Intelligent VR rehabilitation scheme making method and system based on scene adaptive regulation and control

Abstract

The invention provides an intelligent VR rehabilitation scheme making method and system based on scene adaptive regulation and control, and belongs to the technical field of VR rehabilitation training. Comprising the steps of collecting multi-dimensional physiological data of a human body, and performing time alignment; dynamically modeling the time sequence relevance of the multi-dimensional physiological data by adopting a time sequence attention fusion network, generating a multi-dimensional state vector, and packaging the multi-dimensional state vector into a JSON protocol packet; analyzing the JSON protocol packet based on a hierarchical decision-making architecture and executing scene self-adaptive regulation and control; a scene library is driven to be switched among different scenes, and a dynamic rehabilitation scheme matched with the user neural network is generated; physiology and scene real-time coupling is achieved through a cross-modal mapping engine, and quantitative tracking of the rehabilitation process of the user and dynamic optimization of the rehabilitation scheme are achieved. On the basis of the structure of the integrated VR equipment, real-time correspondence between the physiology of the user and the scene is realized through the implementation decision of the multi-source heterogeneous data, and the personalized adaptation level of VR rehabilitation training is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the field of VR rehabilitation training technology, and in particular relates to an intelligent VR rehabilitation program formulation method and system based on scene adaptive regulation. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, virtual reality (VR) technology has demonstrated significant potential in rehabilitation medicine, particularly in neurorehabilitation, motor function recovery, and psychotherapy. Studies have demonstrated that immersive environments can enhance patient motivation for rehabilitation training and neuroplasticity through visual-motor coupling mechanisms. However, current mainstream commercial VR devices (such as the Oculus Quest series) exhibit significant technical deficiencies in real-time biosignal monitoring and dynamic scene adaptation. Existing systems often rely on single-modality behavioral data (such as controller manipulation or head movement trajectories) and lack continuous quantitative analysis of the patient's physiological state (such as EEG activity and autonomic nervous system responses). This results in the inability to optimize training scenarios in real time based on the user's cognitive load or emotional stress, potentially leading to a mismatch between training intensity and patient tolerance.

[0004] In terms of biosignal acquisition technology, existing research attempts to integrate sensors such as EEG and heart rate variability with VR devices, but is limited by hardware form and data fusion bottlenecks. For example, traditional EEG acquisition devices require wearing a multi-electrode headset, which seriously affects the wearing comfort of VR devices; and wrist-worn heart rate monitors have motion artifact interference, making it difficult to ensure signal quality during dynamic rehabilitation training. In addition, existing systems mostly use a time-sharing data processing architecture, which leads to significant deviations in the time-scale alignment of EEG, heart rate, and eye movement signals, making it impossible to meet the real-time requirements of multimodal neural feedback, greatly limiting the response speed of the closed-loop control system.

[0005] In the field of eye tracking technology, solutions based on traditional RGB cameras suffer from inherent limitations in latency and computational overhead, making it difficult to simultaneously achieve high-precision gaze point detection and dynamic rendering optimization in complex VR scenes. While event cameras offer theoretical advantages for eye movement feature extraction due to their ultra-fast response and dynamic visual sensing capabilities, existing solutions have yet to effectively coordinate the asynchronous data flow of event cameras with the real-time rendering architecture of VR engines, and even less practical application in combining them with multi-physiological signal modeling.

[0006] In summary, existing VR rehabilitation systems still face significant technical barriers in terms of hardware integration, multimodal data fusion efficiency, and scene adaptation capabilities, namely: 1) In existing VR-based rehabilitation training methods, scene adjustment lags behind changes in the user's physiological state.

[0007] 2) The physiological sensing module is physically separated from the VR device, which destroys the immersive experience and reduces data synchronization accuracy; 3) Single sensor data is difficult to fully represent the user's cognitive-emotional-motor complex state. Summary of the Invention

[0008] To overcome the shortcomings of the above-mentioned existing technologies, the present invention provides an intelligent VR rehabilitation program formulation method and system based on scene adaptive regulation. Based on the structure of an integrated VR device, through the implementation decision-making of multi-source heterogeneous data, real-time correspondence between user physiology and scene is achieved, greatly improving the personalized adaptation level of VR rehabilitation training.

[0009] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of the present invention provides a method for formulating an intelligent VR rehabilitation program based on scene adaptive regulation.

