Walking function evaluation and training system based on augmented reality

Abstract

The invention discloses a walking function evaluation and training system based on augmented reality, relates to the field of walking function evaluation, and is used for solving the problem that a self-adaptive model cannot meet various patient requirements and experiences, collecting a main configuration parameter deletion height, a main configuration parameter complexity difference, a step length fluctuation value difference and a patient gait symmetry difference. And substituting into a logistic regression formula for calculation to obtain a reasonable evaluation coefficient, comparing the reasonable evaluation coefficient with a preset reasonable threshold value, performing deep scheduling mechanism processing on the adaptive model according to a comparison result to obtain a deep scheduling score, and determining the deep scheduling intensity according to the score. The historical self-adaptive iteration times are obtained by determining the iteration times of the self-adaptive model consistent with the deep scheduling intensity of this time, the training progress index is obtained by collecting the training data of this round, and the continuous iteration result of the self-adaptive model is determined by using fuzzy logic, so that the training difficulty of the patient is reduced, the training probability is improved, and the training experience of the patient is met.

Description

Technical Field

[0001] The present invention relates to the field of walking function evaluation, and more specifically, to an augmented reality-based walking function evaluation and training system. Background Art

[0002] In recent years, with the rapid development of augmented reality (AR) technology, AR has been widely applied in multiple fields such as medical treatment, rehabilitation, and education. Especially in gait analysis and walking function evaluation and training, augmented reality technology has played an important role. Generally, the combination of repetitive transcranial magnetic stimulation and augmented reality gait adaptive training can reflect the impact on the walking function of stroke patients. Stroke patients usually have abnormal gaits due to brain structure damage (such as lesions in the brain, amygdala, epilepsy, etc.), thus affecting their normal walking ability;

[0003] Furthermore, based on three-dimensional gait analysis and sEMG of stroke patients, the effect of gait adaptive training combining repetitive transcranial magnetic stimulation and augmented reality in the International Journal of Physical Therapy: A randomized controlled trial has obtained the effectiveness of the combination of repetitive transcranial magnetic stimulation and augmented reality. The combined treatment can significantly improve the walking ability and motor function of patients, providing new ideas for clinical rehabilitation;

[0004] The existing technology has the following deficiencies:

[0005] Currently, augmented reality technology superimposes virtual elements on the real environment. The parameters configured for the virtual elements usually build an adaptive iterative model based on the mapping relationship between the training difficulty and the main configuration parameters of the virtual elements and apply it to the corresponding processes in rehabilitation training. However, for different patients and different rehabilitation training processes, the main configuration parameters of the virtual elements may not meet the diverse patient needs and experiences based on a simple adaptive model, increasing the training difficulty for patients, raising the probability of training failure, and reducing the patient training experience. Therefore, an augmented reality-based walking function evaluation and training system is proposed.

[0006] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the existing technology, the embodiments of the present invention provide an augmented reality-based walking function evaluation and training system, and solve the problems raised in the above background art by using different product inspection methods.

[0008] To achieve the above object, the present invention provides the following technical solutions. An augmented reality-based walking function evaluation and training system includes a data acquisition module, a rational analysis module, a deep scheduling module, and a scheduling iteration module; the modules are signal-connected to each other;

[0009] The data acquisition module is used to collect the main configuration adjustment coefficient in the adaptive main configuration parameters of two randomly adjacent rounds within a period of time, and obtain the main configuration adjustment state. According to the main configuration adjustment state, the main configuration parameter deletion height and the main configuration parameter complexity difference are obtained. At the same time, the walking states of the patient in two adjacent rounds are collected to obtain the step length fluctuation value difference and the patient gait symmetry difference, and sent to the rational analysis module;

[0010] The rational analysis module is used to receive the main configuration parameter deletion height, the main configuration parameter complexity difference, the step length fluctuation value difference, and the patient gait symmetry difference, substitute them into the data analysis model to obtain a rational evaluation coefficient, and send it to the deep scheduling module;

[0011] The deep scheduling module is used to receive the rational evaluation coefficient, compare it with a preset rational threshold, and according to the comparison result, perform a deep scheduling mechanism process on the adaptive model to obtain a deep scheduling score, determine the deep scheduling intensity according to the score, and send it to the scheduling iteration module;

[0012] The scheduling iteration module is used to receive the deep scheduling intensity, obtain the historical adaptive iteration times by determining the number of adaptive model iterations consistent with the current deep scheduling intensity, obtain the training progress index by collecting the training data of this round, and use fuzzy logic to determine the continuous iteration result of the adaptive model.

