Video stream image optimization method and device

Abstract

According to the video stream image optimization method and device provided by the embodiment of the invention, a dynamic fuzzy kernel generation mechanism is innovatively designed, and the data processing security is ensured through space-time decoupling and physical constraint. A content-motion double-branch network is constructed, and a reliable feature reconstruction system is established in combination with residual connection and space-time transformation. A physical constraint corrector is introduced, and high-quality video output is provided while user privacy is protected through adaptive fusion and rule correction. According to the method, the defects of the traditional technology in the aspects of fuzzy processing, feature decoupling, image optimization and the like are effectively overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing, in particular to a video stream image optimization method and device. BACKGROUND

[0002] The existing video stream image optimization method has obvious shortcomings. The traditional system performs poorly in dynamic blur processing, and fails to effectively combine optical flow and motion features, affecting image quality.

[0003] In addition, the existing technology has a bottleneck in space-time feature decoupling. Most systems lack a perfect content-motion separation mechanism and physical constraint strategy, resulting in unsatisfactory motion area reconstruction.

[0004] The existing system has technical shortcomings in smear and ghost elimination. Lack of sufficient consideration of physical laws, it is difficult to achieve accurate image optimization through feature fusion, affecting visual experience. The solution to these problems is of great significance to improve video quality. SUMMARY

[0005] In view of the problems in the prior art, the present application provides a video stream image optimization method and device, which can effectively solve the shortcomings of traditional technology in blur processing, feature decoupling and image optimization.

[0006] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a video stream image optimization method, comprising:

[0008] Generating a dynamic blur kernel according to the motion speed of the video image and the target object and the optical flow parameter, modeling the optical flow and detecting the occlusion of the video image by using a plurality of convolution units, inputting the video image and the blur kernel into a physical perception module for data preprocessing to generate standardized features with motion trajectory constraints;

[0009] Inputting the standardized features into a space-time decoupling network, the space-time decoupling network including a content branch and a motion branch, the content branch reconstructing static region texture features, the motion branch reconstructing dynamic region motion features, and the reconstruction results of the content branch and the motion branch being fused based on a cross-layer feature connection mechanism to generate space-time decoupling features;

[0010] Inputting the space-time decoupling features into a physical constraint corrector, the physical constraint corrector generating a dynamic weight matrix according to the motion trajectory constraints, adaptively fusing and correcting the physical laws of the space-time decoupling features, and optimizing the corrected features through a residual connection and a space-time transformation network to generate a video output eliminating smear and ghost.

[0011] Further, the method further comprises: performing multi-scale feature matching on the continuous video frames based on a pyramid optical flow algorithm, constructing a motion trajectory matrix of the target object, inputting the motion trajectory matrix into a multi-layer convolutional neural network for feature extraction, the multi-layer convolutional neural network comprising a spatial attention module and a channel attention module, and calculating a dynamic blur kernel parameter based on a spatiotemporal correlation of the motion trajectory matrix.

[0012] The method further comprises: constructing an optical flow model using multiple hollow convolution units, the hollow convolution units comprising a residual connection layer and a normalization layer, performing frame-by-frame optical flow estimation on the video image sequence, generating an occlusion detection map based on consistency analysis of an optical flow vector field and a motion trajectory matrix, and combining the occlusion detection map and the dynamic blur kernel parameter by weighting to obtain a dynamic blur kernel considering the occlusion relationship.

[0013] Further, the method further comprises: inputting the video image and the dynamic blur kernel into a physical perception module, the physical perception module comprising multiple deconvolution layers and up-sampling layers, performing reverse deblurring processing on the video image based on the dynamic blur kernel, the deconvolution layers iteratively reconstructing image features, and the up-sampling layers improving the resolution of the reconstructed features to generate an initial feature map.

[0014] The method further comprises: constructing a motion trajectory constraint model using a bidirectional long short-term memory network, the motion trajectory constraint model comprising a forward propagation layer and a backward propagation layer, inputting the initial feature map into the motion trajectory constraint model, the forward propagation layer extracting forward temporal features, the backward propagation layer extracting backward temporal features, constructing a trajectory constraint condition based on a corresponding relationship between the forward temporal features and the backward temporal features, and performing batch normalization processing on the initial feature map to generate a standardized feature with a motion trajectory constraint.

[0015] Further, the method further comprises: performing static region segmentation on the standardized feature, and constructing a non-local attention module, the non-local attention module comprising a feature transformation subnetwork and a similarity calculation subnetwork, the feature transformation subnetwork mapping the standardized feature to a high-dimensional feature space, and the similarity calculation subnetwork calculating a similarity matrix of each pixel point in the feature map based on a cosine distance to generate a static region feature map.

[0016] The method further comprises: inputting the static region feature map into a texture reconstruction network, the texture reconstruction network comprising multiple densely connected modules and skip connection modules, the densely connected modules extracting multi-level texture information using multi-scale convolution kernels, the skip connection modules fusing texture information at different levels, and performing detail enhancement on the fused features based on a self-attention mechanism to generate reconstructed static region texture features.

[0017] Further, a multi-scale motion encoder is constructed, which includes a pyramid pooling module and a dynamic convolution module, the pyramid pooling module performs multi-scale decomposition on the normalized features, and the dynamic convolution module adaptively adjusts the convolution kernel parameters based on the motion amplitude to reconstruct the features of the dynamic region and generate dynamic region motion features;

[0018] The static region texture features and the dynamic region motion features are input into a feature fusion network, which includes a channel recalibration module and a spatial reweighting module, the channel recalibration module adaptively selects the feature channels, the spatial reweighting module adjusts the feature weights based on the position sensitivity map, the recalibrated features are concatenated and combined through cross-layer residual connection, and spatio-temporal decoupling features are generated.

[0019] Further, a motion physical constraint model is constructed, which includes an acceleration analysis module and a speed prediction module, the acceleration analysis module calculates the acceleration distribution of the target object based on the motion trajectory constraint, and the speed prediction module predicts the motion state at the next time according to the acceleration distribution to generate physical constraint parameters.

[0020] The spatio-temporal decoupling features and the physical constraint parameters are input into a weight generation network, which includes a feature correlation module and a weight calculation module, the feature correlation module performs temporal consistency analysis on the spatio-temporal decoupling features, the weight calculation module evaluates and scores the feature consistency based on the physical constraint parameters, converts the evaluation score into a weight coefficient through normalization processing, and generates a dynamic weight matrix.