[0010] The method for formulating an intelligent VR rehabilitation program based on scene adaptive regulation includes: Collect multi-dimensional physiological data of the human body and perform time alignment on the obtained multi-dimensional physiological data; A temporal attention fusion network is used to dynamically model the temporal correlation of the multi-dimensional physiological data to generate a standardized multi-dimensional state vector; and the multi-dimensional state vector is encapsulated as a JSON protocol package; Parse the JSON protocol package based on a hierarchical decision-making architecture and perform scenario-adaptive control. Specifically, the threshold is initialized based on whether baseline data exists in the parsed content. A sliding time window is used to adjust the initialized threshold in real time, and the threshold range is updated based on historical rehabilitation efficacy coefficients. The drive scene library switches between different scenes and synchronously adjusts the motion trajectory of virtual objects and the sound and light conditions to generate a dynamic rehabilitation plan adapted to the user's neural network; at the same time, through the cross-modal mapping engine, real-time coupling of physiology and scenes is achieved, realizing quantitative tracking of the user's rehabilitation process and dynamic optimization of the rehabilitation plan.

[0011] Furthermore, the temporal attention fusion network is based on a lightweight Transformer encoder architecture and dynamically models the temporal correlation of multi-dimensional physiological data through a multi-head self-attention mechanism.

[0012] Furthermore, the temporal attention fusion network includes a temporal embedding layer, a scene adaptive weight allocation layer and a feature fusion output layer.

[0013] Furthermore, threshold initialization is performed based on whether baseline data exists in the parsed content, including: when baseline data does not exist, calling the scene complexity model to dynamically correct the preset threshold; the scene complexity score is calculated by weighted calculation of object movement speed, sound field density and light and shadow change frequency.

[0014] Furthermore, real-time coupling of physiology and scenarios is achieved through a cross-modal mapping engine, including: first, cross-modal mapping; then, based on the extended modal mapping, a hierarchical transition protocol is adopted, combined with a time series database to generate a cross-scenario physiological indicator evolution map, and the phased transition of the user's rehabilitation effectiveness is visualized through a three-dimensional scatter plot, realizing quantitative tracking of the user's rehabilitation process and dynamic optimization of the rehabilitation plan.

[0015] Furthermore, the hierarchical transition protocol includes two levels: mild over-limit and severe over-limit. In the case of mild over-limit, the operation to be performed is to locally adjust the light source and sound field; in the case of severe over-limit, the operation to be performed is to globally switch the scene type.

[0016] The second aspect of the present invention provides an intelligent VR rehabilitation program formulation system based on scene adaptive regulation.

[0017] An intelligent VR rehabilitation program formulation system based on scene adaptive regulation, including: a VR headset and a flexible biosensor scaffold; The flexible biosensor bracket is integrated into the forehead area of ​​the VR headset and is detachable. Furthermore, the flexible biosensor bracket is embedded with a gold-plated dry electrode array and integrated with an optical heart rate sensor and a skin charge sensor via a flexible circuit board. An event-driven image sensor is embedded on both sides of the nose pad at the lower edge of the VR headset. The gold-plated dry electrode array, optical heart rate sensor and skin charge sensor constitute a multimodal physiological sensing module; the multimodal physiological sensing module is directly connected to a multi-core processor through a low-latency interface, and the multi-core processor integrates a data fusion processing module, a scene adaptive control module and a user feedback module.

[0018] Furthermore, the multimodal physiological perception module is configured to: collect multi-dimensional physiological data of the human body; The data fusion processing module is configured to: perform time alignment on the obtained multi-dimensional physiological data and send the multi-dimensional physiological data after data fusion processing to the cloud; then use the temporal attention fusion network deployed in the cloud to dynamically model the temporal correlation of the multi-dimensional physiological data to generate a standardized multi-dimensional state vector; and encapsulate the obtained multi-dimensional state vector into a JSON protocol package; The scenario adaptive control module is configured to: parse the JSON protocol packet based on a hierarchical decision-making architecture and perform scenario adaptive control. Specifically, it initializes the threshold based on whether baseline data exists in the parsed content, uses a sliding time window to adjust the initialized threshold in real time, and updates the threshold interval based on historical rehabilitation effectiveness coefficients. It then drives the scenario library to switch between different scenarios and synchronously adjusts the motion trajectory of virtual objects and the sound and light conditions to generate a dynamic rehabilitation plan adapted to the user's neural network. The user feedback module is configured to achieve real-time coupling of physiology and scenarios through a cross-modal mapping engine, thereby achieving quantitative tracking of the user's rehabilitation progress and dynamic optimization of the rehabilitation plan. The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the method for formulating an intelligent VR rehabilitation program based on scene adaptive control as described in the first aspect of the present invention.