[0013] In a preferred embodiment, the main configuration adjustment state includes the main configuration parameter deletion height and the main configuration parameter complexity difference; the walking states of the patient in two adjacent rounds include the step length fluctuation value difference and the patient gait symmetry difference;

[0014] The data acquisition module analyzes the reduction of training frequency, the deletion amount of virtual pedestrian quantity, the deletion amount of virtual pedestrian body posture, the deletion amount of virtual environment content, and the deletion amount of the target forward inclination torso angle for different training contents within two adjacent rounds, calculates the difference between the deletion amounts of all main configuration parameters in the current round and the deletion amounts of all main configuration parameters in the previous round to obtain the deletion amount of each main configuration parameter, and performs weighted calculation to obtain the main configuration parameter deletion height Z;

[0015] Through the geometric complexity algorithm, the number of faces of the polyhedron objects in the virtual environment is counted, and the ratio of the number of objects to the volume of the area where the objects are located is calculated to obtain the virtual environment space density. Then, weighted calculation is performed to determine the configuration complexity of the virtual environment. The configuration complexity of the virtual environment in the current round is subtracted from that in the previous round to obtain the difference P in the complexity of the main configuration parameters;

[0016] First, analyze the patient's walking state, calculate the step length of each step, and measure the stepping distances of the left and right feet. Compare the step length differences between adjacent steps in the current round, and count the total step length fluctuation value in the current round. Then subtract the total step length fluctuation value in the previous round from it to obtain the difference U in the step length fluctuation value;

[0017] First, analyze the patient's walking state, and then define it as the ratio of the left step length to the right step length to determine the patient's gait symmetry. Calculate the difference between the total gait symmetry of the patient in the current round and that in the previous round to obtain the difference J in the patient's gait symmetry.

[0018] In a preferred embodiment, substitute the deletion height of the main configuration parameter, the difference in the complexity of the main configuration parameter, the difference in the step length fluctuation value, and the difference in the patient's gait symmetry into the logistic regression formula to calculate the reasonable evaluation coefficient. The specific formula is as follows:

[0019]

[0020] In the formula, L is the result of the logistic regression calculation, that is, the reasonable evaluation coefficient, e is the natural base, y is the linear combination term of the logistic regression model. Specifically, y can be set as:

[0021]

[0022] In the formula, β0 is the bias term, and β1, β2, β3, and β4 are the regression coefficients of the deletion height of the main configuration parameter, the difference in the complexity of the main configuration parameter, the difference in the step length fluctuation value, and the difference in the patient's gait symmetry, respectively.

[0023] In a preferred embodiment, after obtaining the reasonable evaluation coefficient, compare and analyze the reasonable evaluation coefficient with the continuously iterated reasonable threshold;

[0024] If the reasonable evaluation coefficient is greater than or equal to the reasonable threshold, mark the adaptive model deletion mechanism in the current round as unreasonable deletion and generate a scheduling signal;

[0025] If the reasonable evaluation coefficient is less than the reasonable threshold, mark the adaptive model deletion mechanism in the current round as reasonable deletion and generate an end signal.

[0026] In a preferred embodiment, the adaptive model of the current round greater than the reasonable threshold is substituted into the deep scheduling mechanism for processing. From the historical adaptive model, the main configuration parameter library is deleted, and the round of the main configuration parameter of the adaptive model most similar to the main configuration parameter deleted by the current adaptive model is analyzed. Five corresponding vectors for evaluating similarity are obtained by reducing the training frequency, the deletion amount of virtual pedestrian quantity, the deletion amount of virtual pedestrian body posture, the deletion amount of virtual environment content, and the deletion amount of the target forward inclination torso angle.

[0027] The training frequency reduction, the deletion amount of virtual pedestrian quantity, the deletion amount of virtual pedestrian body posture, the deletion amount of virtual environment content, and the deletion amount of the target forward inclination torso angle are vectorized and substituted into the similarity calculation formula to obtain the round of the main configuration parameter of the adaptive model most similar to the main configuration parameter deleted by the current adaptive model.

[0028] In a preferred embodiment, the reasonable evaluation coefficients of the round of the main configuration parameter deleted by the current adaptive model and the round of the main configuration parameter of the most similar adaptive model are compared in size, and the larger reasonable evaluation coefficient is subtracted from the smaller reasonable evaluation coefficient to obtain the difference in the reasonable evaluation coefficient of the similar deletion round.

[0029] The difference in the reasonable evaluation coefficient of the similar deletion round is multiplied by the reasonable evaluation coefficient of the round of the main configuration parameter deleted by the current adaptive model to obtain the deep scheduling score.

[0030] The deep scheduling score is substituted into the historical deep scheduling score library, the ranking of the deep scheduling score is analyzed, and the deep scheduling intensity is obtained according to the ranking.

[0031] If the deep scheduling score value is in the top 20% in the historical deep scheduling score library, the first-level deep scheduling intensity is enabled; otherwise, the second-level deep scheduling intensity is enabled.

[0032] In a preferred embodiment, by collecting the iteration times of multiple adaptive models corresponding to the deep scheduling intensity, the iteration times of all adaptive models are accumulated and the ratio is calculated with the number of adaptive models with increased deletion intensity to obtain the historical adaptive iteration times.

[0033] By collecting the current gait speed minus the ideal gait speed and calculating the ratio with the ideal gait speed, the training progress index is obtained.

[0034] In a preferred embodiment, the historical adaptive iteration times and the training progress index are defined as input variables and are respectively divided into different fuzzy sets.

[0035] The continuous iteration result of the adaptive model is defined as the output variable and is divided into a fuzzy set.

[0036] Formulate fuzzy rules to describe the impact of the historical adaptive iteration times and the training progress index on the continuous iteration results of the adaptive model;

[0037] Conduct fuzzy reasoning according to the fuzzy rules to determine the continuous iteration scheme of the adaptive model.

