaluation model, and uses the action deviation score and action stability index as the action perception semantic vector of the structured motion state data. The feature calculation module is used to couple and map the action perception semantic vector with the physiological load feature vector to obtain the effective motion stimulus intensity vector for each set of structured motion state data. The exercise strategy generation module is used to combine an individual's health status, physical fitness test results, and individual fatigue decay coefficient to construct a multi-objective optimization function, optimize the multi-objective optimization function, select multiple preset exercise mode templates suitable for the individual's health status, and construct an exercise strategy from the selected preset exercise mode templates.

[0046] Example 3: In another embodiment of the present invention, multiple users of the same age and condition are identified, and standard exercise strategies and motion-aware exercise strategies are provided to these users respectively. The comparison of users' 6-minute walking distance and maximum oxygen uptake is observed over multiple exercise cycles, generating data such as... Figure 3 as well as Figure 4 The diagram showing the comparison of exercise strategies illustrates that the standard exercise strategy consists of regular and fixed training movements.

[0047] According to the exercise strategy comparison chart, as the exercise cycle increases, the recovery effect of the motion-aware exercise strategy is significantly better than that of the standard exercise strategy, and the gap between the two continues to widen. Specifically, for the 6-minute walking distance, the motion-aware exercise strategy increased from a 20-meter lead over 7 days to a 65-meter lead over 42 days; for the VO2 max dimension, the motion-aware exercise strategy increased from a 2 ml / (kg・min) lead over 7 days to a 5.5 ml / (kg・min) lead over 42 days, with both the overall improvement magnitude and growth rate being superior.

[0048] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in terms of the scope of the patent invention.

[0049] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0050] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0051] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for optimizing and generating motion strategies based on action perception, characterized in that, The method includes: S1: Collect multi-source motion perception data based on a preset motion mode template, extract the motion timing feature vector and physiological load feature vector from the multi-source motion perception data, and use the preset motion mode template, motion timing feature vector and physiological load feature vector as a set of structured motion state data to form a structured motion state dataset. S2: Based on the structured motion state dataset, use the motion perception evaluation model to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each group of structured motion state data, and use the motion deviation score and motion stability index as the motion perception semantic vector of the structured motion state data. S3: Couple and map the motion perception semantic vector with the physiological load feature vector to obtain the effective motion stimulus intensity vector for each set of structured motion state data; S4: Based on the effective motion stimulus intensity vector of each set of structured motion state data, combined with individual health status, physical fitness test results and individual fatigue decay coefficient, construct a multi-objective optimization function, optimize the multi-objective optimization function, select multiple preset exercise mode templates suitable for individual health status, and use the selected preset exercise mode templates to form an exercise strategy.

2. The motion strategy optimization and generation method for action perception as described in claim 1, characterized in that, Step S1 involves collecting multi-source motion perception data based on a preset motion pattern template, including: S11: Construct multiple sets of preset exercise mode templates based on the standard training action library. Each preset exercise mode template consists of description text for four exercise stages, namely, preparation stage, squatting stage, lowest point stage, and recovery stage. S12: Deploy multimodal sensing devices on standard demonstration personnel, including inertial measurement units, plantar pressure sensors, heart rate sensors, blood oxygen sensors, and surface electromyography sensors. S13: The standard demonstrator completes the specified action under the requirements of the preset movement mode template, and uses the multimodal sensing device to collect multi-source motion perception data to obtain the multi-source motion perception data corresponding to the preset movement mode template. The multi-source motion perception data is then segmented according to the movement stage of the preset movement mode template to obtain multi-source motion sub-perception data for each movement stage. The multi-source motion sub-perception data includes joint angle sequence, plantar pressure sequence, heart rate sequence, blood oxygen saturation sequence, and electromyographic signal.

3. The motion strategy optimization and generation method for action perception as described in claim 2, characterized in that, Step S1, which extracts the action temporal features and physiological load features from the multi-source action sub-sensing data, also includes: S14: Perform first-order difference on the joint angle sequence to obtain a first-order difference sequence of joint angles, and calculate the sequence mean of the first-order difference sequence of joint angles as a joint angular velocity feature. S15: Perform second-order difference on the joint angle sequence to obtain a second-order difference sequence of joint angles, and calculate the sequence mean of the second-order difference sequence of joint angles as a feature of joint angular acceleration. S16: Based on the joint angle sequence, calculate the fluctuation degree of joint angle and the motion coordination coupling index; normalize the plantar pressure sequence and calculate the standard deviation of the normalized plantar pressure sequence as the plantar pressure fluctuation coefficient; splice the joint angular velocity feature, joint angular acceleration feature, fluctuation degree of joint angle, motion coordination coupling index and plantar pressure fluctuation coefficient to obtain the motion temporal feature of the multi-source motion sub-sensing data; S17: Obtain the resting heart rate and maximum heart rate of the standard demonstration personnel, and calculate the heart rate load index of the heart rate sequence based on the resting heart rate and maximum heart rate; S18: Normalize the blood oxygen saturation sequence and electromyography signal, and calculate the standard deviation of the blood oxygen saturation sequence and the root mean square value of the electromyography signal after normalization, which are respectively used as the blood oxygen fluctuation coefficient and the electromyography activation intensity. S19: The heart rate load index, blood oxygen fluctuation coefficient and electromyographic activation intensity are spliced ​​together as the physiological load characteristics of the multi-source sub-motion perception data.