[0021] Further, the spatio-temporal decoupling features and the dynamic weight matrix are input into a feature correction network, which includes an adaptive fusion module and a physical constraint module, the adaptive fusion module performs weighted combination on the spatio-temporal decoupling features based on the dynamic weight matrix, and the physical constraint module performs regularity correction on the fused features according to the motion trajectory constraint to generate a corrected feature map.

[0022] An optimization enhancement network is constructed, which includes a residual learning module and a spatio-temporal transformation module, the residual learning module performs detail compensation on the corrected feature map, the spatio-temporal transformation module performs spatial alignment and temporal smoothing on the compensated features based on deformable convolution, generates optimized video features through multi-layer receptive field fusion and feature re-calibration, and performs deconvolution reconstruction on the optimized video features to obtain a video output with eliminated ghosting and ghosting.

[0023] In a second aspect, the application provides a video stream image optimization device, which comprises:

[0024] The preprocessing module is configured to generate a dynamic blur kernel according to a video image and a motion speed and an optical flow parameter of a target object, model an optical flow of the video image by using a plurality of convolution units, and detect occlusion, input the video image and the blur kernel into a physical perception module for data preprocessing, and generate a standardized feature with a motion trajectory constraint;

[0025] The space-time decoupling module is configured to input the standardized feature into a space-time decoupling network, the space-time decoupling network includes a content branch and a motion branch, the content branch is configured to reconstruct a static region texture feature, the motion branch is configured to reconstruct a dynamic region motion feature, and a reconstruction result of the content branch and the motion branch is fused based on a cross-layer feature connection mechanism to generate a space-time decoupled feature;

[0026] The image optimization module is configured to input the space-time decoupled feature into a physical constraint corrector, the physical constraint corrector is configured to generate a dynamic weight matrix according to the motion trajectory constraint, adaptively fuse and correct the space-time decoupled feature according to a physical law, and optimize the corrected feature by using a residual connection and a space-time transformation network to generate a video output in which ghosting and smearing are eliminated.

[0027] In a third aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the video stream image optimization method when executing the program.

[0028] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the video stream image optimization method.

[0029] In a fifth aspect, the present application provides a computer program product, including a computer program / instruction, wherein the computer program / instruction is executable by a processor to implement the steps of the video stream image optimization method.

[0030] According to the above technical solution, the present application provides a video stream image optimization method and device, which generates a dynamic blur kernel by an innovative design, ensures data processing safety by space-time decoupling and physical constraints, constructs a content-movement dual-branch network, combines residual connection and space-time transformation, establishes a reliable feature reconstruction system, introduces a physical constraint corrector, adaptively fuses and corrects the features according to a law, protects user privacy while providing high-quality video output, and effectively solves the deficiencies of traditional technologies in blur processing, feature decoupling, and image optimization. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0032] Figure 1 The flowchart of the video stream image optimization method in the embodiments of the present application;

[0033] Figure 2 The structural diagram of the video stream image optimization device in the embodiments of the present application;

[0034] Figure 3 The structural diagram of the electronic device in the embodiments of the present application.

[0035] Reference signs:

[0036] Electronic device 9600, central processor 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage part 9142, data storage part 9143, driver program storage part 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0038] The acquisition, storage, use, processing and the like of data in the technical solutions of the present application all conform to the relevant provisions of national laws and regulations.

[0039] In view of the problems in the prior art, the present application provides a video stream image optimization method and device, which innovatively designs a dynamic blur kernel generation mechanism, ensures data processing safety through space-time decoupling and physical constraints, constructs a content-motion double-branch network, combines residual connection and space-time transformation, establishes a reliable feature reconstruction system, introduces a physical constraint corrector, and provides high-quality video output while protecting user privacy through adaptive fusion and rule correction. This method effectively solves the deficiencies of traditional technologies in blur processing, feature decoupling and image optimization.

[0040] In order to effectively solve the deficiencies of traditional technologies in blur processing, feature decoupling and image optimization, etc., an embodiment of a video stream image optimization method is provided, as shown in Figure 1 , which specifically includes the following contents:

[0041] Step S101: generating a dynamic blur kernel according to the motion speed of the video image and the target object and the optical flow parameter, using a plurality of convolution units to model the optical flow and detect occlusions of the video image, inputting the video image and the blur kernel into a physical perception module for data preprocessing to generate standardized features with motion trajectory constraints;

[0042] Optionally, the embodiment innovatively designs a set of feature processing scheme based on physical perception to solve the problems of motion target blur, complex occlusion relationship and inaccurate feature expression in video image processing. In the aspect of feature processing evaluation, the embodiment designs a blur kernel scoring formula: Blur_Score = a x (Motion_Speed) + b x (Flow_Consistency) + g x (Occlusion_Weight), wherein a, b and g are weight coefficients, respectively representing the influence of motion speed, optical flow consistency and occlusion degree. The feature quality score is Feature_Score = (Physical_Constraint x Temporal_Consistency) / (Spatial_Error + Noise_Level), which is used to evaluate the expression ability of the standardized features.

[0043] The embodiment first realizes accurate modeling of motion blur through a deep-optimized dynamic blur kernel generation mechanism. When processing video images, the system analyzes the motion characteristics of the target object, including speed size, direction change and acceleration distribution. For fast-moving targets, the generation process of the blur kernel considers the nonlinear features of the motion trajectory, and by establishing a speed-blur mapping relationship, the shape and size of the blur kernel are dynamically adjusted. When processing complex scenes, the system captures the motion details of the target, including local deformation and overall displacement, by analyzing the optical flow parameters. The design of the blur kernel adopts an adaptive strategy to dynamically adjust the intensity of the blur effect according to the motion state of the target. For multi-target scenes, the system describes the motion dependency relationship between different targets by establishing a target motion correlation graph, ensuring the consistency of the blur kernel generation. This optimized blur kernel generation mechanism significantly improves the expression accuracy of motion blur.