[0038] The technical effects and advantages of the present invention:

[0039] 1. By collecting the deletion height of the main configuration parameters, the complexity difference of the main configuration parameters, the step size fluctuation value difference, and the patient gait symmetry difference, and substituting them into the calculation of the logistic regression formula, the present invention obtains a reasonable evaluation coefficient, compares it with a preset reasonable threshold, and based on the comparison result, performs a deep scheduling mechanism process on the adaptive model to obtain a deep scheduling score, and determines the deep scheduling intensity according to the score, reducing the training difficulty of patients, increasing the training probability, and meeting the training experience of patients.

[0040] 2. By receiving the deep scheduling intensity, determining the adaptive model iteration times consistent with the current deep scheduling intensity to obtain the historical adaptive iteration times, collecting the training data of this round to obtain the training progress index, and using fuzzy logic to determine the continuous iteration results of the adaptive model, the present invention reduces the computational complexity, reduces the scheduling times of the adaptive model, and stabilizes the operation efficiency of the system. Brief Description of the Drawings

[0041] Figure 1 It is a flowchart implementation diagram of the walking function evaluation and training system based on augmented reality of the present invention.

[0042] Figure 2 It is a module schematic diagram of the walking function evaluation and training system based on augmented reality of the present invention. Detailed Embodiment

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0044] Specifically, the effect of the combined repetitive transcranial magnetic stimulation and augmented reality gait adaptive training on walking function was based on three-dimensional gait analysis and sEMG of stroke patients: During the randomized controlled trial, a prospective, randomized, single-blind, parallel-group controlled design was adopted. The participants were randomly divided into an experimental group (receiving combined rTMS and ARGAT treatment) and a control group (receiving only ARGAT), and stroke survivors meeting specific inclusion and exclusion criteria were recruited;

[0045] To ensure the representativeness of the samples and the reliability of the experimental results, the experimental group underwent 20 sessions of combined rTMS and ARGAT treatment, while the control group underwent 20 sessions of ARGAT training only. The intervention lasted for four weeks, five days a week. Then, various assessment tools such as three-dimensional gait analysis (3DGA), surface electromyography (sEMG), Fugl-Meyer assessment of the lower extremities (FMA-LE), and Berg Balance Scale (BBS) were used to comprehensively measure the motor function of the participants;

[0046] A mixed two-way analysis of variance (ANOVA) was used to compare the changes before and after treatment in the two groups, evaluate the main effects and interaction effects of the group and time, and determine the significance of the results through statistical methods;

[0047] Meanwhile, an adaptive model was added to adaptively delete the main configuration parameters for different patients and different rehabilitation training processes to better meet the usage experience of the patients;

[0048] Example 1

[0049] The present invention discloses a walking function evaluation and training system based on augmented reality, as Figures 1 to 2 shown, which includes a data acquisition module, a reasonable analysis module, a deep scheduling module, and a scheduling iteration module; the modules are signal-connected to each other;

[0050] The data acquisition module is used to collect the main configuration adjustment coefficients in the adaptive main configuration parameters of two randomly adjacent rounds within a period of time, and obtain the main configuration adjustment status. Based on the main configuration adjustment status, the main configuration parameter deletion height and the main configuration parameter complexity difference are obtained. At the same time, the walking states of the patients in two adjacent rounds are collected to obtain the step length fluctuation value difference and the patient gait symmetry difference, and they are sent to the reasonable analysis module;

[0051] Among them, the adaptive main configuration parameters include different training items (such as forward flexion, standing, independent walking, etc.), the number of virtual pedestrians generated during independent walking training, the postures of the virtual pedestrians, different virtual environment settings (such as rural, urban, natural scenes, etc.), and the training difficulty is adjusted by the target forward inclination trunk angle;

[0052] It should be noted that the number of virtual pedestrians generated during independent walking training, the postures of the virtual pedestrians, and different virtual environment settings (such as rural, urban, natural scenes, etc.) are all virtual elements superimposed on the real environment. The virtual elements include, but are not limited to, the number of virtual pedestrians, the postures of the virtual pedestrians, and different virtual environment settings, which will not be elaborated here;

[0053] Then, the deletion of the adaptive main configuration parameters includes the deletion for different training items, the deletion of the number of virtual pedestrians, the deletion of the postures of virtual pedestrians (such as reducing the volume of pedestrians to reduce the complexity of training, etc.), the deletion of the virtual environment (such as the deletion of the number of changes in the virtual environment and the deletion of obstacles in the scene during one round of training, etc.), and the deletion of the range of the target forward-tilt torso angle, etc.;

[0054] As can be seen from the above, the adaptive main configuration parameters are set by relevant professional doctors and the experimenters of this experiment, which will not be elaborated here;

[0055] Among them, the main configuration adjustment state includes the deletion height of the main configuration parameters and the complexity difference of the main configuration parameters; the walking states of patients in two adjacent rounds include the difference in the step length fluctuation value and the difference in the gait symmetry of the patients;

