Multi-screen dynamic display control method and device and electronic equipment

Abstract

The invention relates to the technical field of intelligent interaction, and discloses a multi-screen dynamic display control method and device and electronic equipment, and the method comprises the following steps: constructing a seven-order behavior tensor through heterogeneous sensor space-time reference alignment, and employing CP-Tucker mixed decomposition to extract cross-modal intention features; resource allocation parameters are generated based on quantum annealing and classical optimization collaboratively, and screen geometric mapping is dynamically corrected in combination with a neural radiation field; federal learning noise injection and an elastic pixel binding strategy are creatively fused, and real-time rendering control under privacy protection is realized. Through the synergistic effect of space-time synchronous compensation, mixed tensor decomposition, a quantum optimization engine and a differentiable rendering model, dynamic adaptation of display parameters, user behaviors and environment changes is achieved, and the multi-screen collaborative display precision and real-time performance are improved on the premise that privacy security is ensured.

Description

Technical Field

[0001] The present invention relates to the field of intelligent interaction technologies, and specifically to a multi-screen dynamic display control method, device, and electronic device. Background Art

[0002] In the field of intelligent interaction systems, multi-screen dynamic display control technology faces the dual challenges of multi-dimensional perception data fusion and real-time resource scheduling. When integrating heterogeneous sensor data with traditional methods, due to the lack of an accurate spatio-temporal benchmark alignment mechanism, spatio-temporal misalignment errors occur in the process of extracting behavioral characteristics, making it difficult to accurately capture the cross-modal correlation characteristics of user intentions.

[0003] When dealing with the problem of multi-screen collaborative optimization, existing resource allocation algorithms often adopt a static allocation strategy with a fixed threshold, which cannot adapt to network latency fluctuations and dynamic changes in the physical space, and is prone to cause spatial mismatch between the displayed content and user perception.

[0004] Current display calibration technologies mostly rely on preset geometric mapping parameters. When the environmental light changes suddenly or the user's physiological state changes, due to the lack of a biometric coupling mechanism, the update of display parameters lags behind the actual requirements. At the same time, the balance problem between privacy protection and model performance has not been effectively solved. Traditional federated learning schemes adopt a fixed noise injection strategy, which is prone to cause unstable model convergence in dynamic scenarios and difficult to maintain the persistence of multi-device collaborative optimization. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a multi-screen dynamic display control method, device, and electronic device, which solve the problems in the prior art that the spatio-temporal benchmarks of multi-modal perception data are inconsistent, resulting in intention recognition deviation, static resource allocation strategies are difficult to adapt to dynamic environmental changes, display calibration lags behind biometric responses, and it is difficult to synergistically optimize privacy protection and model performance.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A multi-screen dynamic display control method, including the following steps: S1. Synchronously collect multi-modal behavioral data of the user and environmental parameters, and generate multi-source perception data with a unified benchmark through spatio-temporal alignment processing; S2. Based on the multi-source perception data, construct a seventh-order behavioral tensor model containing spatio-temporal correlation features, and perform hybrid tensor decomposition to extract user intention features; S3. According to the characteristic parameters output by the behavioral tensor model, solve the dynamic resource allocation parameters between multiple screens through a quantum-classical hybrid optimization algorithm; S4. Based on the resource allocation parameters, use a neural radiance field to dynamically correct the geometric parameters of the display device, and generate a display mapping relationship that conforms to the physical space topology; S5. Update the multi-modal intent recognition model using the federated learning mechanism according to the display mapping relationship, and feedback the updated model parameters to the tensor decomposition process in step S2; S6. Based on the corrected display parameters and resource allocation scheme, drive the multi-screen system to perform dynamic rendering and content output.

[0007] Preferably, step S1 includes the following steps: Capture the device motion attitude quaternion at a sampling rate of 120 Hz through an inertial measurement unit, generate a depth map with a resolution of 640×480 through a ToF depth camera, and measure the pupil diameter change rate in real time through an eye tracking module; Implement multi-sensor clock synchronization using the IEEE1588v2 protocol, and construct a spatial transformation matrix chain including translation vectors and rotation matrices.

[0008] Preferably, the hybrid tensor decomposition in step S2 satisfies the following formula definition: CP decomposition term: ; Where: is the th eigenvalue; represents the th factor vector of the represents the vector outer product operation; Tucker decomposition term: ; Where: is the core tensor; is the factor matrix of the th modality; represents the th modality product of the tensor and the matrix.