[0019] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for formulating an intelligent VR rehabilitation program based on scene adaptive control as described in the first aspect of the present invention are implemented.

[0020] One or more of the above technical solutions have the following beneficial effects: 1) After collecting the user's physiological data, the present invention first performs time alignment. It then uses a temporal attention fusion network to dynamically model the temporal correlations of multi-dimensional physiological data. Based on a hierarchical decision-making architecture, the JSON protocol packet is parsed and scenario-adaptive control is performed. Specifically, thresholds are initialized based on whether baseline data exists within the parsed content. A sliding time window is used to adjust the initialized thresholds in real time, and the threshold intervals are updated based on historical rehabilitation efficacy coefficients. The present invention then switches between different scenarios by driving the scenario library and synchronously adjusting the motion trajectory of virtual objects and the acoustic and optical conditions to generate a dynamic rehabilitation plan adapted to the user's neural network. Thus, through multiple safeguards, the present invention ensures the synchronization of the user's physiological state with scenario control, effectively preventing lags in scenario control.

[0021] 2) The intelligent VR rehabilitation program development system provided by this invention integrates a flexible biosensor bracket into the forehead area of ​​a VR headset. Furthermore, the flexible biosensor bracket is embedded with a gold-plated dry electrode array, and an optical heart rate sensor and a galvanic skin sensor are integrated via a flexible circuit board. Furthermore, an event-driven image sensor is embedded on either side of the nose pad at the lower edge of the VR headset. This system integrates a multimodal physiological sensing module with the VR device, significantly improving the user's immersive experience and reducing data synchronization accuracy.

[0022] 3) This invention integrates a gold-plated dry electrode array, an optical heart rate sensor, and a galvanic skin sensor embedded in a flexible biosensor scaffold to form a multimodal physiological sensing module. This module is directly connected to a multi-core processor via a low-latency interface, within which it integrates a data fusion processing module, a scene-adaptive control module, and a user feedback module. By leveraging a richer range of sensor data, this invention can more comprehensively characterize the user's cognitive, emotional, and motor complex state compared to existing technologies.

[0023] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0025] Figure 1 This is an overall flow chart of the method for formulating an intelligent VR rehabilitation plan based on scene adaptive control in Example 1 of the present invention.

[0026] Figure 2 This is a structural diagram of a VR head-mounted device in Example 2 of the present invention.

[0027] Figure 3 Schematic diagram of the structure of the flexible biosensor stent in the second embodiment of the present invention. DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0029] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to limit the exemplary embodiments according to the present invention.

[0030] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0031] The overall idea proposed by the present invention is: The present invention provides a method for formulating intelligent VR rehabilitation plans based on scene adaptive regulation. Through hardware-algorithm collaborative design, multi-dimensional physiological data such as EEG, heart rate, skin conduction and eye movement are interacted with VR scenes in real time, and a closed-loop feedback mechanism is constructed to achieve dynamic optimization of rehabilitation training intensity. That is, through this highly integrated multimodal perception system and intelligent control architecture, the personalized adaptation level of VR rehabilitation training is improved.

[0032] Example 1 This embodiment discloses a method for formulating an intelligent VR rehabilitation plan based on scene adaptive regulation.

[0033] like Figure 1 As shown in FIG, the method for formulating an intelligent VR rehabilitation program based on scene adaptive regulation includes: Step S1: collecting multi-dimensional physiological data of a human body and performing time alignment on the obtained multi-dimensional physiological data; Step S2: dynamically modeling the temporal correlation of the multi-dimensional physiological data using a temporal attention fusion network to generate a standardized multi-dimensional state vector; and encapsulating the multi-dimensional state vector into a JSON protocol package; Step S3: Parse the JSON protocol packet based on a hierarchical decision-making architecture and perform scenario-adaptive control. Specifically, the following steps are performed: threshold initialization is performed based on whether baseline data exists in the parsed content; a sliding time window is used to adjust the initialized threshold in real time; and the threshold interval is updated based on historical rehabilitation efficacy coefficients; the scenario library is driven to switch between different scenarios, and the motion trajectory of the virtual object and the sound and light conditions are synchronously adjusted to generate a dynamic rehabilitation plan adapted to the user's neural network. Step S4: Realize real-time coupling of physiology and scenario through a cross-modal mapping engine, and achieve quantitative tracking of the user's rehabilitation process and dynamic optimization of the rehabilitation plan.