[0056] Since the adjacent two rounds of adaptive main configuration parameters are randomly selected within a certain period of time as described above, the selection time is not limited. It can be the first round and the second round of the patient, or any two adjacent rounds in the middle of the patient's training process, etc. The specific selection time and the random selection rule are not limited and will not be elaborated here;

[0057] Specifically, the random selection rule can be based on using programming tools (such as Python, etc.) to generate random selection time points;

[0058] The deletion height of the main configuration parameters refers to the degree of simplification of the virtual environment configuration by the adaptive model after each round of training, that is, the height by which the configuration parameters of the virtual environment are deleted each time during training. Specifically, this reflects the degree of simplifying the training content by adjusting the parameters of the virtual environment (such as the number of virtual pedestrians, the virtual environment scene, the training difficulty, etc.). Its acquisition logic is to analyze the reduction frequency of training, the deletion quantity of the number of virtual pedestrians, the deletion quantity of the postures of virtual pedestrians, the deletion quantity of the virtual environment content, and the deletion quantity of the target forward-tilt torso angle for different training contents within two adjacent rounds, calculate the difference between the deletion quantity of all main configuration parameters in the current round and the deletion quantity of all main configuration parameters in the previous round to obtain the deletion quantity of each main configuration parameter, and perform weighted calculation to obtain the deletion height Z of the main configuration parameters;

[0059] Specifically, the screening of all main configuration parameters is to help patients adapt to training and improve the training process. Compared with adding main configuration parameters, deleting main configuration parameters can better meet the rehabilitation training of stroke patients. While simplifying the training, the adaptive model can find the training mode that best suits the patient;

[0060] Among them, the specific content of the deletion is as follows: reducing the standing time or increasing the walking time in the training content, reducing the number of virtual pedestrians from multiple to a few, reducing the volume of virtual pedestrians in the virtual pedestrian body shape, reducing their movement complexity, reducing the number of obstacles in the scene or simplifying the scene type in the virtual environment content, narrowing the forward inclination angle in the target forward inclination torso angle to reduce the training difficulty, etc.;

[0061] The complexity difference of the main configuration parameters refers to the change in the complexity of the virtual environment configuration in two adjacent rounds of training, which measures the change range of the training difficulty and reflects how the adaptive model adjusts the complexity of the virtual environment to adapt to the patient's rehabilitation process. Its acquisition logic is to use the geometric complexity algorithm to count the number of faces of the polyhedron objects in the virtual environment, calculate the ratio of the number of objects to the volume of the area where the objects are located to obtain the virtual environment space density, perform weighted calculation to determine the complexity of the virtual environment configuration, and subtract the complexity of the virtual environment configuration in the previous round from the complexity of the virtual environment configuration in the current round to obtain the complexity difference P of the main configuration parameters;

[0062] Specifically, the analysis and judgment of each polyhedron object are determined by the preset camera of the experimenter based on augmented reality. The specific area division corresponding to the virtual elements in augmented reality can be based on the experimental data or historical data preset by the experimenter. The rules of the specific area division are not limited and will not be elaborated here;

[0063] Among them, the patient's walking state can generally be divided into two types. One walking state is that after the left or right foot steps out, the rear foot follows and steps out to twice the distance of the front foot step. That is, if the front foot step distance is 50 cm, then the rear foot step distance is 100 cm, which is the normal walking mode. The other walking state is that after the left or right foot steps out, the rear foot follows and steps out to the same distance as the front foot step. That is, if the front foot step distance is 50 cm, then the rear foot step distance is 50 cm. This is a common pathological walking mode for some stroke patients who just need rehabilitation training. For this mode, it cannot be quickly corrected in the early stage of training. Therefore, it is necessary to first identify the patient's walking state and then conduct subsequent analysis to avoid the situation of wrongly analyzing an overly large difference value, resulting in data distortion;

[0064] The step length fluctuation value difference refers to the inconsistency of the step lengths in the patient's gait. Especially in the rehabilitation process of stroke patients, due to the combination of pathological gait (such as pathological gait patterns) and training, the patient's step lengths show large fluctuations, resulting in inconsistent steps, indicating that the parameter deletion of the adaptive model is unreasonable. Its acquisition logic is to first analyze the patient's walking state, calculate the step length of each step, measure the step distances of the left and right feet, compare the step length differences of adjacent steps in the current round, count the total step length fluctuation value in the current round, and subtract the total step length fluctuation value in the previous round to obtain the step length fluctuation value difference U;

[0065] Specifically, the formula for calculating the fluctuation difference of the step length is expressed as follows:

[0066]

[0067] In the formula, ΔL is the step length difference between adjacent steps, N is the number of steps required in one round, L i is the step length of the i-th step, is the average value of the step length;

[0068] It should be noted that the number of walking steps within each round should be kept consistent to clearly observe the patient's training situation. The specific value of the number of walking steps within a specific round is determined by the experimenter and the specific condition of the patient, which will not be elaborated here;