[0009] Preferably, the quantum optimization process in step S3 constructs the following Hamiltonian: ; Where: represents the association strength between screen and , which is determined by the Frobenius norm of the corresponding dimension of the behavior tensor; is the physical screen spacing; is the dynamic time delay threshold, which is adaptively adjusted according to the network load rate; , is the Pauli operator of the -th qubit.

[0010] Preferably, the step S4 includes the following steps: Establish a neural radiance field model, and generate the screen pixel color through cumulative calculation of the transmittance, where the transmittance is related to the optical properties of the screen material; Dynamically adjust the update rate of the radiance field parameters according to the real-time collected user physiological characteristic signals, where the physiological characteristic signals include at least one of pupil diameter and heart rate variability; Introduce a screen curvature differential constraint term to ensure that the geometric correction process meets the physical space continuity condition.

[0011] Preferably, the step S5 includes the following steps: During the device-side training process, inject adaptive noise that is non-linearly related to the gradient norm into the model gradient, where the noise intensity decays exponentially as the gradient magnitude increases; When aggregating the confused gradients on the edge server, adopt a periodic decay strategy to dynamically adjust the learning rate, where the learning rate remains high in the initial stage of training to accelerate convergence and gradually decreases in the later stage to improve the model stability; Encrypt and transmit the updated model parameters to the tensor decomposition in step S2 for reconstructing the core dimension of the behavior tensor.

[0012] Preferably, the step S6 includes the following steps: Construct a virtual pixel resource pool, define the dynamic mapping relationship between virtual pixels and the multi-physical screen, where the alpha-channel blending weight of each virtual pixel is jointly optimized according to the screen physical spacing and network delay; Perform environment-aware dynamic tone compression, and jointly adjust the HDR compression curve based on the ambient light field intensity distribution and user biometric signals, where the compression intensity is positively correlated with the mean ambient light intensity; Implement the physical screen brightness conservation constraint, and ensure that the sum of the brightness outputs of each screen pixel does not exceed the preset threshold through normalization processing.

[0013] Preferably, the calculation method of the screen correlation is: ; where is an adjustable integration time window, represents the slice of the tensor in the screen dimension.

[0014] The multi-screen dynamic display control device includes: A multi-source data synchronous acquisition module, configured to perform spatio-temporal alignment processing of user behavior data and environmental parameters through a heterogeneous sensor array; A spatio-temporal tensor modeling module, configured to construct and decompose a seventh-order behavior tensor model based on a hybrid tensor decomposition algorithm; A quantum-classical collaborative optimization module, configured to solve screen resource allocation parameters with spatio-temporal constraints; A biophysical coupling correction module, configured to dynamically generate a geometric mapping relationship of a display device according to a neural radiation field; A privacy gradient aggregation module, configured to implement adaptive noise injection and gradient obfuscation of a federated learning model; A dynamic resource rendering engine module, configured to perform elastic pixel mapping and dynamic tone rendering of a multi-screen system.

[0015] An electronic device, comprising: A processor; A memory storing a computer program; When the processor executes the program, the above-mentioned method is implemented.

[0016] The present invention provides a multi-screen dynamic display control method, device and electronic device. It has the following beneficial effects: 1. Through the construction of heterogeneous sensor clock synchronization and spatial transformation chain and the combination of CP-Tucker hybrid tensor decomposition, the present invention solves the problem of intention recognition deviation caused by data fragmentation in the existing multi-screen system behavior modeling, and overcomes the defect of insufficient capture of cross-dimensional correlation features by traditional single-modal modeling.

[0017] 2. Based on the construction of a quantum annealing Hamiltonian and classical post-processing fine-tuning, the present invention breaks through the bottleneck that traditional heuristic algorithms are prone to fall into local optimum under complex constraints, can realize the global optimization of multi-screen resource allocation, and avoids the insufficient adaptability of the fixed threshold strategy to network delay fluctuations.

[0018] 3. Through the fusion of a differentiable rendering model and a parameter update strategy driven by pupil diameter, the present invention solves the problem of error accumulation caused by environmental mutations in traditional geometric correction methods, and overcomes the limitation that static mapping models cannot adapt to dynamic changes in biometric features.