[0034] Based on the above process, the present invention, building on the structure of an integrated VR device, achieves real-time alignment between user physiology and the scene through the implementation of multi-source heterogeneous data, significantly improving the personalized adaptation level of VR rehabilitation training. To facilitate understanding of the technical solution of the present invention, the specific implementation methods of the technical solution of the present invention are further explained and illustrated below.

[0035] In step S1, multi-dimensional physiological data of the human body is collected and time-aligned. This can be achieved by the following methods: Step S1-1: Collect multi-dimensional physiological data of the human body.

[0036] This invention modifies a VR headset based on a commercial VR hardware platform (such as the Oculus Quest series), utilizing the modified intelligent VR system to collect multi-dimensional physiological data from the user wearing the headset, namely, EEG signals, heart rate signals, electrodermal signals, and eye movement event streams. Specifically, the intelligent VR system uses an integrated EEG sensor (gold-plated dry electrode array) to collect EEG signals; an integrated optical heart rate sensor to collect the user's heart rate; an integrated electrodermal sensor to collect electrodermal signals; and an integrated eye event-driven image sensor to collect the user's eye movement event stream.

[0037] Among them, the EEG sensors (gold-plated dry electrode arrays) are distributed according to international standard sites; the optical heart rate sensor uses dual-wavelength photoelectric volumetric pulse wave acquisition technology; the galvanic skin sensor measures changes in skin impedance through interdigital electrodes; and the eye event-driven image sensor is based on an asynchronous pixel architecture, which enables each pixel to independently respond to brightness changes and output an event stream to generate high-precision eye movement trajectories.

[0038] Step S1-2: Time-align the obtained multi-dimensional physiological data.

[0039] A heterogeneous computing architecture is used to achieve efficient collaborative processing of multi-source biosignals (generated multi-dimensional physiological data). Multi-core processors are deployed on the embedded side of the intelligent VR system. Hardware timestampers are used to align EEG, heart rate, electrodermal, and eye movement event streams to eliminate multi-sensor physical latency. Signal synchronization in this heterogeneous computing architecture involves: first, unified marking by a hardware timestamper; then, high-speed signal transmission via a dedicated interface; and finally, dynamic calibration by the multi-core processor to achieve multi-source signal synchronization.

[0040] After the original signal is preprocessed, the following data operations are performed based on the multi-core processor: A. For EEG signals, extract the power spectrum percentages of the four frequency bands: delta (1-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), and beta (12-30 Hz). Delta waves are associated with deep relaxation or sleep, theta waves reflect meditation or low alertness, alpha waves represent wakefulness and relaxation, and beta waves correspond to high cognitive load.

[0041] By extracting EEG signals from these four frequency bands, we can cover the full spectrum of user states, from relaxation to high concentration, and thus comprehensively quantify cognitive load, providing a multi-dimensional basis for scenario-based adaptive control. In the specific implementation process, the power spectrum proportion is calculated based on the extracted frequency bands. This involves applying a fast Fourier transform (FFT) to each time window to convert the time domain signal into a frequency domain power spectrum. Finally, a normalization process is performed to calculate the power spectrum proportion.

[0042] B. For the ECG signal, calculate the time domain indicator SDNN (standard deviation of the normal RR interval) and the frequency domain indicator LF / HF ratio; LF represents the low-frequency power ratio of the frequency domain indicator, and HF represents the high-frequency power ratio of the frequency domain indicator. The R wave is the highest point peak of the QRS complex in the ECG (corresponding to the contraction caused by ventricular electrical excitation), and the RR interval represents the time interval (in milliseconds) between two adjacent R waves, reflecting the duration of a heartbeat cycle.

[0043] The time-domain metric SDNN is calculated as the standard deviation of normal RR intervals, reflecting overall fluctuations in heart rate variability (HRV) and assessing the stability of the autonomic nervous system. The frequency-domain metric (LF / HF ratio), as the ratio of low-frequency to high-frequency power, quantifies the balance between sympathetic (excitement) and parasympathetic (relaxation) nervous systems. This allows for a comprehensive assessment of a user's emotional arousal (e.g., stress level), providing a physiological basis for scenario manipulation. The time-domain metric SDNN and the frequency-domain metric LF / HF serve as standardized metrics for HRV analysis, quantifying the state of the autonomic nervous system from the time and frequency domains, respectively.

