Tactile perception control method for improving immersion of multi-element interaction scene

Abstract

The invention provides a tactile perception control method for improving the immersion of a multi-element interaction scene, and the method comprises the steps: building a load field of typical load elements and a vulnerable characteristic model of a virtual world personnel target based on the load calculation and vulnerable characteristic analysis of the interaction of real world load elements and the personnel target; according to the possible interaction relationship between the load element and the personnel target, designing and resolving the intersection calculation working condition of the load element and the personnel target, and completing the generation of an acting load data set and an acting position data set; training of the dynamic load prediction model and the action position prediction model is completed based on an ADABOOST algorithm and a decision tree machine learning algorithm, and then prediction results are evaluated and analyzed. And according to the bearing characteristics of the virtual world personnel target vulnerability characteristic model, a tactile perception control strategy is established, and mapping of virtual and real space acting loads is realized. And the strategy is deployed to various wearable tactile feedback equipment control ends, so that feedback of tactile perception of the experiencer can be realized.

Description

Technical Field

[0001] The present invention belongs to the fields of metaverse, digital twin, virtual-reality fusion and human-computer interaction technology, and relates to a tactile perception control method for enhancing the immersiveness of multi-faceted interactive scenarios. Background Art

[0002] In virtual-reality fusion fields like digital twins and the metaverse, visual and auditory perception is relatively mature, but tactile perception and feedback remain somewhat limited. With the continuous advancement of market demand and technology, some VR-based shooter games have initially attempted to use a single vibration pattern to simulate the tactile feedback of a user being struck. However, this vibration pattern fails to consider the load mechanisms and transmission relationships in the real physical world, resulting in a lack of professionalism and seriousness in these virtual-reality interaction systems. Furthermore, in scenarios like virtual-reality fusion training, multiple people need to interact with multiple elements simultaneously. The time required to calculate tactile feedback impacts the user experience of these virtual-reality interaction systems. Therefore, a tactile perception control method is urgently needed to enhance the immersiveness of multi-faceted interaction scenarios. This method, firstly, addresses the mechanical mechanism of tactile perception to improve realistic simulation and mapping feedback; secondly, it leverages machine learning techniques to achieve delay-free tactile feedback response. Summary of the Invention

[0003] Purpose of the invention: The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide a tactile perception control method for enhancing the immersiveness of multiple interactive scenarios, comprising the following steps:

[0004] Step 1: Construct a load field of a load element in a real scene; the load element contains one or more load action units, and the load field of the load element includes the mass, center of gravity position, action direction and action load of each load action unit;

[0005] Step 2: Based on the human body's response to loads and the arrangement of tactile units in a wearable tactile feedback device, a vulnerability model of a virtual world human target is constructed to output load assessment analysis at different locations; the wearable tactile feedback device includes a wearable tactile feedback device with electrode stimulation, motor vibration, gas pressure, or a combination of these modes.

[0006] Step 3: The load field of the load element in step 1 and the human target vulnerability characteristic model in step 2 are transformed into the same coordinate space according to the coordinate transformation rule; in the coordinate space, the coordinate origin is defined at the projection position of the center of gravity of the human target vulnerability characteristic model on the ground, and the human target outline size and the effective action distance of the load element are used as the range space to construct the load element and human target intersection calculation condition; the intersection calculation condition is to set N load element action points in the range space using the Monte Carlo method; at each load element action point position, according to the influence of the load element's own rotation, M load conditions are set according to the linear difference rule, that is, N*M load element and human target intersection calculation conditions are generated; M and N are integers greater than 1;

[0007] Step 4: Solve the load element and personnel target intersection calculation conditions constructed in step 3, and construct an applied load data set and an applied position data set. For the load element and personnel target intersection calculation conditions constructed in step 3, respectively solve the dynamic displacement response of the personnel target in the process of being acted upon by the load element and the position information of the load element acting on the personnel target in each condition, and integrate all the calculation results to form an applied load data set and an applied position data set.

[0008] Step 5: Build a dynamic load prediction model: Preprocess the applied load dataset and divide it into a training sample set and a test sample set. Use the training sample set as input to train the dynamic load prediction model based on the ADABOOST algorithm. The prediction results are evaluated and analyzed using the absolute mean error, root mean square error, and coefficient of determination.