[0044] The embodiment innovatively realizes the optical flow modeling and occlusion detection mechanism. The design of multiple convolution units adopts a hierarchical structure, and different levels of convolution units are responsible for feature extraction of different scales. In the optical flow modeling process, the system constructs a pixel-level motion vector field through spatial gradient analysis and temporal correlation calculation. The design of the convolution unit adopts a multi-scale strategy, and the shallow convolution unit is responsible for extracting local motion features, and the deep convolution unit focuses on global motion patterns. For the detection of occluded areas, the system identifies potential occlusion boundaries by analyzing the discontinuity and motion consistency of the optical flow field. Temporal constraints are introduced in the occlusion detection process to improve the accuracy of occlusion detection by analyzing the feature correspondence between consecutive frames. This optimized feature extraction mechanism provides reliable input for subsequent data preprocessing.

[0045] The embodiment realizes the standardization of features through an optimized physical perception mechanism. The physical perception module adopts a multi-level feature extraction strategy when processing input data. For video images, the system describes the motion law of the target by establishing a physical model, including inertia, resistance and external force influence. The application process of the blur kernel considers the physical constraints to ensure that the deblurring result conforms to the actual motion characteristics. Motion trajectory constraints are introduced in the feature extraction process to constrain the features in terms of temporal consistency by establishing a trajectory prediction model. This optimized physical perception mechanism significantly improves the physical reasonableness of the features.

[0046] The embodiment establishes a complete feature processing framework. The generation of standardized features adopts a multi-stage optimization strategy, first analyzing the motion characteristics of the target through a physical model, and then constructing a feature extraction network based on the motion characteristics. The system establishes a feature quality evaluation mechanism to monitor the expression effect of the features in real time, supporting dynamic adjustment of the feature extraction strategy. This optimized feature processing mechanism provides high-quality feature support for subsequent video processing tasks. The embodiment provides an innovative solution for feature processing, through optimized blur kernel generation and physical perception, the system realizes the whole process optimization from raw video to standardized features. This reliable processing mechanism significantly improves the quality of the features, especially in handling fast motion and complex occlusion scenes, the generated features exhibit stronger physical consistency and temporal correlation.

[0047] The innovation of the embodiment mainly lies in the adaptive generation of dynamic blur kernels and the physical constraint expression of features. Through multi-level feature extraction and physical perception mechanisms, the system realizes accurate modeling of motion blur and standardized expression of features, providing reliable feature support for video image processing tasks. Especially in high-speed motion scenes, this feature processing method based on physical constraints significantly improves the robustness and accuracy of the system.

[0048] Step S102: inputting the standardized features into a space-time decoupling network, the space-time decoupling network comprising a content branch and a motion branch, the content branch being configured to reconstruct static region texture features, the motion branch being configured to reconstruct dynamic region motion features, and a cross-layer feature connection mechanism being configured to fuse reconstruction results of the content branch and the motion branch to generate space-time decoupled features;

[0049] Optionally, the embodiment is directed to the problems of feature confusion between static regions and dynamic regions, complex space-time feature correlation, and insufficient feature fusion in video processing, and a set of space-time decoupling network based on double branches is innovatively designed. In terms of feature decoupling evaluation, the embodiment designs a decoupling quality score formula: Decoupling_Score = a x (Static_Fidelity) + β x (Motion_Accuracy) + γ x (Feature_Independence), wherein a, β, and γ are weight coefficients, respectively representing static fidelity, motion accuracy, and feature independence. The feature fusion score is Fusion_Score = (Cross_Layer_Correlation x Feature_Complementarity) / (Information_Loss + Interference_Level), which is used to evaluate the effect of feature fusion.

[0050] The embodiment first realizes accurate reconstruction of static regions through a deeply optimized content branch. When processing standardized features, the content branch adopts a multi-level feature extraction architecture, and different levels are responsible for capturing texture information of different scales. For the identification of static regions, the system analyzes the temporal consistency and spatial correlation of the features to construct a static region confidence map. An adaptive receptive field mechanism is introduced in the feature extraction process, which dynamically adjusts the size of the convolution kernel to accurately capture texture of different scales. When processing complex textures, the system establishes a multi-scale feature pyramid to realize hierarchical expression of texture details. The reconstruction process adopts an iterative optimization strategy, and each iteration is based on the difference between the current reconstruction result and the original features to dynamically adjust the reconstruction parameters. In order to maintain the continuity of the texture, the system introduces a local consistency constraint to ensure that the reconstruction result maintains smooth transition in the local region. This optimized static feature reconstruction mechanism significantly improves the accuracy of texture expression.

[0051] The embodiment innovatively realizes the feature reconstruction mechanism of the motion branch. When processing dynamic regions, the motion branch adopts a time-aware feature extraction strategy. For the extraction of motion features, the system constructs a motion feature field by analyzing the feature changes between consecutive frames. The feature extraction network adopts a multi-resolution design, with low-resolution layers responsible for capturing overall motion trends and high-resolution layers focusing on local motion details. When processing complex motion, the system decomposes complex motion into basic motion components by establishing a motion decomposition model. Motion continuity constraints are introduced in the reconstruction process to ensure the temporal smoothness of the reconstruction results. For fast motion regions, the system increases the temporal resolution of feature extraction through an adaptive sampling strategy to improve the accuracy of motion capture. This optimized dynamic feature reconstruction mechanism provides reliable motion expression for subsequent feature fusion.

[0052] The embodiment realizes effective feature fusion through an optimized cross-layer feature connection mechanism. The feature fusion process adopts a multi-level connection strategy, and different levels of features are combined through adaptive weights. The design of the fusion network considers the complementarity of features, and guides the selective fusion of features by establishing a feature correlation graph. To reduce information redundancy, the system introduces a feature screening mechanism to retain the most discriminative feature combinations. A residual learning strategy is used in the fusion process to maintain the original feature information through a skip connection. This optimized fusion mechanism significantly improves the expression ability of spatio-temporal decoupled features.

[0053] The embodiment establishes a complete feature decoupling framework. The generation of spatio-temporal decoupled features adopts a multi-stage optimization strategy, first extracting static and dynamic features through a dual-branch network, and then realizing effective feature fusion through a cross-layer connection mechanism. The system establishes a feature quality evaluation mechanism to monitor the decoupling effect in real time and supports dynamic adjustment of network parameters. This optimized decoupling mechanism provides high-quality feature support for subsequent video processing tasks. The embodiment provides an innovative solution for feature decoupling, through optimized dual-branch networks and feature fusion, the system realizes the whole process optimization from standardized features to spatio-temporal decoupled features. This reliable decoupling mechanism significantly improves the expression ability of features, especially when processing complex scenes, decoupled features exhibit stronger discriminability and adaptability.