[0044] C. For the skin conductance signal, extract the peak value of the skin conductance response (SCR); D. Generate gaze heatmap based on eye movement event stream.

[0045] First, the VR scene is divided into a fine-grained pixel grid and the number of fixations in each grid unit is counted. Then, the Gaussian kernel density estimation algorithm is used to smooth the diffusion of discrete fixations to form a continuous heat distribution map. The Gaussian kernel density estimation formula is as follows: ; in, represents the gaze point coordinates, Represents the bandwidth parameter, and the output is a continuous thermal distribution map.

[0046] Then, the obtained heat distribution map is divided into N equal areas (such as 100×100 pixels), and the fixation probability of each area is calculated. ; Among them, the fixation probability is the ratio of the number of regional fixations to the total number of fixations. Finally, the fixation entropy is calculated based on the Shannon entropy formula ,Right now: .

[0047] In step S2, a temporal attention fusion network is used to dynamically model the temporal correlation of the multi-dimensional physiological data to generate a standardized multi-dimensional state vector; and the multi-dimensional state vector is encapsulated as a JSON protocol package.

[0048] After processing in step S1, the multi-core processor sends the resulting data to a temporal attention fusion network (TAFN) deployed in the cloud. Based on a lightweight Transformer encoder architecture, the TAFN dynamically models the temporal correlations of multi-dimensional physiological data through a multi-head self-attention mechanism. The TAFN includes a temporal embedding layer, a scene-adaptive weight allocation layer, and a feature fusion output layer. In this embodiment, the multi-head self-attention mechanism uses four attention heads.

[0049] Specifically, input features (EEG power spectrum, heart rate variability index, electrodermal response (SCR) amplitude, and eye gaze entropy) are positionally encoded and injected with temporal information. An attention score is then calculated using a query-key-value matrix. This is combined with the scene type vector (e.g., high cognitive load scenes are coded as 1,0,0) and the user's real-time status (e.g., task completion) to dynamically assign a weight coefficient (e.g., increasing the EEG weight to 0.6). This weighted fusion generates a standardized multidimensional state vector, including a cognitive load index (based on the weighted value of the EEG beta / theta power ratio and eye gaze entropy), emotional arousal (a fusion of electrodermal response (SCR) amplitude and heart rate (LF / HF) ratio), and a motor coordination score (based on the dispersion of eye movement trajectories). This data is then encapsulated into a JSON protocol packet and pushed back to the scene adaptive control module within the intelligent VR system via WebSocket in real time.

[0050] In step S3, the JSON protocol packet is parsed based on the hierarchical decision architecture and scenario adaptive control is performed.

[0051] The scene adaptive control module in the intelligent VR system realizes dynamic optimization of rehabilitation stimulation based on a hierarchical decision-making architecture. After receiving the multi-dimensional state vector output by the data fusion processing module: First, threshold initialization is performed based on whether baseline data exists in the parsed content. Specifically, threshold initialization is performed based on the user type (neural injury / movement disorder) and rehabilitation stage (early / middle / late). A. When baseline data exists, that is, when the resting EEG alpha wave ratio is >30% and the heart rate SDNN baseline value is >50ms, calculate the initial threshold for: ; in, is the normalized value of the EEG alpha wave power ratio (based on the ratio of alpha wave power to total power of the full frequency band in the baseline data), is the baseline value of the heart rate variability time domain indicator SDNN (unit: ms); weight coefficient and Dynamically allocated based on user type (default is 0.5 each).

[0052] B. When there is no baseline data, the scene complexity model is called to dynamically correct the preset threshold for: ; in, is the preset maximum complexity, log is the natural logarithm function; scene complexity score The speed of the object v (m / s), sound field density d (number of sound sources / cubic meter) and frequency of light and shadow changes (times / second) is obtained by weighted calculation, that is: ; Next, in the real-time control phase, a sliding time window is used to monitor the state vector. In this embodiment, the sliding time window used is a 10-second window with a 1-second step size. If the cognitive load index exceeds the current threshold within three consecutive windows, , then the threshold is adjusted upward according to the gradient rule, that is: ; in, is the mean of the cognitive load index within the window, and the coefficient 0.05 is the learning rate (control threshold adjustment amplitude); Indicates the threshold after being increased according to the gradient rule.