4. The motion strategy optimization and generation method for action perception as described in claim 3, characterized in that, Based on the action temporal features and physiological load features of the multi-source action sub-sensing data, a structured motion state dataset is constructed. The structured motion state dataset consists of several sets of structured motion state data, wherein each set of structured motion state data includes a preset motion mode template, an action temporal feature vector, and a physiological load feature vector. The action temporal feature vector is obtained by concatenating the action temporal features of the multi-source action sub-sensing data for each motion stage, and the physiological load feature vector is obtained by concatenating the physiological load features of the multi-source action sub-sensing data for each motion stage.

5. The motion strategy optimization and generation method for action perception as described in claim 1, characterized in that, In step S2, the motion perception evaluation model is used to calculate the motion deviation score and motion stability index between the motion temporal feature vector and the preset motion mode template in each set of structured motion state data, including: S21: The motion perception evaluation model includes an input layer, a template semantic embedding layer, a motion deviation scoring calculation layer, and a motion stability calculation layer. The input layer receives the motion temporal feature vector from the structured motion state data and a preset motion mode template. It also performs word segmentation and word vector conversion on the descriptive text in the preset motion mode template to obtain the word vector sequence of the preset motion mode template in the preparation stage, squatting stage, lowest point stage, and recovery stage. S22: The template semantic embedding layer receives word vector sequences of a preset motion mode template in the preparation stage, squatting stage, lowest point stage, and recovery stage, and extracts semantic encoding features of the word vector sequences in the preparation stage, squatting stage, lowest point stage, and recovery stage respectively. S23: The motion deviation scoring calculation layer calculates the stage deviation between the semantic coding features and the motion temporal features of any motion stage, and performs weighted and normalized processing on the stage deviation to obtain the motion deviation score between the motion temporal feature vector in the structured motion state data and the preset motion mode template. S24: The motion stability calculation layer calculates the degree of fluctuation of the motion temporal features in the motion phase, extracts the motion coordination coupling index from the motion temporal features, and calculates the standard deviation of the extracted motion coordination coupling index as the coordination stability term of the motion phase. S25: Based on the degree of fluctuation of the action timing characteristics and the cooperative stability term of the motion phase, the nonlinear function is used to calculate the motion stability index between the action timing feature vector and the preset motion mode template.

6. The motion strategy optimization and generation method for action perception as described in claim 1, characterized in that, Step S3 involves coupling and mapping the action-aware semantic vector with the physiological load feature vector, including: Based on the motion perception semantic vector and physiological load feature vector of the structured motion state data, a coupling mapping method is used to map the physiological load features of different motion stages to obtain the effective motion stimulus intensity of different motion stages. The effective motion stimulus intensities of different motion stages are concatenated as the effective motion stimulus intensity vector of the structured motion state data. The mapping formula for the effective motion stimulus intensity of different motion stages is as follows: ； ； in, The effective motion stimulus intensity of the m-th group of structured motion state data in the e-th motion phase is represented by M, where M represents the number of groups of structured motion state data. The first to fourth motion phases are the preparation phase, squatting phase, lowest point phase, and recovery phase, respectively. The numbers represent the heart rate load index, blood oxygen variability coefficient, and electromyographic activation intensity of the m-th group of structured exercise data in the e-th exercise phase, respectively. This represents the motion deviation score for the m-th group of structured motion state data. The motion stability index represents the m-th group of structured motion state data. This indicates the weighting of the contributions from the heart, lungs, and muscles. Indicates the stability modulation coefficient; This represents the motion deviation term of the m-th group of structured motion state data in the e-th motion stage. This represents the penalty coefficient for motion deviation. Indicates the blood oxygen fluctuation suppression coefficient. This represents an exponential function with the natural constant as its base.

7. The motion strategy optimization and generation method for action perception as described in claim 1, characterized in that, In step S4, a multi-objective optimization function is constructed by combining individual health status, physical fitness test results, and individual fatigue decay coefficient, including: S41: Collect the user's individual health status and physical fitness test results, wherein the individual health status includes the normalized representation of maximum oxygen uptake, skeletal muscle mass and heart rate variability, and the physical fitness test results are the normalized representation of the distance walked by the user within a specified time. S42: Based on the effective motion stimulus intensity vector of the structured motion state data, the user's individual health status, and the physical fitness test results, construct the structured motion state data relative to the user's individual fatigue decay coefficient; S43: To maximize the efficiency of motion stimulation, a multi-objective optimization function is constructed. The multi-objective optimization function selects several sets of structured motion state data from the structured motion state dataset and optimizes the multi-objective optimization function with the goal of maximizing the motion efficiency of the preset motion mode template corresponding to the selected structured motion state data. The selected and optimized preset motion mode template is then used to form a motion strategy.

8. The motion strategy optimization and generation method for action perception as described in claim 7, characterized in that, The optimization of the multi-objective optimization function in step S4 further includes: S431: Binary number the structured motion state data in the structured motion state dataset and randomly generate multiple populations, wherein each population contains multiple different sets of structured motion state data that satisfy the preset number of motion mode templates included in the motion strategy. S432: Use the preset movement pattern template in the population as the input of the multi-objective optimization function, and use the output of the multi-objective optimization function as the fitness of the population; S433: With the goal of maximizing the fitness of the population, the population is crossovered and mutated based on the genetic algorithm to iteratively generate a new population until the preset maximum number of iterations is reached, and the preset movement pattern template in the population with the highest fitness is selected to form the movement strategy.

9. A motion strategy optimization and generation system with motion perception, characterized in that, The motion strategy generation system includes a data processing module, a feature extraction module, a feature calculation module, and a motion strategy generation module, to implement a motion strategy optimization generation method for motion perception as described in any one of claims 1-8.

Images