[0009] Step 6: Constructing an action location prediction model: Preprocess the action location dataset and divide it into a training sample set and a test sample set. Using the training sample set as input, the action location prediction model is trained based on a decision tree algorithm. The prediction results are evaluated and analyzed using the absolute mean error, root mean square error, and coefficient of determination.

[0010] Step 7: Establish a tactile perception control strategy. The dynamic load prediction model of step 5 and the action position prediction model of step 6 realize the relationship mapping of the action load in virtual and real space, and realize feedback of the experiencer's tactile perception by driving the wearable tactile feedback device. The tactile perception control strategy includes feedback action load and feedback action position.

[0011] In step 1, the load field of the load element covers multi-source data such as theoretical derivation, finite element numerical simulation, physical experimental testing and prior historical information.

[0012] Step 4 includes:

[0013] In step 4-1, it is assumed that the typical load element contains W acting load units, where the j-th acting load unit applies a load to the virtual world human target vulnerability model at the i-th load point. The specific solution method includes:

[0014] Calculate the dynamic displacement response. The dynamic displacement response calculation formula of the personnel target vulnerability characteristic model after being acted upon by the load action unit is:

[0015]

[0016] Where d represents the differential sign, Z represents the displacement response of the virtual world human target vulnerability model during loading, and m j is the mass of the jth load acting unit, V z is the impact velocity of the load acting unit; σ e is the equivalent compressive linear elastic limit of the virtual world human target vulnerability model under quasi-static conditions, β is the influence factor of the load element shape, S i represents the contact area under the action point of the i-th load action unit; ρ represents the equivalent density of the target action position of the personnel;

[0017] Calculate the action position of the load action unit; based on the initial action direction information of the load action unit and environmental factors, modify the action direction of the load action unit, and determine the intersection position of the interaction with the personnel target vulnerability characteristic model based on the radiographic detection principle;

[0018] Step 4-2: Repeat step 4-1 to complete the calculation of the intersection of load elements and personnel targets, and summarize the calculation results into the action load data set and action position data set.

[0019] Step 5 includes:

[0020] Step 5-1: Data preparation and preprocessing: clean the load data set, handle missing values and outliers, normalize the data, obtain data features on the same scale, and then randomly divide it into training sample set and test sample set;

[0021] Step 5-2: Determine the input, including the labeled training sample set, the self-learning algorithm, and the maximum number of iterations;

[0022] Step 5-3, parameter initialization, the initial weight of each training sample is set to be equal, that is, the initial weight is 1 / N um , where N um is the total number of samples;

[0023] Step 5-4, train the weak classifier, give the sample weight, calculate the classification error rate, and calculate the weight α of the weak classifier based on the classification error ratet :

[0024]

[0025] where error t Represents the error rate of the current weak classifier;

[0026] Step 5-5, update the sample weights, update the sample weights according to the prediction results of the current weak classifier, and the update rules are:

[0027]

[0028] in, is the weight of training sample i in round t, y i is the true label of training sample i, ht(xi) is the prediction result of the weak classifier in round t, x i is the characteristic value of training sample i; exp is the natural exponential function;

[0029] Step 5-5, build a strong classifier. In each round of training, a weak classifier and its weight α are obtained. t , the final strong classifier is the weighted voting result of all weak classifiers, and the formula is:

[0030]

[0031] Where H(x) is the final prediction of the strong classifier and T is the number of weak classifiers;

[0032] Steps 5-6, model evaluation.

[0033] Steps 5-6 include: using the test sample set for prediction, using the absolute mean error, root mean square error, and coefficient of determination to evaluate the model performance of the prediction results, and obtaining the final prediction model. The dynamic load prediction model expression is:

[0034] P=f(A0,m j ,μ,t n ,σ e )

[0035] Where P is the dynamic displacement response generated by the load unit, A0 is the cross-sectional area of the load unit, μ is the friction coefficient, and t n is the action time;

[0036] The trained dynamic load prediction model can be used for actual prediction tasks.

[0037] Step 6 includes:

[0038] Step 6-1: Data preparation and preprocessing: clean the action position dataset, handle missing values and outliers, normalize the data to obtain data features on the same scale, and then randomly divide it into training sample set and test sample set;

[0039] Step 6-2: The location data belongs to discrete classes. Select the classification decision tree model and train the model using the training sample set.