[0069] The patient gait symmetry difference refers to the coordination of the patient's left and right foot gaits, especially the symmetry of the steps. In normal walking, the step lengths and step frequencies of the left and right feet should be close. In the pathological walking mode, due to the intervention of stroke and training, the gait symmetry of the left and right feet should tend to be balanced. The symmetry difference reflects the irrationality of the parameters deleted by the adaptive model. Its acquisition logic is to first analyze the patient's walking state, then define it as the ratio of the left foot step length to the right foot step length, determine the patient's gait symmetry, and calculate the difference between the total gait symmetry of the patient in the current round and the total gait symmetry of the patient in the previous round to obtain the patient gait symmetry difference J;

[0070] The reasonable analysis module is used to receive the main configuration parameter deletion height, the main configuration parameter complexity difference, the step length fluctuation value difference, and the patient gait symmetry difference, substitute them into the data analysis model, obtain the reasonable evaluation coefficient, and send it to the deep scheduling module;

[0071] Specifically, the data analysis model refers to the calculation of the logistic regression formula;

[0072] Normalize the main configuration parameter deletion height, the main configuration parameter complexity difference, the step length fluctuation value difference, and the patient gait symmetry difference. All input variables will be transformed into the same range to ensure the balanced contribution of each input to the model. Specifically, the normalization method is to normalize through the [0,1] interval, and the specific formula is expressed as:

[0073]

[0074] In the formula, Z is the main configuration parameter deletion height, Z norm is the normalized main configuration parameter deletion height, Z min is the minimum value of the main configuration parameter deletion height, Z max is the maximum value of the main configuration parameter deletion height;

[0075] Among them, the differences in the complexity of the main configuration parameters, the differences in the step size fluctuation values, and the differences in the patient's gait symmetry are also normalized using the above formula, which will not be elaborated here;

[0076] After normalization, the values of the deletion height of the main configuration parameters, the differences in the complexity of the main configuration parameters, the differences in the step size fluctuation values, and the differences in the patient's gait symmetry are all within the range of [0, 1]. The system can compare and make decisions on a unified scale, thereby improving the accuracy and efficiency of the logistic regression model;

[0077] Substitute the deletion height of the main configuration parameters, the differences in the complexity of the main configuration parameters, the differences in the step size fluctuation values, and the differences in the patient's gait symmetry into the logistic regression formula to calculate the reasonable evaluation coefficient. The specific formula is expressed as follows:

[0078]

[0079] In the formula, L is the result of the logistic regression calculation, that is, the reasonable evaluation coefficient, e is the natural base, and y is the linear combination term of the logistic regression model. Specifically, y can be set as:

[0080]

[0081] In the formula, β0 is the bias term, and β1, β2, β3, and β4 are the regression coefficients of the deletion height of the main configuration parameters, the differences in the complexity of the main configuration parameters, the differences in the step size fluctuation values, and the differences in the patient's gait symmetry respectively;

[0082] Among them, the deletion height of the main configuration parameters, the differences in the complexity of the main configuration parameters, the differences in the step size fluctuation values, and the differences in the patient's gait symmetry are all data manifestations directly expressing the rationality of the adaptive model deletion mechanism;

[0083] It can be seen from the formula that the higher the deletion height of the main configuration parameters, the differences in the complexity of the main configuration parameters, the differences in the step size fluctuation values, and the differences in the patient's gait symmetry, the lower the rationality of the adaptive model deletion mechanism and the higher the training difficulty. On the contrary, it indicates that the rationality of the adaptive model deletion mechanism is higher and the training difficulty is more in line with the current patient's needs;

[0084] The deep scheduling module is used to receive the reasonable evaluation coefficient, compare it with the preset reasonable threshold, and based on the comparison result, perform deep scheduling mechanism processing on the adaptive model to obtain a deep scheduling score, determine the deep scheduling intensity based on the score, and send it to the scheduling iteration module;

[0085] The logic for obtaining a reasonable threshold is to collect a distribution sample set of the historical adaptive model deletion result database, divide the data set into a training set and a test set, set the evaluation index and clustering algorithm, and in each round of cross-validation, train the model on the training set and evaluate the model performance on the test set. Then, adjust the reasonable threshold based on the performance of the validation set. Therefore, the reasonable threshold is constantly iterated and updated.

[0086] In the present invention, clustering algorithm is a kind of unsupervised learning algorithm, which is used to divide the adaptive model pruning results in the data set into groups or clusters with similarities; a common one is K-means clustering, which divides the pruning results in the data set into K clusters, so that the distance between each pruning result and the center point (center of mass) of the cluster to which it belongs is minimized, and finally the pruning result is measured with the corresponding reasonable result by the Euclidean distance, so as to set a reasonable threshold;

[0087] After obtaining the reasonable evaluation coefficient, compare and analyze the reasonable evaluation coefficient with the continuously iterated reasonable threshold value;

[0088] If the reasonable evaluation coefficient is greater than or equal to the reasonable threshold, the adaptive model pruning mechanism of the current round is marked as unreasonable pruning, and a scheduling signal is generated;

[0089] If the reasonable evaluation coefficient is less than the reasonable threshold, the adaptive model pruning mechanism of the current round is marked as reasonable pruning, and an end signal is generated;

[0090] Substitute the adaptive model of the current round that is greater than the reasonable threshold into the deep scheduling mechanism for processing. The specific steps are as follows:

[0091] From the historical adaptive model deleted main configuration parameter library, analyze the adaptive model deleted main configuration parameter rounds that are most similar to the current adaptive model deleted main configuration parameters, and use the reduction in training frequency, virtual pedestrian number deletion, virtual pedestrian posture deletion, virtual environment content deletion, and target forward leaning torso angle deletion as five groups of corresponding vectors for evaluating similarity;

[0092] It should be noted that the historical adaptive model deletion main configuration parameter library refers to the database that stores the model configuration and related parameters used in the past adaptive training process. The historical data includes the parameter adjustments of different adaptive models in previous training rounds, such as training frequency, number of virtual pedestrians, virtual pedestrian posture, virtual environment content, target forward leaning torso angle, etc., which will not be elaborated here;

[0093] The frequency of reducing training, the amount of virtual pedestrians reduced, the amount of virtual pedestrian posture reduced, the amount of virtual environment content reduced, and the amount of target forward leaning trunk angle reduced are vectorized and substituted into the similarity calculation formula to obtain the adaptive model deleted main configuration parameter round that is most similar to the current adaptive model deleted main configuration parameter;

[0094] Specifically, the similarity calculation can be obtained based on the Euclidean distance or cosine similarity. The specific similarity calculation formula is prior art and will not be elaborated here;

[0095] Compare the reasonable evaluation coefficient of the current adaptive model's main configuration parameter deletion round with that of its most similar adaptive model's main configuration parameter deletion round. Subtract the smaller reasonable evaluation coefficient from the larger one to obtain the reasonable evaluation coefficient difference of the similar deletion rounds;

[0096] Furthermore, if the reasonable evaluation coefficient of the current adaptive model's main configuration parameter deletion round is the same as that of its most similar adaptive model's main configuration parameter deletion round, then the reasonable evaluation coefficient of the current adaptive model's main configuration parameter deletion round is the deep scheduling score;

[0097] Multiply the reasonable evaluation coefficient difference of the similar deletion rounds by the reasonable evaluation coefficient of the current adaptive model's main configuration parameter deletion round to obtain the deep scheduling score;

[0098] Substitute the deep scheduling score into the historical deep scheduling score library, analyze the ranking of the deep scheduling score, and obtain the deep scheduling intensity according to the ranking;

[0099] It should be noted that the historical deep scheduling score library refers to a database used to store the deep scheduling scores of each round in the past training process. This library records the adjustment results (including reasonable evaluation coefficients and adaptive configuration parameters) of all adaptive models in the past training, calculates the deep scheduling scores based on these data, analyzes the deep scheduling scores of the current training round, and compares them with the past scores to determine the optimal deep scheduling intensity;

[0100] Specifically, if the deep scheduling score value is in the top 20% in the historical deep scheduling score library, then enable the first-level deep scheduling intensity; otherwise, enable the second-level deep scheduling intensity;

[0101] Among them, both the first-level deep scheduling intensity and the second-level deep scheduling intensity represent the deletion intensity determined when the next adaptive model starts. The higher the deletion intensity, the higher the deletion degree of the training frequency, the number of virtual pedestrians, the postures of virtual pedestrians, the content of the virtual environment, and the target forward leaning torso angle;

[0102] The present invention collects the height of the main configuration parameter deletion, the difference in the complexity of the main configuration parameter, the difference in the step length fluctuation value, and the difference in the patient's gait symmetry, substitutes them into the calculation of the logistic regression formula, obtains a reasonable evaluation coefficient, compares it with a preset reasonable threshold, and based on the comparison result, performs a deep scheduling mechanism process on the adaptive model to obtain a deep scheduling score, and determines the deep scheduling intensity based on the score, reducing the training difficulty of the patient, increasing the training probability, and meeting the training experience of the patient.

[0103] Embodiment 2

[0104] In Embodiment 1 of the present invention, the height of the main configuration parameter deletion, the difference in the complexity of the main configuration parameter, the difference in the step length fluctuation value, and the difference in the patient's gait symmetry are mainly exemplified and substituted into the data analysis model to obtain a reasonable evaluation coefficient, which is compared with a preset reasonable threshold. Based on the comparison result, a deep scheduling mechanism process is performed on the adaptive model to obtain a deep scheduling score, and the operation strategy of determining the deep scheduling intensity based on the score is described; however, in Embodiment 1, only the rationality of this round is analyzed from two adjacent rounds, and the deletion intensity of the next round is determined, but for the subsequent iterative process, whether the current deletion intensity can be continuously used is not limited. Obviously, this will lead to complex calculations, an increase in the number of scheduling times, and a reduction in the operating efficiency of the system; in response to the above problems, Embodiment 2 of the present invention is further refined;

[0105] The scheduling iteration module is used to receive the deep scheduling intensity, obtain the historical adaptive iteration times by determining the number of adaptive model iterations consistent with the current deep scheduling intensity, obtain the training progress index by collecting the training data of this round, and use fuzzy logic to determine the continuous iteration result of the adaptive model;

[0106] Specifically, the deep scheduling intensity is divided into two types, namely the first-level deep scheduling intensity and the second-level deep scheduling intensity; therefore, in the historical deep scheduling score library, there are multiple adaptive models corresponding to the consistent first-level deep scheduling intensity and the second-level deep scheduling intensity;