[0019] 4. The present invention adopts an adaptive noise and virtual pixel mixing weight optimization technology related to gradient norm, maintains the effectiveness of model update on the premise of ensuring user privacy, and eliminates the risk of model performance degradation caused by fixed noise intensity in existing federated learning schemes. Description of the Drawings

[0020] Figure 1 It is a flowchart of the method of the present invention; Figure 2 Schematic diagram of the device structure of the present invention. Specific implementation manners

[0021] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Embodiment: Please refer to the attached Figure 1 , the embodiments of the present invention provide a multi-screen dynamic display control method, including the following steps: S1. Synchronously collect the multi-modal behavior data and environmental parameters of the user, and generate multi-source perception data with a unified benchmark through spatio-temporal alignment processing; To achieve precise synchronous acquisition of multi-modal data and spatio-temporal benchmark alignment, this step adopts a heterogeneous sensor collaborative perception architecture, and realizes physical space mapping and timing consistency control of multi-source information through a hierarchical data processing flow. The specific implementation process is as follows: First, configure a multi-modal sensor array, including an inertial measurement unit, a time-of-flight depth camera, and a near-infrared eye tracking module. The inertial measurement unit preferably adopts a nine-axis microelectromechanical system, and calculates the device attitude quaternion through the fusion of an accelerometer, a gyroscope, and a magnetometer. The sampling frequency is set to 120 Hz to ensure the continuity of motion capture. The time-of-flight depth camera is preferably configured with an infrared light source in the 940 nm band, generates a depth map with a resolution of 640×480 by using the phase measurement principle, and realizes millimeter-level ranging accuracy through adaptive adjustment of the modulation frequency. The eye tracking module integrates a high-frame-rate near-infrared camera, analyzes the change rate of the pupil diameter based on the corneal reflection vector, and the sampling interval is preferably 33 ms to match the human visual persistence characteristics.

[0023] To eliminate the spatio-temporal benchmark differences of multi-source data, a hierarchical alignment mechanism is established. At the time sequence synchronization level, the IEEE1588v2 precise clock protocol is adopted to construct a distributed clock network. Each sensor node deploys a hardware timestamp module, and realizes microsecond-level synchronization accuracy through a master-slave clock architecture. Specifically, the master clock node periodically broadcasts synchronization messages, and the slave nodes calculate the clock offset based on the message transmission delay, and compensate the crystal oscillator frequency offset through a linear regression algorithm. Preferably, the synchronization period is set to 2 seconds to balance the network load and synchronization accuracy requirements.

[0024] At the spatial registration level, a rigid transformation relationship chain between sensor coordinate systems is constructed. Taking the inertial measurement unit coordinate system as the global reference system, the spatial transformation parameters of the depth camera and the eye movement module relative to the inertial unit are obtained through external parameter calibration. A transformation model from the depth camera coordinate to the global reference is established: ; where represents the three-dimensional point coordinates in the global coordinate system, is the original coordinate in the depth camera coordinate system, is the rotation matrix, is the translation vector. The external parameter calibration preferably adopts the checkerboard joint calibration method, and the least squares optimization problem is solved through the corresponding relationship of feature points under multiple postures.

[0025] For the eye movement tracking module, a mapping relationship from the pupil center coordinate to the global reference is established: ; where is the line-of-sight direction vector in the eye movement module coordinate system, and are the corresponding rotation matrix and translation vector. Preferably, a non-linear optimization algorithm is used to compensate for the coordinate deviation caused by lens distortion, and a polynomial distortion model is constructed: ; where is the ideal projection coordinate, is the actual observed coordinate, is the radial distortion coefficient, is the tangential distortion coefficient, .

[0026] In the data fusion stage, the extended Kalman filter is used to achieve the spatio-temporal alignment of multi-source perception data. A state vector is established, where is the attitude quaternion, is the angular velocity, is the linear acceleration, is the position vector. The prediction equation is derived based on the inertial measurement unit data: ; The observation equation fuses the depth camera and eye movement data: ; where is the non-linear observation function, is the observation noise. The multi-sensor confidence weighted fusion is realized through the covariance matrix adaptive adjustment, and the weight of the inertial data is automatically increased when the confidence of the depth data is lower than the threshold.