[0040] Step 6-3: Adjust the parameters to optimize the maximum depth of the tree (max_depth), the minimum number of samples for internal node re-splitting (min_samples_split), and the minimum number of samples on leaf nodes (min_samples_leaf). (In the tuning process, key parameters such as the maximum depth and the minimum number of samples on leaf nodes are adjusted, and optimization is performed through grid search. The grid search algorithm belongs to the existing technology.)

[0041] Step 6-4, model evaluation.

[0042] Step 6-4 includes: evaluating the accuracy of the model by calculating the accuracy, precision, and recall of the training model to obtain the final action position prediction model. Taking position prediction as an example: there are 92 actual target points in 1000 position points, the model correctly identifies 90 of them (misses 10), and misjudges 10 non-target points. The accuracy is 90% (90 / 100) and the recall is 90% (90 / 100), reaching the preset threshold (precision ≥ 85%, recall ≥ 85%), that is, the model accuracy meets the requirements. The trained action position prediction model can be used for actual prediction tasks.

[0043] Step 7 includes:

[0044] Step 7-1: Determine the tactile feedback action response information; perform feature analysis based on the dynamic displacement response of the personnel target vulnerability characteristic model, and determine the output excitation range of the tactile unit in the wearable tactile feedback device based on personnel safety and tactile perception tolerance. The tactile perception control strategy is shown in the following formula:

[0045]

[0046] Among them F n is the feedback load of the nth tactile unit in the wearable tactile feedback device, is the amplitude of the output feedback load, ω n is the frequency of the output feedback load, t n is the action time of the output feedback load;

[0047] The amplitude of the output feedback load and the frequency ω of the output feedback load nThe method for obtaining is as follows: perform the empirical mode decomposition (EMD) method on the dynamic displacement response P of the load-acting unit output by the dynamic load prediction model in steps 5-6, remove the signal components caused by non-impact loads such as high frequency and noise, perform fast Fourier transform (FT) analysis on the low-frequency signal data, and obtain the amplitude and frequency corresponding to the low-order frequency of the dynamic displacement response P;

[0048] Step 7-2, determining the tactile feedback action position information: calling the action position prediction model obtained in step 6, and outputting the ID information of the wearable tactile feedback device triggering the vibration feedback unit;

[0049] Step 7-3: driving the wearable tactile feedback device to complete the tactile perception feedback according to the tactile perception feedback action response information and the tactile perception feedback action position information.

[0050] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the method.

[0051] The present invention also provides a storage medium storing a computer program or instruction, which executes the steps of the method when the computer program or instruction is run on a computer.

[0052] The present invention has the following beneficial effects:

[0053] (1) The present invention provides a tactile perception control method for enhancing the immersiveness of multi-faceted interactive scenarios. Based on the solution of interaction loads between real-world objects and the analysis of the vulnerability characteristics of human targets, a load / target intersection prediction model is constructed by applying a variety of machine learning algorithms. The load and position of the human targets in the scenario are solved and output in real time. Based on the established tactile perception control strategy, the drive control and management of different types of wearable tactile perception feedback devices can be realized to meet the application of tactile perception in virtual-reality fusion interactive systems.

[0054] (2) The load action intersection prediction model constructed by the present invention covers a dynamic load prediction model and an action position prediction model. It can use the type of load field generated by the element, the arbitrary relative position with the personnel target, and the physical world environment impact parameters as input, and can solve and output the information such as the action load and action position borne by the personnel target in the scene in real time, which can meet the timeliness requirements of the virtual-reality interaction system.

[0055] (3) The construction of the position intersection and personnel injury database of the present invention is based on theoretical analysis, finite element numerical simulation, experimental testing and other means, and relies on target vulnerability analysis technology. The relevant data has high fidelity, which provides strong support for the seriousness and professionalism of the virtual-reality fusion system. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more apparent.

[0057] Figure 1 It is a flow chart of the method of the present invention.

[0058] Figure 2 It is a schematic diagram of the vulnerability characteristic model of personnel targets in the virtual world.

[0059] Figure 3 It is a schematic diagram of the calculation condition where load elements and personnel targets intersect.