[0107] The acquisition logic of the historical adaptive iteration times is to collect the iteration times of multiple adaptive models that improve the deletion intensity corresponding to the deep scheduling intensity, accumulate the iteration times of all adaptive models, and calculate the ratio with the number of adaptive models that improve the deletion intensity to obtain the historical adaptive iteration times;

[0108] Specifically, the above-mentioned adaptive model refers to an adaptive model with an additional deep scheduling intensity, that is, an adaptive model with an increased deletion intensity;

[0109] The acquisition logic of the training progress index is to collect the gait speed of this round minus the ideal gait speed and calculate the ratio with the ideal gait speed to obtain the training progress index;

[0110] It should be noted that gait speed is an important indicator for measuring the gait recovery of patients. As the rehabilitation training progresses, the gait speed of patients gradually approaches the normal level. The ideal gait speed is the ideal gait speed preset by the current doctor or the experimenter of this experiment for the current rehabilitation stage, and it can also be the normal gait speed;

[0111] For example, "High", "Low", "Medium" for the historical adaptive iteration times, and "Fast", "Slow", "Average" for the training progress index;

[0112] Formulate a set of fuzzy rules to describe the influence of different input variables on the output variable. The definition of the rules can be based on professional knowledge or obtained through data analysis and experiments. For example:

[0113] Mark the historical adaptive iteration times as X, the training progress index as U, and the continuous iteration result of the adaptive model as C_results;

[0114] Then it can be defined as:

[0115] Rule 1: IF(X is High) AND (U is Fast) THEN (C_results is High)

[0116] Rule 2: IF (U is Low) AND (U is Slow) THEN (C_results is Low) ...

[0118] Conduct fuzzy reasoning according to the fuzzy rules to determine the continuous iteration scheme of the adaptive model;

[0119] It should be noted that the division of the fuzzy set can be adjusted according to the actual situation. For example, although three fuzzy sets are used as examples in this embodiment, in fact, the historical adaptive iteration times and the training progress index can be divided into more than three sets to facilitate more accurate adjustment according to different iteration times;

[0120] Furthermore, for the judgment of high, medium, and low of the historical adaptive iteration times and the training progress index, thresholds can be set for judgment according to the actual situation. For example, when the historical adaptive iteration times exceed 65%, it is marked as "High", and when the training progress index is higher than 72%, it is marked as "Low", etc., which will not be elaborated here;

[0121] The present invention receives the deep scheduling intensity, determines the number of adaptive model iterations consistent with the current deep scheduling intensity to obtain the historical adaptive iteration times, collects the training data of this round to obtain the training progress index, and uses fuzzy logic to determine the continuous iteration result of the adaptive model, reducing computational complexity, reducing the scheduling times of the adaptive model, and stabilizing the operating efficiency of the system.

[0122] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0123] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0124] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0125] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0126] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0127] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0128] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0130] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0131] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the said claims.

Claims

1. The walking function assessment and training system based on augmented reality is characterized by: It includes data acquisition module, rational analysis module, deep scheduling module and scheduling iteration module; signal connection between each module; The data acquisition module is used to collect the main configuration adjustment coefficients of the adaptive main configuration parameters of two random adjacent rounds within a period of time, and obtain the main configuration adjustment state, obtain the main configuration parameter deletion height and the main configuration parameter complexity difference according to the main configuration adjustment state, and at the same time collect the patient's walking state of two adjacent rounds to obtain the step length fluctuation value difference and the patient's gait symmetry difference, and send them to the reasonable analysis module; The reasonable analysis module is used to receive the main configuration parameter deletion height, the main configuration parameter complexity difference, the step length fluctuation value difference and the patient's gait symmetry difference, substitute them into the data analysis model, obtain the reasonable evaluation coefficient, and send it to the deep scheduling module; The deep scheduling module is used to receive the reasonable evaluation coefficient and compare it with the preset reasonable threshold. According to the comparison result, the adaptive model is processed by the deep scheduling mechanism to obtain the deep scheduling score. The deep scheduling intensity is determined according to the score and sent to the scheduling iteration module. The scheduling iteration module is used to receive the deep scheduling intensity, obtain the historical number of adaptive iterations by determining the number of adaptive model iterations consistent with the current deep scheduling intensity, obtain the training progress index by collecting the current round of training data, and use fuzzy logic to determine the continuous iteration results of the adaptive model.