[0027] Finally, a unified benchmark multi-source perception data stream is generated. The data structure includes fields such as timestamp, spatial coordinates, depth information, physiological parameters, and ambient light intensity. The data stream realizes temporal alignment through a circular buffer, and the buffer size is preferably 500 ms to accommodate the maximum expected transmission delay.

[0028] S2. Construct a seventh-order behavior tensor model containing spatio-temporal correlation features based on multi-source perception data, and perform hybrid tensor decomposition to extract user intention features; To achieve the accurate extraction of user intention features, in this step, a behavior tensor model integrating multi-dimensional spatio-temporal correlation is constructed, and the potential behavior patterns are separated through a hybrid tensor decomposition algorithm. The specific implementation process first establishes a seventh-order tensor data structure, and then alternately performs CP decomposition and Tucker decomposition to extract local and global features. Finally, the two types of decomposition results are fused to generate an intention feature vector.

[0029] The dimension definition of the seventh-order behavior tensor is based on the multi-source perception data generated in step S1. Each dimension respectively represents spatial coordinates, line-of-sight direction, gesture features, ambient light intensity, sound field distribution, physiological signals, and temporal information. Preferably, the temporal dimension is constructed using a sliding window mechanism, and the window length is adaptively adjusted according to the user behavior continuity requirement. The tensor element values are obtained through multi-source data normalization to ensure the dimensional consistency of each modality.

[0030] The CP decomposition process is used to capture independent behavior patterns, and its mathematical expression is: ; where represents the weight coefficient of the th feature component, is the th factor vector of the th modality, represents the vector outer product operation. The decomposition rank is preferably automatically determined based on the Akaike information criterion, suppressing noise interference while retaining important behavior patterns. The Tucker decomposition process is used to extract cross-modal correlation features, and its mathematical expression is: ; where is the core tensor, are the factor matrices of each modality, represents the product operation of the tensor in the th modality. The core tensor dimension is preferably set to 30%-50% of the corresponding original dimension to achieve feature dimensionality reduction.

[0031] During the hybrid decomposition process, CP decomposition and Tucker decomposition are alternately iterated. First, the factor matrices of CP decomposition are initialized and used as the initial values for Tucker decomposition. In each iteration, the CP decomposition results are fixed and the Tucker core tensor is updated, and then the weight coefficients of CP decomposition are optimized in the reverse direction. Preferably, the alternating least squares method is used to update the parameters, and the objective function is defined as: ; where is the regularization coefficient, which is used to control the sparsity of the factor matrices. The objective function is solved by the projected gradient descent method, and the step size is adaptively adjusted according to the condition number of the Hessian matrix.

[0032] In the feature fusion stage, the local pattern features of CP decomposition and the global correlation features of Tucker decomposition are linearly combined according to the information entropy weights. Define the information entropy of the th CP component as: ; The fusion weight coefficient is dynamically adjusted according to the information entropy: ; where is the temperature coefficient, preferably set to the reciprocal of the current iteration number to balance exploration and exploitation. The final intention feature vector is generated by weighted summation: ; where represents the vector concatenation operation, is the unfolding matrix of the core tensor in the seventh mode, is the principal component vector of the temporal factor matrix.

[0033] The hybrid decomposition algorithm is implemented in a distributed computing framework, preferably using the resilient distributed dataset mechanism of ApacheSpark. The tensor slices are assigned to different computing nodes, and each node performs local decomposition calculations in parallel, and the driver periodically performs global parameter aggregation. To reduce the communication overhead, a data transmission protocol based on residual compression is designed to transmit only the parameters whose gradient changes exceed the threshold.

[0034] S3. According to the feature parameters output by the behavior tensor model, solve the dynamic resource allocation parameters between multiple screens through a quantum-classical hybrid optimization algorithm; To achieve the dynamic optimization of multi-screen resource allocation, this step constructs an optimization model based on quantum-classical hybrid computing, and converts the intention features extracted in step S2 into physical screen parameter configurations. The specific implementation is to first establish a discrete optimization problem with spatio-temporal constraints, then approximate the optimal solution through quantum annealing search, and finally fine-tune the parameters by combining classical optimization algorithms.