[0060] Figure 4 It is a schematic diagram of the interaction between the load action unit and the personnel target.

[0061] Figure 5 、 Figure 6 、 Figure 7 This is a schematic diagram of the output results of the action position prediction model.

[0062] Figure 8 It is a tactile perception control strategy mapping diagram. DETAILED DESCRIPTION

[0063] An embodiment of the present invention provides a tactile perception control method for enhancing immersion in a multi-faceted interactive scenario, comprising the following steps:

[0064] Step 1: Construct a load field of a load element in a real scene; the load element contains one or more load action units, and the load field of the load element includes the mass, center of gravity position, action direction and action load of each load action unit;

[0065] Step 2: Based on the human body's response to loads and the arrangement of tactile units in a wearable tactile feedback device, a vulnerability model of a virtual world human target is constructed to output load assessment analysis at different locations; the wearable tactile feedback device includes a wearable tactile feedback device with electrode stimulation, motor vibration, gas pressure, or a combination of these modes.

[0066] Step 3: The load field of the load element in step 1 and the human target vulnerability characteristic model in step 2 are transformed into the same coordinate space according to the coordinate transformation rule; in the coordinate space, the coordinate origin is defined at the projection position of the center of gravity of the human target vulnerability characteristic model on the ground, and the human target outline size and the effective action distance of the load element are used as the range space to construct the load element and human target intersection calculation condition; the intersection calculation condition is to set N load element action points in the range space using the Monte Carlo method; at each load element action point position, according to the influence of the load element's own rotation, M load conditions are set according to the linear difference rule, that is, N*M load element and human target intersection calculation conditions are generated; M and N are integers greater than 1;

[0067] Step 4: Solve the load element and personnel target intersection calculation conditions constructed in step 3, and construct an applied load data set and an applied position data set. For the load element and personnel target intersection calculation conditions constructed in step 3, respectively solve the dynamic displacement response of the personnel target in the process of being acted upon by the load element and the position information of the load element acting on the personnel target in each condition, and integrate all the calculation results to form an applied load data set and an applied position data set.

[0068] Step 5: Build a dynamic load prediction model: Preprocess the applied load dataset and divide it into a training sample set and a test sample set. Use the training sample set as input to train the dynamic load prediction model based on the ADABOOST algorithm. The prediction results are evaluated and analyzed using the absolute mean error, root mean square error, and coefficient of determination.

[0069] Step 6: Constructing an action location prediction model: Preprocess the action location dataset and divide it into a training sample set and a test sample set. Using the training sample set as input, the action location prediction model is trained based on a decision tree algorithm. The prediction results are evaluated and analyzed using the absolute mean error, root mean square error, and coefficient of determination.

[0070] Step 7: Establish a tactile perception control strategy. The dynamic load prediction model of step 5 and the action position prediction model of step 6 realize the relationship mapping of the action load in virtual and real space, and realize feedback of the experiencer's tactile perception by driving the wearable tactile feedback device. The tactile perception control strategy includes feedback action load and feedback action position.

[0071] In step 1, the load field of the load element covers multi-source data such as theoretical derivation, finite element numerical simulation, physical experimental testing and prior historical information.

[0072] Step 4 includes:

[0073] In step 4-1, it is assumed that the typical load element contains W acting load units, where the j-th acting load unit applies a load to the virtual world human target vulnerability model at the i-th load point. The specific solution method includes:

[0074] Calculate the dynamic displacement response. The dynamic displacement response calculation formula of the personnel target vulnerability characteristic model after being acted upon by the load action unit is:

[0075]

[0076] Where d represents the differential sign, Z represents the displacement response of the virtual world human target vulnerability model during loading, and m j is the mass of the jth load acting unit, V z is the impact velocity of the load acting unit; σ e is the equivalent compressive linear elastic limit of the virtual world human target vulnerability model under quasi-static conditions, β is the influence factor of the load element shape, S i represents the contact area under the action point of the i-th load action unit; ρ represents the equivalent density of the target action position of the personnel;

[0077] Calculate the action position of the load action unit; based on the initial action direction information of the load action unit and environmental factors, modify the action direction of the load action unit, and determine the intersection position of the interaction with the personnel target vulnerability characteristic model based on the radiographic detection principle;

[0078] Step 4-2, repeat step 4-1 to complete the calculation of the intersection of load elements and personnel targets, and summarize the calculation results into the action load data set and action position data set.