[0053] Subsequently, in the long-term optimization phase, the threshold interval is updated through transfer learning in combination with the historical rehabilitation efficacy coefficient. To ensure that the threshold dynamically matches the user's recovery progress. Specifically, the historical recovery efficiency coefficient is: ; in, represents the historical rehabilitation efficacy coefficient, is the historical threshold change amplitude, Δ t is the time span. Based on the historical rehabilitation effectiveness coefficient, the threshold interval can be updated through transfer learning to ensure that the threshold dynamically matches the user's rehabilitation progress. Specifically, the threshold interval is: =[0.4 ,1.2 ]; in, represents the updated threshold interval, Represents the historical rehabilitation efficacy coefficient.

[0054] Ultimately, the driving scene library intelligently switches between natural ecology (low stimulation), urban traffic (medium stimulation), and snowfield (high stimulation), and simultaneously adjusts the complexity of the virtual object's motion trajectory (such as the vehicle speed in the urban scene ±15%), the ambient light color temperature (6500K→4500K), and the low-frequency proportion of the sound field (20Hz-200Hz increased by 10dB), forming a dynamic rehabilitation plan that is highly adapted to the user's neuroplasticity. Intelligent switching of the scene library is achieved by matching the real-time cognitive load index with the dynamic threshold interval. Specifically: the system is based on the updated threshold interval [0.4×E_eff, 1.2×E_eff]; where E_eff is the historical rehabilitation efficiency coefficient, and periodically detects the sub-interval of the cognitive load index ( 0.4×E_eff) selects natural ecology, in the sub-interval (0.4~0.8×E_eff) selects urban transportation, in the sub-interval ( When the threshold exceeds 0.8 × E_eff, the snowfield scene is selected, and the scene type is forcibly downgraded when the threshold is severely exceeded (ΔT ≥ 0.2). The development of the rehabilitation plan relies on a closed-loop linkage between scene switching and parameter adjustment. After intelligently switching scene types, the virtual object movement speed, ambient light color temperature, and low-frequency energy of the sound field are dynamically adjusted. A cross-modal mapping engine then converts physiological signals into scene parameter changes in real time, ultimately generating a dynamic rehabilitation plan adapted to the user's progress in neuroplasticity.

[0055] In step S4, real-time coupling of physiology and scenario is achieved through a cross-modal mapping engine, enabling quantitative tracking of the user's rehabilitation process and dynamic optimization of the rehabilitation plan.

[0056] The user feedback module within the intelligent VR system achieves real-time physiological-scene coupling through a cross-modal mapping engine. Specifically, it maps the EEG concentration index (β / θ power ratio) to the cloud movement rate of a dynamic weather system (linearly corresponding to 0.5-2.0 m / s), converts the heart rate variability (LF / HF) ratio into the pulse frequency of a virtual light source (exponentially adjustable from 0.1-5 Hz), uses the peak galvanic skin response (SCR) amplitude to drive the ripple density of the water surface (increasing by 5 ripples / ㎡ every 0.1 μs), and uses eye movement heat maps to trigger enhanced rendering of localized details (texture resolution in the focal area is increased to 8K). The cross-modal mapping engine achieves closed-loop feedback by converting physiological signals into virtual scene parameters in real time. Essentially, it acts as a translator from physiological signals to virtual parameters, converting the user's neural state into actionable scene variables through quantitative rules.

[0057] When the data fusion result exceeds the adaptive threshold, a graded transition protocol is activated, combined with a time series database, to generate a cross-scenario physiological indicator evolution map. The user's rehabilitation effectiveness is visualized through a three-dimensional scatter plot, enabling quantitative tracking of the user's rehabilitation progress and dynamic optimization of the rehabilitation plan. This allows real-time monitoring of the user's rehabilitation effectiveness and dynamic optimization of the rehabilitation plan. The primary advantage lies in improving the personalization and adaptability of rehabilitation training, thereby avoiding mismatches between training intensity and user tolerance. Specifically, the graded transition protocol includes two levels: mild and severe. Mild excesses involve local adjustments to the light source and sound field (e.g., reducing the pulse frequency by 10%). Severe excesses involve a global scene switch (e.g., switching from a city scene to a snowy field scene). The transition process uses a gradual in-and-out algorithm to avoid visual jerks. Furthermore, when the adaptive threshold ΔT is less than 0.2, it is considered a mild excess; when the adaptive threshold ΔT is ≥ 0.2, it is considered a severe excess.