[0035] The objective function of the optimization problem is defined as: ; where represents the feature vector of the -th screen (from the output of step S2), is a binary assignment variable, is the content block attribute vector, characterizes the association strength between the screen and the content , is the regularization coefficient. The association strength is calculated using: ; where is the slice of the behavior tensor in the screen dimension and the content dimension , is the dynamic integration window, whose value is adaptively adjusted according to the network delay variance : ; where is the reference integration time, is the decay coefficient. The delay variance is obtained by calculating the sample variance of the historical transmission delay sequence through a sliding window.

[0036] Map the above optimization problem to a quantum annealing model and construct the following Hamiltonian: ; where and are the Pauli operators of the -th qubit, represents the physical distance between the screen and , is the adaptive time delay threshold. The time delay threshold is dynamically updated according to the current network load rate : ; The quantum annealing process is preferably implemented on a D-Wave quantum processor, and the annealing time is set according to the problem size : ; After annealing is completed, the first low-energy samples are collected, and a candidate solution set is generated through a majority voting mechanism. The candidate solutions are input into a classical optimization module for post-processing, and the objective function is adjusted to: ; where is the sample weight (negatively correlated with the energy value), is the total variation regularization term. The coordinate descent method is used for iterative optimization, and each time a single screen allocation variable is updated: ; Optimization result is output as the final resource allocation parameter to step S4. To improve the computational efficiency, a distributed computing framework is implemented, and the qubit mapping problem is decomposed into multiple sub-problems. Each sub-problem corresponds to a local screen cluster, and parallel annealing calculations are performed through Apache Spark, and the driver periodically performs global solution aggregation. The data sharding strategy is based on clustering the physical locations of the screens to ensure that the coupling strength between sub-problems is lower than the threshold.

[0037] S4. Based on the resource allocation parameters, use the neural radiance field to dynamically correct the geometric parameters of the display device, and generate a display mapping relationship that conforms to the physical space topology; To achieve the dynamic correction of the geometric parameters of the display device, in this step, a differentiable rendering model based on the neural radiance field is constructed, and the resource allocation parameters generated in step S3 are converted into the geometric mapping relationship of the physical screen. Specifically, an implicit neural representation of the screen surface is first established, and then the radiance field parameters are optimized by integrating biophysical characteristics, and finally a display mapping matrix that conforms to spatial continuity is generated.

[0038] The neural radiance field model is implemented through a multi-layer perceptron (MLP). The network input includes the spatial coordinates and the viewing direction , and the output is the volume density and the color value . Preferably, the network structure includes 8 hidden layers, the number of neurons in each layer decays exponentially with depth, and the activation function uses a combination of ReLU and Softplus: ; where is the hidden feature of the th layer, , and , are trainable parameters. When accumulating colors at the sampling points along the ray, the hierarchical volume rendering formula is used: ; where the transmittance is related to the volume density and the sampling interval and satisfies: ; Define the parameter update rate for fusing biophysical features with the physiological signal (such as the change rate of pupil diameter): ; where is the base learning rate, is a small constant to prevent division by zero. When the pupil diameter exceeds the threshold, the sampling point density is automatically increased , preferably adjusted dynamically according to , is the reference value of the physiological signal.

[0039] Introduce the screen surface differential constraint term to ensure that the surface after geometric correction satisfies physical continuity. Define the curvature constraint loss function: ; where is the implicit function of the screen surface, represents the second-order derivative operator of the surface. This constraint term is jointly optimized with the rendering loss : ; where is the trade-off coefficient, preferably determined by grid search on the validation set. The Adam optimizer is used in the optimization process, and the momentum parameters , are set to 0.9 and 0.99.

[0040] The dynamic correction process is executed in the following time sequence: First, load the resource allocation parameters output in step S3 and initialize the neural radiance field model parameters. Subsequently, collect real-time physiological signals and calculate the adaptive learning rate and sampling density. In each iteration, randomly sample a set of rays from the screen surface and calculate the color accumulation value, and update the network parameters by backpropagation. After every 50 iterations, evaluate the mean square error of the curvature constraint term, and if it exceeds the threshold, increase the value.

[0041] In terms of implementation details, the screen surface implicit function is represented by the signed distance function (SDF), and its zero isosurface corresponds to the physical screen surface. The SDF network shares underlying features with the radiance field network and realizes parameter interaction through the cross-attention mechanism. Preferably, the model inference is deployed on the NPU accelerator, and the OpenGLES3.2 interface is called during real-time rendering to generate the geometric mapping matrix.