[0079] Step 5 includes:

[0080] Step 5-1: Data preparation and preprocessing: clean the load data set, handle missing values and outliers, normalize the data, obtain data features on the same scale, and then randomly divide it into training sample set and test sample set;

[0081] Step 5-2: Determine the input, including the labeled training sample set, the self-learning algorithm, and the maximum number of iterations;

[0082] Step 5-3, parameter initialization, the initial weight of each training sample is set to be equal, that is, the initial weight is 1 / N um , where N um is the total number of samples;

[0083] Step 5-4, train the weak classifier, give the sample weight, calculate the classification error rate, and calculate the weight α of the weak classifier based on the classification error ratet :

[0084]

[0085] where error t Represents the error rate of the current weak classifier;

[0086] Step 5-5, update the sample weights, update the sample weights according to the prediction results of the current weak classifier, and the update rules are:

[0087]

[0088] in, is the weight of training sample i in round t, y i is the true label of training sample i, ht(xi) is the prediction result of the weak classifier in round t, x i is the characteristic value of training sample i; exp is the natural exponential function;

[0089] Step 5-5, build a strong classifier. In each round of training, a weak classifier and its weight α are obtained. t , the final strong classifier is the weighted voting result of all weak classifiers, and the formula is:

[0090]

[0091] Where H(x) is the final prediction of the strong classifier and T is the number of weak classifiers;

[0092] Steps 5-6, model evaluation.

[0093] Steps 5-6 include: using the test sample set for prediction, using the absolute mean error, root mean square error, and coefficient of determination to evaluate the model performance of the prediction results, and obtaining the final prediction model. The dynamic load prediction model expression is:

[0094] P=f(A0,m j ,μ,t n ,σ e )

[0095] Where P is the dynamic displacement response generated by the load unit, A0 is the cross-sectional area of the load unit, μ is the friction coefficient, and t n is the action time;

[0096] The trained dynamic load prediction model can be used for actual prediction tasks.

[0097] Step 6 includes:

[0098] Step 6-1: Data preparation and preprocessing: clean the action position dataset, handle missing values and outliers, normalize the data to obtain data features on the same scale, and then randomly divide it into training sample set and test sample set;

[0099] Step 6-2: The location data belongs to discrete classes. Select the classification decision tree model and train the model using the training sample set.

[0100] Step 6-3: Adjust the parameters to optimize the maximum depth of the tree (max_depth), the minimum number of samples for internal node re-splitting (min_samples_split), and the minimum number of samples on leaf nodes (min_samples_leaf). (In the tuning process, key parameters such as the maximum depth and the minimum number of samples on leaf nodes are adjusted, and optimization is performed through grid search. The grid search algorithm belongs to the existing technology.)

[0101] Step 6-4, model evaluation.

[0102] The trained action position prediction model can be used for actual prediction tasks.

[0103] Step 6-4 involves evaluating the model's accuracy by calculating the trained model's accuracy, precision, and recall, resulting in the final action location prediction model. For example, consider location prediction: out of 1,000 locations, there are 92 actual target points. The model correctly identifies 90 of these points (missing 10) and misidentifies 10 as non-target points. This results in a precision of 90% (90 / 100) and a recall of 90% (90 / 100), meeting the preset thresholds (precision ≥ 85%, recall ≥ 85%), indicating that the model meets the required accuracy.

[0104] Step 7 includes:

[0105] Step 7-1: Determine the tactile feedback action response information; perform feature analysis based on the dynamic displacement response of the personnel target vulnerability characteristic model, and determine the output excitation range of the tactile unit in the wearable tactile feedback device based on personnel safety and tactile perception tolerance. The tactile perception control strategy is shown in the following formula:

[0106]

[0107] Among them F n is the feedback load of the nth tactile unit in the wearable tactile feedback device, is the amplitude of the output feedback load, ω n is the frequency of the output feedback load, t n is the action time of the output feedback load;