[0054] The innovation of the embodiment mainly lies in the decoupled expression of spatio-temporal features and the adaptive fusion of features. By constructing a dual-branch network and a cross-layer connection mechanism, the system realizes accurate modeling of static and dynamic regions, providing reliable feature support for video processing tasks. Especially when processing scenes with complex motion patterns, this decoupled feature processing method significantly improves the expression ability and processing accuracy of the system.

[0055] Step S103: inputting the space-time decoupling features into a physical constraint modifier, generating a dynamic weight matrix according to the motion trajectory constraint, adaptively fusing the space-time decoupling features and correcting the physical law, optimizing the corrected features through a residual connection and a space-time transformation network, and generating a video output eliminating ghosting and ghosting.

[0056] Optionally, the embodiment is directed to the problem of ghosting and ghosting in video processing, and a set of feature correction scheme based on physical constraints is innovatively designed. In the aspect of feature correction evaluation, the embodiment designs a correction effect score formula: Correction_Score = a x (Physics_Constraint) + b x (Feature_Consistency) + g x (Motion_Smoothness), wherein a, b and g are weight coefficients, respectively representing the satisfaction degree of physical constraint, feature consistency and motion smoothness. The output quality score is: Quality_Score = (Detail_Preservation x Artifact_Removal) / (Motion_Blur + Ghost_Effect), which is used to evaluate the quality of the video output.

[0057] The embodiment first realizes accurate correction of features through a deep optimized physical constraint correction mechanism. When processing the space-time decoupling features, the physical constraint modifier adopts a multi-level constraint modeling strategy. According to the motion trajectory constraint obtained in the previous step, the modifier first establishes a physical motion model to describe the motion law of the target. For high-speed moving targets, the system predicts the motion state of the target by analyzing the acceleration distribution and speed change. The construction of physical constraint conditions considers multiple aspects, including motion continuity, physical feasibility and environmental interaction. Based on these constraint conditions, the system generates a dynamic weight matrix for adaptive fusion of features. The generation process of the weight matrix adopts an attention mechanism, which dynamically adjusts the weight distribution of different regions by calculating the correlation between features and physical constraints. This optimized constraint modeling mechanism significantly improves the accuracy of feature correction.

[0058] The embodiment innovatively realizes feature fusion and correction mechanism. In the adaptive fusion stage, the system combines the space-time decoupling features through a dynamic weight matrix. The fusion process considers the spatial correlation and temporal consistency of the features, and describes the dependence between different features by establishing a feature correlation graph. For the area that may produce smearing, the system dynamically adjusts the fusion weight by analyzing the motion trajectory and velocity field, and suppresses the smearing effect. The physical law correction process adopts an iterative optimization strategy, and each iteration adjusts the features based on physical constraints. Time smoothing constraints are introduced in the correction process to ensure the time continuity of the correction results. This optimized fusion mechanism provides a reliable foundation for subsequent feature optimization.

[0059] The embodiment realizes the final optimization of features through an optimized network structure. The design of residual connection and space-time transformation network adopts a multi-branch architecture. The residual learning branch is responsible for compensating and enhancing the detailed information, and preserves the high-frequency details of the image by establishing a residual mapping relationship. The space-time transformation branch realizes the spatial alignment and temporal smoothing of features through deformable convolution. During the transformation process, the system adopts a multi-scale receptive field fusion strategy to capture image information at different scales. The feature re-calibration process considers physical constraints to ensure that the optimization results comply with the actual motion laws. This optimized network structure significantly improves the quality of video output.

[0060] The embodiment establishes a complete video optimization framework. The generation of video output adopts a multi-stage optimization strategy, first correcting the features through a physical constraint modifier, and then enhancing the details through an optimization network. The system establishes a video quality evaluation mechanism to monitor the output effect in real time, supporting dynamic adjustment of the optimization strategy. This optimized processing mechanism significantly improves the video quality, especially in handling complex motion scenes, the output results show better visual effect and physical rationality. The embodiment provides an innovative solution for video optimization, through optimized physical constraints and feature correction, the system realizes the whole process optimization from feature processing to video output. This reliable optimization mechanism significantly improves the video quality, especially in handling fast motion and complex scenes, the output video shows clearer details and fewer artifacts.

[0061] The innovation of the embodiment mainly lies in the dynamic application of physical constraints and the adaptability of feature optimization. Through the physical constraint modifier and the optimization network, the system realizes feature correction and video quality improvement based on physical laws, providing reliable optimization support for video processing tasks. Especially in handling high-speed motion scenes, this optimization method based on physical constraints significantly improves the performance of the system, effectively eliminating visual artifacts such as smearing and ghosting. For example, when processing sports competition videos, the system can accurately capture the rapid movements of athletes while maintaining the clarity and continuity of the picture.

[0062] From the above description, the video stream image optimization method provided by the embodiments of the present application can ensure data processing security through the innovative design of the dynamic blur kernel generation mechanism, through space-time decoupling and physical constraints. A content-motion double-branch network is constructed, residual connection and space-time transformation are combined, and a reliable feature reconstruction system is established. A physical constraint modifier is introduced, which provides high-quality video output while protecting user privacy through adaptive fusion and regular correction. This method effectively solves the deficiencies of traditional technologies in blur processing, feature decoupling, and image optimization.

[0063] In an embodiment of the video stream image optimization method of the present application, the following content can also be included:

[0064] Step S201: Perform multi-scale feature matching on consecutive video frames based on a pyramid optical flow algorithm, construct a motion trajectory matrix of the target object, input the motion trajectory matrix into a multi-layer convolutional neural network for feature extraction, the multi-layer convolutional neural network includes a spatial attention module and a channel attention module, calculate dynamic blur kernel parameters based on the space-time correlation of the motion trajectory matrix;

[0065] Step S202: Construct an optical flow model using multiple atrous convolution units, the atrous convolution units include a residual connection layer and a normalization layer, perform frame-by-frame optical flow estimation on the video image sequence, generate an occlusion detection map based on the consistency analysis of the optical flow vector field and the motion trajectory matrix, and combine the occlusion detection map and the dynamic blur kernel parameters to obtain a dynamic blur kernel considering the occlusion relationship.