[0042] Geometric mapping matrix Each element stores pixel coordinate transformation parameters, including a translation vector and a rotation quaternion. When updating the matrix, regions with a curvature change rate exceeding a threshold are preferentially processed, and the query is accelerated through a spatial hash table. The external parameter calibration uses an offline calibration method assisted by a checkerboard to establish a rigid transformation relationship between the screen coordinate system and the global reference.

[0043] S5. According to the display mapping relationship, use the federated learning mechanism to update the multi-modal intent recognition model, and feedback the updated model parameters to the tensor decomposition process in step S2; To achieve privacy protection and dynamic adaptation of model updates, this step constructs a distributed training framework based on federated learning, uses the display mapping relationship generated in step S4 as a supervision signal, and iteratively optimizes the multi-modal intent recognition model. Specifically, local training is first performed on the device side and adaptive noise is injected, then the gradients are aggregated on the edge server and the learning strategy is adjusted, and finally the encrypted parameters are sent back to the tensor decomposition module in step S2.

[0044] The differential privacy protection mechanism is adopted in the local training process, and the gradient perturbation equation is defined as: ; where is the original gradient of the th device in the th round of training, is the benchmark noise intensity, is the gradient norm sensitivity parameter. The noise standard deviation is adaptively adjusted according to the gradient amplitude: when the noise intensity is reduced by the exponential decay factor : ; Preferably, the decay coefficient is set to the reciprocal of the current training round to ensure that the noise intensity approaches zero in the later stage of training. The sensitivity parameter ϵϵ is dynamically calibrated through the gradient statistical histogram and updated every 5 rounds of training.

[0045] When the edge server aggregates the confused gradients, a weighted average strategy with momentum acceleration is adopted: ; where is the data volume weight of device , represents the local sample quantity, is the momentum factor. The momentum factor is adaptively adjusted according to the variance of the aggregated gradients: ; The learning rate scheduling adopts a periodic cosine annealing strategy, and the The round learning rate is: ; where is the annealing period, preferably set to 1 / 5 of the total number of training rounds. When it is detected that the loss function plateau exceeds 3 cycles, the learning rate is automatically reset to to jump out of the local optimum.

[0046] The updated model parameters are transmitted to the tensor decomposition module through AES-256 encryption for reconstructing the core dimensions of the behavior tensor. The parameter injection process is achieved through tensor rank correction, and the update equation of the core tensor GG is defined as: ; where are the model parameters in the projection vector of the th modality, is the injection intensity coefficient. The projection vector is obtained through singular value decomposition: ; Preferably, the first principal components are retained, and the value is determined according to the cumulative contribution rate of eigenvalues ≥ 95%. The injection intensity is proportional to the corresponding singular value , and the proportionality factor decays linearly with the number of training rounds.

[0047] In terms of implementation details, local training adopts an asynchronous parallel architecture, and each device independently executes forward inference and backpropagation. The gradient aggregation server is deployed at the edge node and receives encrypted gradients through the gRPC protocol. The parameter injection module and the tensor decomposer share the memory space, and zero-copy data transmission is achieved through the DMA channel. The privacy protection strength is controlled by -differential privacy budget, and the privacy loss is calculated for each round of training: ; where is the noise standard deviation in the th round, is the failure probability. When exceeds the preset threshold, the training process is automatically terminated and the model reset mechanism is triggered.

[0048] S6. Based on the corrected display parameters and resource allocation scheme, drive the multi-screen system to perform dynamic rendering and content output.

[0049] To achieve dynamic rendering control for a multi-screen system, in this step, a virtual pixel resource pool is constructed and environment-aware tone mapping is implemented. The geometric mapping relationship generated in step S4 and the model parameters updated in step S5 are converted into actual rendering instructions. Specifically, first, a dynamic binding relationship between virtual pixels and physical screens is established. Subsequently, the ambient light field and biometric features are fused to perform tone compression. Finally, output stability is ensured through brightness constraints.

[0050] The construction of the virtual pixel resource pool is based on the resource allocation parameters of step S3, defining virtual pixels , where is the set of candidate physical screens. The alpha-channel blending weight of each virtual pixel is optimized by the following formula: ; where is the distance between the virtual pixel and the physical screen , is the network transmission delay, is the scaling factor. The delay is obtained by calculating the exponentially weighted moving average of the historical delay sequence through a sliding window: ; where is the current measured delay, is the forgetting factor. Preferably, the delay window length is set to 3 times the network round-trip time to balance the requirements of real-time performance and stability.