[0108] The amplitude of the output feedback load and the frequency ω of the output feedback load n The method for obtaining is as follows: perform the empirical mode decomposition (EMD) method on the dynamic displacement response P of the load-acting unit output by the dynamic load prediction model in steps 5-6, remove the signal components caused by non-impact loads such as high frequency and noise, perform fast Fourier transform (FT) analysis on the low-frequency signal data, and obtain the amplitude and frequency corresponding to the low-order frequency of the dynamic displacement response P;

[0109] Step 7-2, determining the tactile feedback action position information: calling the action position prediction model obtained in step 6, and outputting the ID information of the wearable tactile feedback device triggering the vibration feedback unit;

[0110] Step 7-3: driving the wearable tactile feedback device to complete the tactile perception feedback according to the tactile perception feedback action response information and the tactile perception feedback action position information.

[0111] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the method.

[0112] The present invention also provides a storage medium storing a computer program or instruction, which executes the steps of the method when the computer program or instruction is run on a computer.

[0113] like Figure 1 As shown, another specific embodiment of the present invention provides a tactile perception control method for enhancing the immersiveness of a multi-faceted interactive scene. A typical load element contains 600 load action units, and the load element action range is approximately 0 to 6 meters. The specific method includes:

[0114] Step 1: With the geometric center of the load element as the coordinate origin, the CAE numerical simulation method is used to calculate and solve the scattering characteristic information of each load action unit within the time range of 0 to 50 μs (time step is 1 μs). The characteristic information includes the mass of each load action unit and the velocity, acceleration and center of gravity position at each step.

[0115] Step 2, such as Figure 2 As shown, the wearable haptic feedback device used is a vest-style feedback device that deploys 92 vibration motors as tactile units. The human vulnerability model is divided into seven regions based on the head, limbs, and front and back chest. The two front and back chest regions are further divided based on the deployment of the 92 tactile units. The coordinate origin of this virtual world human target vulnerability model is set to the projection of the model's center of gravity on the ground.

[0116] Step 3: Map the load elements in step 1 to the coordinate space of the virtual world personnel target vulnerability model, and define the projection of the center of gravity of the virtual world personnel target vulnerability model in step 2 on the ground as the coordinate origin; use the personnel outline size and the effective action distance of the load as the range space, and use the Monte Carlo method to set 36,860 load element action points in the range space; at each element load action point position, each load action point position has 100 calculation conditions that rotate around the local coordinate X, Y, and Z axes, generating a total of 3,686,000 load action point calculation conditions, as shown in the following figure: Figure 3 shown.

[0117] Step 4: Calculate the dynamic displacement response of the 600 load elements acting on the virtual world human target vulnerability model using the dynamic displacement response calculation formula from Step 4-1. Follow Steps 4-1 and 4-2 to solve the intersection of each load element and the human target at each of the 36,860 load element action points.

[0118] Taking the calculation condition where the center of gravity coordinates of the virtual world personnel target vulnerability model are <0, 0, 95>, the starting position coordinates of the load element are <0, 0, 120>, and the rotation angle around the load element's own coordinates is <0, 50°, 0> as an example, the calculation results of this condition can be solved based on the compiled calculation program. Figure 4 shown.

[0119] According to steps 5 to 7, the dynamic displacement prediction model of the load action unit, the action position prediction model and the tactile perception control strategy are constructed. The effectiveness of this method is verified by arbitrarily inputting the coordinates of the load element action point, posture information and personnel target position information to obtain the dynamic displacement response, action position prediction information and tactile perception control results generated by the interaction between the load unit and the personnel vulnerability. Figure 5 、 Figure 6 、 Figure 7 shown.

[0120] like Figure 8As shown, the tactile perception control method of the present invention can be used for driving and controlling various types of wearable tactile feedback devices such as electrode stimulation, motor vibration, gas pressure and combination modes. Through the tactile perception control strategy, the load action type, action time, action intensity and combat frequency are mapped in sequence to the stimulation mode, microcurrent intensity, stimulation frequency and duration of the electrode stimulation type tactile unit; the vibration waveform, vibration amplitude, vibration frequency and vibration time of the motor vibration type tactile unit; the inflation mode, pressure intensity, pressurization frequency and punching time of the gas pressure type tactile unit. The combination mode can also be a combination mode of any two or three of the electrode stimulation type, motor vibration type and gas pressure type. In addition, the method supports the solution of tactile perception feedback when multiple elements (such as environmental elements, threat elements, multiple collision effects, etc.) act on the human target at the same time. Under the premise of ensuring a certain prediction accuracy, setting the solution time to within 50ms can meet the real-time requirements of the virtual-reality interaction system, significantly improving the solution time of the existing technology.