[0058] At the same time, a cross-scenario physiological indicator evolution map is generated based on the time series database, and the phased transition of the user's rehabilitation efficiency is visualized through a three-dimensional scatter plot, so as to achieve quantitative tracking and dynamic optimization of the rehabilitation process. That is, the user's physiological indicator time series data stored in the database is used to extract key indicators (such as cognitive load index and emotional arousal), and their phased transitions are visualized through a three-dimensional scatter plot to achieve dynamic tracking of the rehabilitation process and program optimization.

[0059] Example 2 This embodiment discloses an intelligent VR rehabilitation program formulation system based on scene adaptive regulation.

[0060] An intelligent VR rehabilitation program formulation system based on scene adaptive regulation, including: a VR headset and a flexible biosensor scaffold; The flexible biosensor scaffold is integrated into Figure 2 The forehead area of ​​the VR head-mounted device shown is detachable; at the same time, Figure 3 As shown, a gold-plated dry electrode array is embedded in the flexible biosensor bracket, and an optical heart rate sensor and a skin charge sensor are integrated through a flexible circuit board; an event-driven image sensor is embedded on both sides of the nose pad at the lower edge of the VR head-mounted device; The gold-plated dry electrode array, optical heart rate sensor, and galvanic skin sensor constitute a multimodal physiological sensing module, which is directly connected to a multi-core processor via a low-latency interface. As an optional embodiment, the multi-core processor used is the NVIDIA Jetson AGX Orin. The multi-core processor integrates a data fusion processing module, a scene adaptive control module, and a user feedback module. Specifically: The multimodal physiological perception module is configured to: collect multi-dimensional physiological data of the human body; The data fusion processing module is configured to: perform time alignment on the obtained multi-dimensional physiological data and send the multi-dimensional physiological data after data fusion processing to the cloud; then use the temporal attention fusion network deployed in the cloud to dynamically model the temporal correlation of the multi-dimensional physiological data to generate a standardized multi-dimensional state vector; and encapsulate the obtained multi-dimensional state vector into a JSON protocol package; The scenario adaptive control module is configured to: parse the JSON protocol packet based on a hierarchical decision-making architecture and perform scenario adaptive control. Specifically, it initializes the threshold based on whether baseline data exists in the parsed content, uses a sliding time window to adjust the initialized threshold in real time, and updates the threshold interval based on historical rehabilitation effectiveness coefficients. It then drives the scenario library to switch between different scenarios and synchronously adjusts the motion trajectory of virtual objects and the sound and light conditions to generate a dynamic rehabilitation plan adapted to the user's neural network. The user feedback module is configured to achieve real-time coupling of physiology and scenarios through a cross-modal mapping engine, thereby achieving quantitative tracking of the user's rehabilitation progress and dynamic optimization of the rehabilitation plan.

[0061] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0062] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for formulating an intelligent VR rehabilitation program based on scene adaptive regulation as described in the first embodiment of the present disclosure.

[0063] Example 4 The purpose of this embodiment is to provide an electronic device.

[0064] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for formulating an intelligent VR rehabilitation program based on scene adaptive control as described in the first embodiment of the present disclosure are implemented.

[0065] The steps involved in the apparatuses of Examples 2, 3, and 4 above correspond to those of Method Example 1. For detailed implementations, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any method of the present invention.

[0066] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0067] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. An intelligent VR rehabilitation program formulation method based on scene adaptive regulation is characterized by: include: Collect multi-dimensional physiological data of the human body and perform time alignment on the obtained multi-dimensional physiological data; A temporal attention fusion network is used to dynamically model the temporal correlation of the multi-dimensional physiological data to generate a standardized multi-dimensional state vector; Encapsulate the multidimensional state vector into a JSON protocol package; Parse the JSON protocol package based on a hierarchical decision-making architecture and perform scenario-adaptive control. Specifically, the threshold is initialized based on whether baseline data exists in the parsed content. A sliding time window is used to adjust the initialized threshold in real time, and the threshold range is updated based on historical rehabilitation efficacy coefficients. The drive scene library switches between different scenes and synchronously adjusts the motion trajectory of virtual objects and the sound and light conditions to generate a dynamic rehabilitation plan adapted to the user's neural network; at the same time, through the cross-modal mapping engine, real-time coupling of physiology and scenes is achieved, realizing quantitative tracking of the user's rehabilitation process and dynamic optimization of the rehabilitation plan.