[0051] Environment-aware dynamic tone compression is jointly adjusted based on the ambient light field intensity and the user's pupil diameter . Define the high-dynamic range compression curve: ; where is the mean ambient light intensity, is the compression intensity coefficient. The coefficient is adaptively adjusted according to biometric features: ; where is the reference pupil diameter, is the adjustment sensitivity parameter. When a rapid contraction of the pupil diameter is detected, the value is automatically increased to enhance the dark details.

[0052] The physical screen brightness conservation constraint is achieved through normalization to ensure that the total output brightness does not exceed the hardware threshold . Define the normalization factor: ; where is the original brightness of the -th screen, and is a small constant to prevent division by zero. The final output brightness is scaled proportionally: ; The normalization factor is dynamically updated in the rendering pipeline. When a sudden change in ambient light intensity exceeds the threshold, an emergency brightness limit mechanism is triggered. This mechanism directly writes the brightness upper limit value through the hardware register, bypassing the software layer processing delay.

[0053] The implementation process is executed in the following time sequence: First, load the virtual pixel resource pool configuration and establish pixel binding relationships according to the geometric mapping matrix in step S4. Subsequently, collect the ambient light sensor data and the pupil diameter of the eye tracking module, and calculate the dynamic compression parameters. In the rendering pipeline, each virtual pixel performs tone compression and alpha blending calculations in parallel to generate an intermediate rendering result. Finally, apply the brightness normalization constraint and output to the physical screen through the DisplayPort1.4a interface.

[0054] In terms of hardware acceleration, the alpha blending calculation is implemented through the texture sampling unit of the GPU, and bilinear interpolation is used to optimize the spatial continuity. The tone compression curve is pre-computed through a look-up table (LUT) and stored in the video memory, and is quickly accessed through the texture buffer during real-time rendering. The brightness normalization factor is calculated through a dedicated circuit implemented by FPGA to ensure sub-microsecond response latency.

[0055] The triggering conditions for updating the dynamic mapping relationship include: the physical position of the screen changes by more than the threshold, the network delay volatility exceeds the set value, or the gradient of the ambient light intensity change reaches the critical point. During the update process, a double-buffer mechanism is adopted to avoid rendering tearing, and the old and new mapping matrices are switched through atomic operations.

[0056] Please refer to the appendix Figure 2 , the present invention also provides a multi-screen dynamic display control device, including: A multi-source data synchronous acquisition module for performing spatio-temporal alignment processing of user behavior data and environmental parameters through a heterogeneous sensor array; A spatio-temporal tensor modeling module for constructing and decomposing a seventh-order behavior tensor model based on the hybrid tensor decomposition algorithm; A quantum-classical collaborative optimization module for solving the screen resource allocation parameters with spatio-temporal constraints; A biophysical coupling correction module for dynamically generating the geometric mapping relationship of the display device according to the neural radiance field; A privacy gradient aggregation module for implementing adaptive noise injection and gradient confusion of the federated learning model; A dynamic resource rendering engine module for performing elastic pixel mapping and dynamic tone rendering in a multi-screen system.

[0057] The present invention also provides an electronic device, comprising: a processor; a memory storing a computer program, which when processed by the computer program, executes the method as described above.

[0058] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-screen dynamic display control method, characterized in that It includes the following steps: S1. Synchronously collect the multi-modal behavior data of the user and environmental parameters, and generate multi-source perception data with a unified benchmark through spatio-temporal alignment processing; S2. Based on the multi-source perception data, construct a seventh-order behavior tensor model containing spatio-temporal correlation features, and perform hybrid tensor decomposition to extract user intention features; S3. According to the feature parameters output by the behavior tensor model, solve the dynamic resource allocation parameters between multiple screens through a quantum-classical hybrid optimization algorithm; S4. Based on the resource allocation parameters, use a neural radiance field to dynamically correct the geometric parameters of the display device, and generate a display mapping relationship that conforms to the physical space topology; S5. According to the display mapping relationship, adopt a federated learning mechanism to update the multi-modal intention recognition model, and feedback the updated model parameters to the tensor decomposition process in step S2; S6. Based on the corrected display parameters and resource allocation scheme, drive the multi-screen system to perform dynamic rendering and content output.