[0066] Optionally, the embodiments of the present application innovatively design a feature processing scheme based on multi-scale optical flow to solve the problems of inaccurate motion target tracking, imperfect occlusion detection, and unreasonable blur kernel generation in traditional video processing. In the evaluation of optical flow estimation, the present embodiment designs a matching quality scoring formula: Flow_Score = α × (Feature_Similarity) + β × (Motion_Consistency) + γ × (Scale_Coverage), where α, β, and γ are weight coefficients, representing feature similarity, motion consistency, and scale coverage, respectively. The occlusion detection score is: Occlusion_Score = (Flow_Discontinuity × Trajectory_Consistency) / (Noise_Level + Estimation_Error), which is used to evaluate the accuracy of occlusion detection.

[0067] The embodiment first realizes multi-scale feature matching through a depth-optimized pyramid optical flow algorithm. When processing consecutive video frames, the system constructs a multi-level image pyramid, with each level corresponding to a different spatial resolution. For the tracking of moving targets, the system adopts a top-down matching strategy, first performing coarse matching at the low-resolution level to obtain initial motion estimation, and then gradually performing fine matching at the high-resolution level. Local deformation constraints are introduced during feature matching, and a local affine transformation model is established to describe the non-rigid deformation of the target. For fast-moving targets, the system dynamically adjusts the search range of feature matching through an adaptive search window mechanism. The construction of the motion trajectory matrix adopts temporal consistency constraints, and by analyzing the correspondence between features in consecutive frames, the continuity of the trajectory is ensured. This optimized feature matching mechanism significantly improves the accuracy of motion estimation.

[0068] The embodiment innovatively realizes an attention-enhanced feature extraction mechanism. When processing the motion trajectory matrix, the multi-layer convolutional neural network adopts an architecture that combines spatial attention and channel attention. The spatial attention module adaptively highlights the feature expression of important regions by calculating the spatial response map of the feature map. For the positioning of moving targets, the attention mechanism focuses on the spatial distribution and local structural features of the target. The channel attention module enhances discriminative motion features by learning the dependency between feature channels. When dealing with complex backgrounds, the attention mechanism can effectively suppress background interference and improve the extraction quality of target features. The calculation of dynamic blur kernel parameters is based on spatio-temporal correlation analysis, and by establishing a motion state transition model, the motion trend of the target is predicted. This optimized feature extraction mechanism provides reliable feature support for subsequent optical flow estimation.

[0069] The embodiment realizes accurate optical flow estimation through an optimized dilated convolution network. The construction of the optical flow model adopts multiple dilated convolution units, with different dilation rate convolution kernels responsible for capturing motion information at different scales. The design of the residual connection layer ensures efficient transmission of feature information, and the jump connection mechanism reduces the problem of gradient disappearance. The normalization layer adopts a batch normalization strategy to improve the training stability of the network by standardizing the feature distribution. When performing frame-by-frame optical flow estimation, the system establishes temporal consistency constraints to ensure the temporal continuity of the optical flow field. The implementation of occlusion detection is based on the consistency analysis of the optical flow vector field and the motion trajectory matrix, and by identifying the discontinuous regions of the optical flow field and the breakpoints of the trajectory, an occlusion detection map is generated. The final dynamic blur kernel is generated through a weighted combination mechanism, which integrates the occlusion relationship into the generation process of the blur kernel. This optimized optical flow estimation mechanism significantly improves the accuracy of occlusion detection.

[0070] The embodiment establishes a complete fuzzy kernel generation framework. The dynamic fuzzy kernel generation considering the occlusion relationship adopts a multi-stage optimization strategy. First, motion information is obtained through multi-scale feature matching. Then, the fuzzy kernel is accurately modeled based on optical flow estimation and occlusion detection. The system monitors the generation effect in real time by establishing a fuzzy kernel quality evaluation mechanism, and supports dynamic adjustment of parameters. This optimized generation mechanism provides high-quality fuzzy kernel expression for subsequent video processing tasks. The embodiment provides an innovative solution for fuzzy kernel generation. Through optimized feature matching and optical flow estimation, the system realizes the whole process optimization from video sequence to dynamic fuzzy kernel. This reliable generation mechanism significantly improves the quality of the fuzzy kernel, especially in complex motion and occlusion scenes, the generated fuzzy kernel shows stronger adaptability and accuracy.

[0071] The innovation of the embodiment mainly lies in the adaptive processing of multi-scale feature matching and the accurate modeling of occlusion relationship. By constructing a multi-level feature processing network and an optimized optical flow estimation model, the system realizes accurate tracking of moving targets and accurate detection of occlusion relationship, providing reliable fuzzy kernel support for video deblurring tasks. Especially in complex scenes, this multi-scale based feature processing method significantly improves the robustness and accuracy of the system.

[0072] In an embodiment of the video stream image optimization method of the present application, the following contents can also be specifically included:

[0073] Step S301: input the video image and the dynamic fuzzy kernel into a physical perception module, the physical perception module includes multiple deconvolution layers and up-sampling layers, perform reverse deblurring processing on the video image based on the dynamic fuzzy kernel, the deconvolution layer iteratively reconstructs image features, and the up-sampling layer improves the resolution of the reconstructed features to generate an initial feature map;

[0074] Step S302: construct a motion trajectory constraint model using a bidirectional long short-term memory network, the motion trajectory constraint model includes a forward propagation layer and a backward propagation layer, input the initial feature map into the motion trajectory constraint model, the forward propagation layer extracts forward time sequence features, the backward propagation layer extracts backward time sequence features, construct a trajectory constraint condition based on the corresponding relationship between the forward time sequence features and the backward time sequence features, perform batch normalization processing on the initial feature map to generate a standardized feature with motion trajectory constraint.