[0121] The present invention provides a tactile perception control method for enhancing immersion in multi-faceted interactive scenarios. While there are numerous methods and approaches for implementing this technical solution, the foregoing description represents only a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Any components not specified in this embodiment may be implemented using existing technologies.

Claims

1. A tactile perception control method for enhancing immersion in multi-faceted interactive scenarios, characterized in that: The following steps are involved: Step 1: Construct a load field of a load element in a real scene; the load element contains one or more load action units, and the load field of the load element includes the mass, center of gravity position, action direction and action load of each load action unit; Step 2: Based on the human body's response to loads and the arrangement of tactile units in a wearable tactile feedback device, a vulnerability model of a virtual world human target is constructed to output load assessment analysis at different locations; the wearable tactile feedback device includes a wearable tactile feedback device with electrode stimulation, motor vibration, gas pressure, or a combination of these modes. Step 3: The load field of the load element described in Step 1 and the human target vulnerability characteristic model described in Step 2 are converted into the same coordinate space according to the coordinate transformation rule; in the coordinate space, the coordinate origin is defined at the projection position of the center of gravity of the human target vulnerability characteristic model on the ground, and the human target outline size and the effective action distance of the load element are used as the range space to construct the load element and human target intersection calculation condition; the intersection calculation condition is to set N load element action points in the range space using the Monte Carlo method; At each load element action point, according to the influence of the load element's own rotation, M load conditions are set according to the linear difference law, that is, N*M load elements and personnel target intersection calculation conditions are generated; M and N are integers greater than 1; Step 4: Solve the load element and personnel target intersection calculation conditions constructed in step 3, and construct an applied load data set and an applied position data set. For the load element and personnel target intersection calculation conditions constructed in step 3, respectively solve the dynamic displacement response of the personnel target in the process of being acted upon by the load element and the position information of the load element acting on the personnel target in each condition, and integrate all the calculation results to form an applied load data set and an applied position data set. Step 5: Build a dynamic load prediction model: Preprocess the applied load dataset and divide it into a training sample set and a test sample set. Use the training sample set as input to train the dynamic load prediction model based on the ADABOOST algorithm. The prediction results are evaluated and analyzed using the absolute mean error, root mean square error, and coefficient of determination. Step 6: Constructing an action location prediction model: Preprocess the action location dataset and divide it into a training sample set and a test sample set. Using the training sample set as input, the action location prediction model is trained based on a decision tree algorithm. The prediction results are evaluated and analyzed using the absolute mean error, root mean square error, and coefficient of determination. Step 7: Establish a tactile perception control strategy. The dynamic load prediction model of step 5 and the action position prediction model of step 6 realize the relationship mapping of the action load in virtual and real space, and realize feedback of the experiencer's tactile perception by driving the wearable tactile feedback device. The tactile perception control strategy includes feedback action load and feedback action position.

2. The method according to claim 1, characterized in that In step 1, the load field of the load element covers multi-source data such as theoretical derivation, finite element numerical simulation, physical experimental testing and prior historical information.

3. The method according to claim 2, characterized in that Step 4 includes: In step 4-1, it is assumed that the typical load element contains W acting load units, where the j-th acting load unit applies a load to the virtual world human target vulnerability model at the i-th load point. The specific solution method includes: Calculate the dynamic displacement response. The dynamic displacement response calculation formula of the personnel target vulnerability characteristic model after being acted upon by the load action unit is: Where d represents the differential sign, Z represents the displacement response of the virtual world human target vulnerability model during loading, and m j is the mass of the jth load acting unit, V z is the impact velocity of the load acting unit; σ e is the equivalent compressive linear elastic limit of the virtual world human target vulnerability model under quasi-static conditions, β is the influence factor of the load element shape, S i represents the contact area under the action point of the i-th load action unit; ρ represents the equivalent density of the target action position of the personnel; Calculate the action position of the load action unit; based on the initial action direction information of the load action unit and environmental factors, modify the action direction of the load action unit, and determine the intersection position of the interaction with the personnel target vulnerability characteristic model based on the radiographic detection principle; Step 4-2, repeat step 4-1 to complete the calculation of the intersection of load elements and personnel targets, and summarize the calculation results into the action load data set and action position data set.