2. The method for formulating an intelligent VR rehabilitation program based on scene adaptive control according to claim 1, characterized in that: The temporal attention fusion network is based on a lightweight Transformer encoder architecture and dynamically models the temporal correlation of multi-dimensional physiological data through a multi-head self-attention mechanism.

3. The method for formulating an intelligent VR rehabilitation program based on scene adaptive control according to claim 2, characterized in that: The temporal attention fusion network includes a temporal embedding layer, a scene adaptive weight allocation layer and a feature fusion output layer.

4. The method for formulating an intelligent VR rehabilitation program based on scene adaptive control according to claim 1, characterized in that: The threshold is initialized based on whether there is baseline data in the parsed content, including: when there is no baseline data, calling the scene complexity model to dynamically correct the preset threshold; the scene complexity score is calculated by weighting the object movement speed, sound field density and light and shadow change frequency.

5. The method for formulating an intelligent VR rehabilitation program based on scene adaptive control according to claim 1, characterized in that: Real-time coupling of physiology and scenarios is achieved through a cross-modal mapping engine, including: first, cross-modal mapping; then, based on the extended modal mapping, a hierarchical transition protocol is adopted, combined with a time series database to generate a cross-scenario physiological indicator evolution map, and the phased transition of the user's rehabilitation effectiveness is visualized through a three-dimensional scatter plot, realizing quantitative tracking of the user's rehabilitation process and dynamic optimization of the rehabilitation plan.

6. The method for formulating an intelligent VR rehabilitation program based on scene adaptive control according to claim 5, characterized in that: The hierarchical transition protocol includes two levels: mild over-limit and severe over-limit. In the case of mild over-limit, the operation to be performed is to locally adjust the light source and sound field; in the case of severe over-limit, the operation to be performed is to globally switch the scene type.

7. The intelligent VR rehabilitation program formulation system based on scene adaptive control is characterized by: include: VR headsets and flexible biosensing scaffolds; The flexible biosensor bracket is integrated into the forehead area of ​​the VR headset and is detachable. Furthermore, the flexible biosensor bracket is embedded with a gold-plated dry electrode array and integrated with an optical heart rate sensor and a skin charge sensor via a flexible circuit board. An event-driven image sensor is embedded on both sides of the nose pad at the lower edge of the VR headset. The gold-plated dry electrode array, optical heart rate sensor and skin charge sensor constitute a multimodal physiological sensing module; the multimodal physiological sensing module is directly connected to a multi-core processor through a low-latency interface, and the multi-core processor integrates a data fusion processing module, a scene adaptive control module and a user feedback module.

8. The intelligent VR rehabilitation program formulation system based on scene adaptive control according to claim 7, characterized in that: The multimodal physiological perception module is configured to: collect multi-dimensional physiological data of the human body; The data fusion processing module is configured to: perform time alignment on the obtained multi-dimensional physiological data, and send the multi-dimensional physiological data after data fusion processing to the cloud; The temporal attention fusion network deployed in the cloud then dynamically models the temporal correlation of the multi-dimensional physiological data to generate a standardized multi-dimensional state vector. The obtained multi-dimensional state vector is then encapsulated as a JSON protocol package. The scenario-adaptive control module is configured to: parse the JSON protocol packet based on a hierarchical decision-making architecture and perform scenario-adaptive control. Specifically, it initializes the threshold based on whether baseline data exists in the parsed content, uses a sliding time window to adjust the initialized threshold in real time, and updates the threshold interval based on historical rehabilitation effectiveness coefficients; Then, the scene library is driven to switch between different scenes, and the motion trajectory of virtual objects and the sound and light conditions are synchronously adjusted to generate a dynamic rehabilitation plan adapted to the user's neural network; The user feedback module is configured to achieve real-time coupling of physiology and scenarios through a cross-modal mapping engine, thereby achieving quantitative tracking of the user's rehabilitation progress and dynamic optimization of the rehabilitation plan.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for formulating an intelligent VR rehabilitation program based on scene adaptive control are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps in the method for formulating an intelligent VR rehabilitation program based on scene adaptive control are implemented as described in any one of claims 1 to 6.

Images