[0075] Optionally, the embodiment is directed to the problem of inaccurate feature reconstruction, insufficient temporal constraints, and unreasonable feature standardization in video deblurring processing, and innovatively designs a set of feature processing scheme based on physical perception. In the aspect of feature processing evaluation, the embodiment designs a deblurring effect scoring formula: Deblur_Score = a x (Feature_Reconstruction) + β x (Resolution_Enhancement) + γ x (Physical_Consistency), wherein a, β, γ are weight coefficients, respectively representing feature reconstruction quality, resolution enhancement effect and physical consistency. The feature standardization score is: Normalization_Score = (Temporal_Consistency x Trajectory_Constraint) / (Feature_Distortion + Noise_Level), which is used to evaluate the quality of standardized features.

[0076] The embodiment first realizes video deblurring processing through a deeply optimized physical perception module. When processing input data, the physical perception module adopts a multi-level feature reconstruction strategy. The design of the deconvolution layer is based on a deep learning architecture, which realizes the inverse recovery of blurred features through multi-level feature mapping. Each deconvolution layer is equipped with an attention mechanism, which dynamically adjusts the reconstruction weight by calculating the feature response map. The iterative reconstruction process adopts a progressive strategy, focusing on restoring main structural features in the initial stage, and gradually optimizing detailed information in subsequent iterations. For complex blur scenarios, the system analyzes the characteristics of the dynamic blur kernel and adaptively adjusts the deconvolution parameters. Physical constraints are introduced in the reconstruction process to ensure that the reconstruction result conforms to the actual imaging process. The up-sampling layer uses a learnable up-sampling kernel to improve the resolution through multi-scale feature fusion. This optimized deblurring mechanism significantly improves the quality of feature reconstruction.

[0077] The embodiment innovatively realizes a motion trajectory constraint mechanism. The design of the bidirectional long short-term memory network adopts a symmetric architecture, with the forward propagation layer and the backward propagation layer processing the forward and backward dependency relationships of the temporal information, respectively. The forward propagation layer selectively retains and updates the temporal features through the gating mechanism, focusing on the continuity and predictability of motion. The backward propagation layer supplements the causal relationship of motion through reverse analysis, enhancing the integrity of trajectory constraints. The network training process introduces a temporal consistency loss, which compares the corresponding relationship of the forward and backward features to optimize the feature extraction effect. For complex motion scenarios, the system establishes a multi-scale feature pyramid to capture features of different time scales. This optimized constraint mechanism provides reliable temporal support for feature standardization.

[0078] The embodiment realizes the standard expression of features through an optimized feature normalization mechanism. The batch normalization process considers the statistical characteristics and physical constraints of features in the normalization process. The calculation of the normalization parameter adopts an adaptive strategy to dynamically adjust the normalization coefficient according to the feature distribution. In order to maintain the discriminability of the features, the system introduces local response normalization, which preserves local feature differences while maintaining global consistency. The trajectory constraint condition in the normalization process is realized by establishing a feature correspondence graph, ensuring that the normalization result meets the motion continuity requirement. This optimized normalization mechanism significantly improves the expression ability of features.

[0079] The embodiment establishes a complete feature processing framework. The generation of normalized features adopts a multi-stage optimization strategy, first reconstructing features through a physical perception module, and then normalizing features through motion trajectory constraints. The system establishes a feature quality evaluation mechanism to monitor the processing effect in real time and supports dynamic adjustment of the processing strategy. This optimized processing mechanism provides high-quality feature support for subsequent video processing tasks. The embodiment provides an innovative solution for feature processing, which realizes the whole-process optimization from blurred images to normalized features through optimized physical perception and trajectory constraints. This reliable processing mechanism significantly improves the quality of features, especially in processing complex motion scenes, the generated features exhibit stronger physical reasonableness and temporal consistency.

[0080] The innovation of the embodiment mainly lies in the feature reconstruction of physical perception and the constraint expression of motion trajectory. Through multi-level feature processing and bidirectional temporal modeling, the system realizes accurate reconstruction of blurred features and effective constraint of motion features, providing reliable feature support for video processing tasks. Especially in processing high-speed motion scenes, this feature processing method based on physical constraints significantly improves the reconstruction quality and temporal consistency of the system.

[0081] In an embodiment of the video stream image optimization method of the present application, the following contents can also be specifically included:

[0082] Step S401: Perform static region segmentation on the normalized features, and construct a non-local attention module, which includes a feature transformation subnetwork and a similarity calculation subnetwork. The feature transformation subnetwork maps the normalized features to a high-dimensional feature space, and the similarity calculation subnetwork calculates a similarity matrix of each pixel point in the feature map based on the cosine distance, generating a static region feature map.

[0083] Step S402: input the static region feature map into a texture reconstruction network, the texture reconstruction network comprising a plurality of densely connected modules and a jump connected module, the densely connected module extracting multi-level texture information using multi-scale convolution kernels, the jump connected module performing feature fusion on texture information at different levels, performing detail enhancement on the fused features based on a self-attention mechanism, and generating a reconstructed static region texture feature.

[0084] Optionally, the embodiment is directed to the problems of inaccurate static region segmentation, incomplete texture reconstruction, and insufficient feature expression in video processing, and innovatively designs a set of static feature reconstruction scheme based on non-local attention. In terms of feature reconstruction evaluation, the embodiment designs a reconstruction quality score formula: Reconstruction_Score = a x (Region_Accuracy) + b x (Texture_Fidelity) + g x (Detail_Preservation), wherein a, b, and g are weight coefficients, respectively representing region accuracy, texture fidelity, and detail preservation. The feature enhancement score is: Enhancement_Score = (Local_Detail x Global_Coherence) / (Information_Loss + Artifact_Level), which is used to evaluate the effect of feature enhancement.

[0085] The embodiment first realizes accurate segmentation of static regions through a depth-optimized non-local attention mechanism. In processing standardized features, the feature transformation subnetwork adopts a multi-level mapping architecture, which maps the features to a high-dimensional space through nonlinear transformation. An adaptive weight mechanism is introduced in the mapping process to dynamically adjust the transformation parameters according to the importance of the features. For complex texture regions, the system realizes hierarchical expression of features by establishing a multi-scale feature pyramid. In the high-dimensional feature space, the system analyzes the local structure and global distribution of the features to construct a feature similarity map. The similarity calculation subnetwork uses an improved cosine distance metric to improve the accuracy of similarity calculation by considering the spatial position relationship and feature distribution characteristics of the pixel points. For the processing of region boundaries, the system introduces a boundary perception module to improve the accuracy of boundary positioning by analyzing the gradient changes of the features. This optimized attention mechanism significantly improves the accuracy of static region recognition.