2. The multi-screen dynamic display control method according to claim 1, characterized in that The step S1 includes the following steps: Capture the device motion attitude quaternion at a sampling rate through an inertial measurement unit, generate a resolution depth map through a ToF depth camera, and measure the pupil diameter change rate in real time through an eye tracking module; Adopt the IEEE1588v2 protocol to achieve multi-sensor clock synchronization, and construct a spatial transformation matrix chain including a translation vector and a rotation matrix.

3. The multi-screen dynamic display control method according to claim 1, characterized in that The hybrid tensor decomposition in the step S2 satisfies the following formula definition: CP decomposition term: ； Among them, is the th eigenvalue, represents the th factor vector of the th mode, represents the outer product operation of vectors; Tucker decomposition term: ； Among them, is the core tensor; is the factor matrix of the th modality, represents the th modality product of the tensor and the matrix.

4. The multi-screen dynamic display control method according to claim 1, wherein The quantum optimization process in the step S3 constructs the following Hamiltonian: ； Where: Indicates the screen The association strength with is determined by the Frobenius norm of the corresponding dimension of the behavior tensor; is the physical screen spacing; is the dynamic time delay threshold, which is adaptively adjusted according to the network load rate; , is the Pauli operator for the th qubit.

5. The multi-screen dynamic display control method according to claim 1, wherein The step S4 includes the following steps: Establish a neural radiance field model, and generate the screen pixel color through transmittance accumulation calculation, where the transmittance is related to the optical properties of the screen material; Dynamically adjust the update rate of the radiance field parameters according to the user's physiological characteristic signals collected in real time, where the physiological characteristic signals include at least one of pupil diameter and heart rate variability; Introduce a screen surface differential constraint term to ensure that the geometric correction process conforms to the physical space continuity condition.

6. The multi-screen dynamic display control method according to claim 1, characterized in that The step S5 includes the following steps: During the training process on the device side, inject adaptive noise that is non-linearly related to the gradient norm into the model gradient, where the noise intensity decays exponentially as the gradient amplitude increases; When aggregating the confused gradients on the edge server, adopt a periodic decay strategy to dynamically adjust the learning rate, where the learning rate remains high in the initial stage of training to accelerate convergence and gradually decreases in the later stage to improve the model stability; Encrypt and transmit the updated model parameters to the tensor decomposition in step S2 for reconstructing the core dimension of the behavior tensor.

7. The multi-screen dynamic display control method according to claim 1, characterized in that The step S6 includes the following steps: Construct a virtual pixel resource pool, define the dynamic mapping relationship between virtual pixels and multi-physical screens, where the alpha channel blending weight of each virtual pixel is jointly optimized according to the physical distance between the screens and the network delay; Perform environment-aware dynamic tone compression, and jointly adjust the HDR compression curve based on the ambient light field intensity distribution and the user's biometric signals, where the compression intensity is positively correlated with the mean ambient light intensity; Implement the physical screen brightness conservation constraint, and ensure that the sum of the brightness outputs of each screen pixel does not exceed a preset threshold through normalization processing.

8. The multi-screen dynamic display control method according to claim 4, wherein The screen correlation degree is calculated as follows: ； wherein is an adjustable integral time window, represents a slice of the tensor in the screen dimension.

9. A multi-screen dynamic display control device, characterized by the multi-screen dynamic display control method according to any one of claims 1-8. Including: A multi-source data synchronous acquisition module for performing spatio-temporal alignment processing of user behavior data and environmental parameters through a heterogeneous sensor array; A spatio-temporal tensor modeling module for constructing and decomposing a seventh-order behavior tensor model based on a hybrid tensor decomposition algorithm; A quantum-classical collaborative optimization module for solving screen resource allocation parameters with spatio-temporal constraints; A biophysical coupling correction module for dynamically generating the geometric mapping relationship of a display device according to a neural radiance field; A privacy gradient aggregation module for implementing adaptive noise injection and gradient obfuscation of a federated learning model; A dynamic resource rendering engine module for performing elastic pixel mapping and dynamic tone rendering of a multi-screen system.

10. An electronic device, characterized in that, Including: A processor; A memory storing a computer program; When the processor executes the program, the method described in any one of claims 1-8 is implemented.

Images