4. The method according to claim 3, characterized in that Step 5 includes: Step 5-1: Data preparation and preprocessing: clean the load data set, handle missing values and outliers, normalize the data, obtain data features on the same scale, and then randomly divide it into training sample set and test sample set; Step 5-2: Determine the input, including the labeled training sample set, the self-learning algorithm, and the maximum number of iterations; Step 5-3, parameter initialization, the initial weight of each training sample is set to be equal, that is, the initial weight is 1 / N um , where N um is the total number of samples; Step 5-4, train the weak classifier, give the sample weight, calculate the classification error rate, and calculate the weight α of the weak classifier based on the classification error rate t : where error t Represents the error rate of the current weak classifier; Step 5-5, update the sample weights, update the sample weights according to the prediction results of the current weak classifier, and the update rules are: in, is the weight of training sample i in round t, y i is the true label of training sample i, ht(xi) is the prediction result of the weak classifier in round t, x i is the characteristic value of training sample i; exp is the natural exponential function; Step 5-5, build a strong classifier. In each round of training, a weak classifier and its weight α are obtained. t , the final strong classifier is the weighted voting result of all weak classifiers, and the formula is: Where H(x) is the final prediction of the strong classifier and T is the number of weak classifiers; Steps 5-6, model evaluation.

5. The method according to claim 4, characterized in that Steps 5-6 include: using the test sample set for prediction, using the absolute mean error, root mean square error, and coefficient of determination to evaluate the model performance of the prediction results, and obtaining the final prediction model. The dynamic load prediction model expression is: P=f(A0,m j ,m,t n ,s e ) Where P is the dynamic displacement response generated by the load unit, A0 is the cross-sectional area of the load unit, μ is the friction coefficient, and t n is the action time; The trained dynamic load prediction model can be used for actual prediction tasks.

6. The method according to claim 5, characterized in that Step 6 includes: Step 6-1: Data preparation and preprocessing: clean the action position dataset, handle missing values and outliers, normalize the data to obtain data features on the same scale, and then randomly divide it into training sample set and test sample set; Step 6-2: The location data belongs to discrete classes. Select the classification decision tree model and train the model using the training sample set. Step 6-3: Adjust the parameters to optimize the maximum depth of the tree (max_depth), the minimum number of samples for internal node re-splitting (min_samples_split), and the minimum number of samples on the leaf node (min_samples_leaf). Step 6-4, model evaluation.

7. The method according to claim 6, characterized in that Step 6-4 includes: evaluating the accuracy of the model by calculating the accuracy, precision, and recall rate of the training model to obtain the final action position prediction model.

8. The method according to claim 7, characterized in that Step 7 includes: Step 7-1: Determine the tactile feedback action response information; perform feature analysis based on the dynamic displacement response of the personnel target vulnerability characteristic model, and determine the output excitation range of the tactile unit in the wearable tactile feedback device based on personnel safety and tactile perception tolerance. The tactile perception control strategy is shown in the following formula: Among them F n is the feedback load of the nth tactile unit in the wearable tactile feedback device, is the amplitude of the output feedback load, ω n is the frequency of the output feedback load, t n is the action time of the output feedback load; The amplitude of the output feedback load and the frequency ω of the output feedback load n The method for obtaining is as follows: perform empirical mode decomposition on the dynamic displacement response P of the load action unit output by the dynamic load prediction model in steps 5-6, remove the signal components caused by non-impact loads, perform fast Fourier transform analysis on the low-frequency signal data, and obtain the amplitude and frequency corresponding to the low-order frequency of the dynamic displacement response P; Step 7-2, determining the tactile feedback action position information: calling the action position prediction model obtained in step 6, and outputting the ID information of the wearable tactile feedback device triggering the vibration feedback unit; Step 7-3: driving the wearable tactile feedback device to complete the tactile perception feedback according to the tactile perception feedback action response information and the tactile perception feedback action position information.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is run on a computer, the steps of the method according to any one of claims 1 to 8 are executed.

Images