[0086] The embodiment innovatively realizes the texture reconstruction mechanism of dense connection. The texture reconstruction network adopts multiple dense connection modules for feature extraction when processing static region feature maps. Each dense connection module contains multiple parallel convolution branches, different branches use different scale convolution kernels to capture multi-scale texture features. The design of the convolution kernel considers the directionality and periodicity of the texture, and enhances the feature extraction of the main texture direction through an adaptive weight mechanism. The feature fusion process adopts a gradual strategy, first combining features in local areas, then gradually expanding the fusion range to ensure the continuity of texture reconstruction. The design of the skip connection module adopts a multi-level connection strategy, which maintains the transmission of detailed information by establishing direct paths between features at different levels. This optimized reconstruction mechanism provides a reliable texture basis for subsequent feature enhancement.

[0087] The embodiment realizes feature detail enhancement through an optimized self-attention mechanism. When processing the fused features, the system uses a multi-head attention mechanism, with different attention heads responsible for capturing different types of texture features. The calculation of attention weights considers the local correlation and global consistency of the features, and describes the relationship between different regions by establishing a feature dependency graph. For detail enhancement, the system adaptively adjusts the enhancement strength by analyzing the frequency distribution of the features. An edge preservation constraint is introduced during the enhancement process to ensure that the enhanced features maintain the sharpness of the edges. This optimized enhancement mechanism significantly improves the quality of texture feature expression.

[0088] The embodiment establishes a complete static feature reconstruction framework. The generation of texture features uses a multi-stage optimization strategy, first implementing region segmentation through a non-local attention module, then reconstructing texture based on a dense connection network, and finally implementing detail enhancement through a self-attention mechanism. The system monitors the reconstruction effect in real time through a feature quality evaluation mechanism, supporting dynamic adjustment of parameters. This optimized reconstruction mechanism provides high-quality static feature expression for subsequent video processing tasks. The embodiment provides an innovative solution for static feature reconstruction, through optimized attention mechanisms and feature reconstruction, the system realizes the whole process optimization from standardized features to static texture features. This reliable reconstruction mechanism significantly improves the expression ability of features, especially in handling complex texture scenes, the reconstructed features exhibit stronger detail expression ability and spatial consistency.

[0089] The innovation of the embodiment mainly lies in the adaptive processing of non-local attention and the feature reconstruction of dense connection. By constructing a multi-level feature processing network and an optimized attention mechanism, the system realizes accurate segmentation of static regions and high-quality reconstruction of texture, providing reliable static feature support for video processing tasks. Especially in handling scenes with rich texture details, this attention-based feature reconstruction method significantly improves the expression ability and reconstruction quality of the system.

[0090] In an embodiment of the video stream image optimization method of the present application, the following can also be specifically included:

[0091] Step S501: constructing a multi-scale motion encoder, the multi-scale motion encoder including a pyramid pooling module and a dynamic convolution module, the pyramid pooling module performing multi-scale decomposition on normalized features, and the dynamic convolution module adaptively adjusting convolution kernel parameters based on motion amplitude to reconstruct features of dynamic regions and generate dynamic region motion features;

[0092] Step S502: inputting the static region texture features and the dynamic region motion features into a feature fusion network, the feature fusion network including a channel recalibration module and a spatial reweighting module, the channel recalibration module adaptively selecting feature channels, the spatial reweighting module adjusting feature weights based on a position sensitivity map, and the recalibrated features being concatenated and combined through cross-layer residual connection to generate spatio-temporal decoupling features.

[0093] Optionally, the embodiment innovatively designs a set of feature processing scheme based on multi-scale coding to solve the problems of inaccurate dynamic feature extraction, insufficient feature fusion, and spatio-temporal feature coupling in video processing. In the aspect of feature processing evaluation, the embodiment designs an encoding quality score formula: Encoding_Score = a x (Scale_Coverage) + β x (Motion_Adaptability) + γ x (Feature_Discriminability), where a, β, and γ are weight coefficients, respectively representing scale coverage, motion adaptability, and feature discriminability. The fusion effect score is Fusion_Score = (Channel_Selection x Spatial_Weighting) / (Information_Loss + Feature_Interference), which is used to evaluate the quality of feature fusion.

[0094] The embodiment first realizes accurate extraction of dynamic features through a depth-optimized multi-scale motion encoder. The pyramid pooling module adopts a multi-level feature decomposition strategy when processing standardized features. For motion features of different scales, the system aggregates features through an adaptive pooling window to ensure multi-scale expression of features. An attention mechanism is introduced in the pooling process to dynamically adjust the pooling weights by calculating the feature response map. For fast-moving targets, the system increases the pooling level to improve the feature field range. The feature decomposition process considers spatial correlation by establishing a feature dependency graph to describe the relationship between features of different scales. The dynamic convolution module uses a deformable convolution design, and the convolution kernel parameters are dynamically adjusted according to the motion amplitude. For complex motion patterns, the system analyzes the motion gradient field to adaptively adjust the shape and size of the convolution kernel. This optimized encoding mechanism significantly improves the quality of dynamic feature extraction.

[0095] The embodiment innovatively realizes a feature fusion mechanism. The feature fusion network adopts a strategy combining channel recalibration and spatial reweighting when processing static and dynamic features. The channel recalibration module realizes adaptive selection of feature channels by learning the dependency between feature channels. For important feature channels, the system increases the channel weight to enhance its contribution in the fusion process. The channel attention mechanism is introduced in the calibration process to dynamically adjust the channel weight by calculating the channel response map. The spatial reweighting module constructs a spatial weight map of the feature based on position sensitivity analysis. For target regions, the system enhances feature expression by increasing spatial weight. The reweighting process considers the spatial continuity of the feature, and the smoothness constraint ensures the rationality of the weight distribution. This optimized fusion mechanism provides reliable support for feature decoupling.