2. The augmented reality-based walking function assessment and training system according to claim 1, characterized in that: The main configuration adjustment status includes the main configuration parameter deletion height and the main configuration parameter complexity difference; the patient walking status of two consecutive rounds includes the step length fluctuation value difference and the patient gait symmetry difference; The data acquisition module analyzes the frequency of reduced training, the amount of virtual pedestrian reduction, the amount of virtual pedestrian posture reduction, the amount of virtual environment content reduction, and the amount of target forward leaning torso angle reduction for different training contents in two adjacent rounds, and calculates the difference between the amount of all parameters of the main configuration reduction in the current round and the amount of all parameters of the main configuration reduction in the previous round to obtain the amount of each main configuration parameter reduction, and performs weighted calculation to obtain the main configuration parameter reduction height Z; Through the geometric complexity algorithm, the number of faces of the virtual environment polyhedron objects is counted, and the ratio of the number of objects to the volume of the area where the objects are located is calculated to obtain the virtual environment space density, and the virtual environment configuration complexity is determined by weighted calculation, and the virtual environment configuration complexity of the current round is subtracted from the virtual environment configuration complexity of the previous round to obtain the main configuration parameter complexity difference P; Prioritize analyzing the patient's walking state, calculate the stride length of each step, measure the stepping distance of the left foot and the right foot, compare the stride length difference of adjacent steps in the current round, calculate the total fluctuation value of the stride length in the current round, and subtract it from the total fluctuation value of the stride length in the previous round to obtain the difference U of the stride length fluctuation value; The patient's walking state is analyzed first, and then defined as the ratio of the left foot length to the right foot length to determine the patient's gait symmetry. The difference between the total gait symmetry of the patient in the current round and the total gait symmetry of the patient in the previous round is calculated to obtain the patient's gait symmetry difference J.

3. The augmented reality-based walking function assessment and training system according to claim 2, characterized in that: Substitute the main configuration parameter deletion height, the main configuration parameter complexity difference, the step length fluctuation value difference and the patient gait symmetry difference into the logistic regression formula to calculate the reasonable evaluation coefficient. The specific formula is expressed as follows: In the formula, L is the result of logistic regression calculation, that is, the reasonable evaluation coefficient, e is the natural base, and y is the linear combination term of the logistic regression model. Specifically, y can be set as: Where β0 is the bias term, β1, β2, β3 and β4 are the regression coefficients of the main configuration parameter deletion height, the main configuration parameter complexity difference, the step length fluctuation value difference and the patient gait symmetry difference, respectively.

4. The augmented reality-based walking function assessment and training system according to claim 3, characterized in that: After obtaining the reasonable evaluation coefficient, compare and analyze the reasonable evaluation coefficient with the continuously iterated reasonable threshold value; If the reasonable evaluation coefficient is greater than or equal to the reasonable threshold, the adaptive model pruning mechanism of the current round is marked as unreasonable pruning, and a scheduling signal is generated; If the reasonable evaluation coefficient is less than the reasonable threshold, the adaptive model pruning mechanism of the current round is marked as reasonable pruning, and an end signal is generated.

5. The augmented reality-based walking function assessment and training system according to claim 4, characterized in that: Substitute the current round of adaptive models that are greater than a reasonable threshold into the deep scheduling mechanism for processing, analyze the adaptive model deletion main configuration parameter round that is most similar to the current adaptive model deletion main configuration parameter from the historical adaptive model deletion main configuration parameter library, and use the reduction in training frequency, virtual pedestrian number deletion, virtual pedestrian posture deletion, virtual environment content deletion, and target forward leaning torso angle deletion as five groups of corresponding vectors for evaluating similarity; The frequency of reducing training, the amount of virtual pedestrians reduced, the amount of virtual pedestrian posture reduced, the amount of virtual environment content reduced, and the amount of target forward leaning torso angle reduced are vectorized and substituted into the similarity calculation formula to obtain the round of adaptive model main configuration parameter reduction that is most similar to the current adaptive model main configuration parameter reduction.

6. The augmented reality-based walking function assessment and training system according to claim 5, characterized in that: Compare the reasonable evaluation coefficient of the current adaptive model's deletion of main configuration parameters with the reasonable evaluation coefficient of its most similar adaptive model's deletion of main configuration parameters, subtract the smaller reasonable evaluation coefficient from the larger reasonable evaluation coefficient, and obtain the difference in reasonable evaluation coefficients of similar deletion rounds; The difference in reasonable evaluation coefficients of similar deletion rounds is multiplied by the reasonable evaluation coefficient of the current adaptive model deletion round of main configuration parameter to obtain the deep scheduling score; Substitute the deep scheduling score into the historical deep scheduling score library, analyze the ranking of the deep scheduling score, and obtain the deep scheduling strength based on the ranking; If the deep scheduling score value is in the top 20% of the historical deep scheduling score library, the first-level deep scheduling intensity is enabled; otherwise, the second-level deep scheduling intensity is enabled.

7. The augmented reality-based walking function assessment and training system according to claim 6, characterized in that: The number of iterations of multiple adaptive models with increased pruning strength corresponding to the deep scheduling strength is collected, the number of iterations of all adaptive models is accumulated and the ratio is calculated with the number of adaptive models with increased pruning strength to obtain the historical number of adaptive iterations; The training progress index is obtained by collecting the current round gait speed minus the ideal gait speed and calculating the ratio with the ideal gait speed.

8. The augmented reality-based walking function assessment and training system according to claim 7, characterized in that: The historical adaptive iteration number and training progress index are defined as input variables and divided into different fuzzy sets respectively; The continuous iteration results of the adaptive model are defined as output variables and divided into fuzzy sets; Formulate fuzzy rules to describe the impact of historical adaptive iteration counts and training progress index on the continuous iteration results of the adaptive model; Perform fuzzy reasoning based on fuzzy rules and determine the continuous iteration plan of the adaptive model.

Images