[0096] The embodiment realizes spatio-temporal decoupling through an optimized feature combination mechanism. The design of cross-layer residual connection adopts a multi-level feature transfer strategy. For features of different levels, the system maintains the independence and integrity of the features through residual learning. The feature selection mechanism is introduced in the connection process to dynamically adjust the connection strength by calculating the feature correlation. The cascading combination adopts a gradual strategy, maintaining independent expression of features in the initial stage and gradually realizing effective fusion of features in the subsequent stage. The combination process considers spatio-temporal consistency by establishing a feature correspondence relationship to ensure the continuity of decoupled features. This optimized combination mechanism significantly improves the effect of feature decoupling.

[0097] The embodiment establishes a complete feature processing framework. The generation of spatio-temporal decoupled features adopts a multi-stage optimization strategy. First, dynamic features are extracted through a multi-scale encoder, and then static and dynamic features are effectively combined through a feature fusion network. The system monitors the processing effect in real time by establishing a feature quality evaluation mechanism, and supports dynamic adjustment of the processing strategy. This optimized processing mechanism provides high-quality feature support for subsequent video processing tasks. The embodiment provides an innovative solution for feature processing. Through the optimized feature encoding and fusion mechanism, the system realizes the whole-process optimization from feature extraction to spatio-temporal decoupling. This reliable processing mechanism significantly improves the quality of features, especially in processing complex motion scenes, the decoupled features exhibit stronger discriminability and adaptability.

[0098] The innovation of the embodiment mainly lies in the dynamic encoding of multi-scale features and the adaptive fusion of features. By constructing a multi-level feature processing network and an optimized fusion mechanism, the system realizes accurate extraction of dynamic features and effective combination of static features, providing reliable feature support for video processing tasks. Especially in processing complex scenes, this decoupled feature processing method significantly improves the expression ability and processing accuracy of the system. For example, in processing sports video, the system can accurately separate the dynamic features of the athletes and the static features of the field, providing clearer visual effects.

[0099] In an embodiment of the video stream image optimization method of the present application, the following contents can also be specifically included:

[0100] Step S601: constructing a motion physical constraint model, the motion physical constraint model includes an acceleration analysis module and a speed prediction module, the acceleration analysis module calculates the acceleration distribution of the target object based on the motion trajectory constraint, the speed prediction module predicts the motion state at the next time according to the acceleration distribution, and generates physical constraint parameters;

[0101] Step S602: inputting the spatio-temporal decoupled features and the physical constraint parameters into a weight generation network, the weight generation network includes a feature correlation module and a weight calculation module, the feature correlation module performs temporal consistency analysis on the spatio-temporal decoupled features, the weight calculation module evaluates and scores the feature consistency based on the physical constraint parameters, converts the evaluation score into a weight coefficient through normalization processing, and generates a dynamic weight matrix.

[0102] Optionally, the embodiment is directed to the problem of inaccurate motion prediction, unreasonable weight allocation and insufficient feature fusion in video processing, and a dynamic weight generation scheme based on physical constraints is innovatively designed. In the aspect of motion prediction evaluation, the embodiment designs a prediction accuracy scoring formula: Prediction_Score = a x (Acceleration_Accuracy) + b x (Velocity_Consistency) + g x (State_Coherence), where a, b, g are weight coefficients, representing acceleration accuracy, velocity consistency and state coherence, respectively. The weight generation score is: Weight_Score = (Feature_Correlation x Physical_Constraint) / (Temporal_Error + Spatial_Drift), which is used to evaluate the rationality of weight allocation.

[0103] The embodiment first realizes accurate prediction of motion state through depth-optimized physical constraint model. In processing motion trajectory constraints, the acceleration analysis module adopts a multi-level feature extraction strategy. For the motion analysis of the target object, the system describes the acceleration variation law of the object by establishing a kinematic model. The generation of the acceleration distribution graph considers multiple physical factors, including inertia, external force and environmental constraints. In processing complex motion, the system decomposes the motion components and analyzes the acceleration characteristics in different directions. For the processing of sudden motion, the system introduces a motion state transition model to predict possible state changes by analyzing historical motion data. The velocity prediction module predicts the motion trend of the target based on the acceleration distribution graph using a dynamics model. Time sequence smoothing constraints are introduced in the prediction process to ensure the continuity of the prediction results. This optimized prediction mechanism significantly improves the accuracy of motion state estimation.

[0104] The embodiment innovatively realizes a weight generation mechanism based on physical constraints. The weight generation network adopts a dual-flow architecture design when processing input data. The feature correlation module constructs a feature dependency graph by analyzing the temporal changes of spatio-temporal decoupling features. The temporal consistency analysis adopts a sliding window strategy to compare the feature changes within consecutive time windows to evaluate the stability of the features. For modeling long-term dependencies, the system maintains the transmission of important feature information by establishing a memory mechanism. The weight calculation module adopts a multi-level evaluation strategy based on physical constraint parameters. The evaluation process considers the matching degree of features and physical constraints, and quantifies the physical rationality of features by establishing a scoring model. This optimized weight generation mechanism provides reliable weight guidance for feature fusion.

[0105] The embodiment realizes the standardization expression of the weight through the optimized normalization processing. The conversion process of the weight coefficient adopts an adaptive normalization strategy, and dynamically adjusts the normalization parameter according to the feature distribution. In order to maintain the smoothness of the weight distribution, the system introduces a local consistency constraint to avoid the dramatic change of the weight distribution. When generating the dynamic weight matrix, the system describes the weight dependence relationship between different regions by establishing a spatial correlation graph. The optimized normalization mechanism significantly improves the rationality of the weight distribution.

[0106] The embodiment establishes a complete weight generation framework. The generation of the dynamic weight matrix adopts a multi-stage optimization strategy, which first predicts the motion state through a physical constraint model, and then realizes the accurate distribution of the weight based on a weight generation network. The system realizes real-time monitoring of the generation effect by establishing a weight quality evaluation mechanism, and supports dynamic adjustment of the parameters. The optimized generation mechanism provides high-quality weight support for subsequent feature fusion. The embodiment provides an innovative solution for weight generation, which realizes the whole-process optimization from motion prediction to weight distribution through optimized physical constraints and weight calculation. The reliable generation mechanism significantly improves the quality of the weight, especially in processing complex motion scenes, the generated weight exhibits stronger physical rationality and temporal consistency.

[0107] The innovation of the embodiment mainly lies in the dynamic modeling of the physical constraint and the adaptive generation of the weight. By constructing a multi-level constraint model and an optimized weight generation network, the system realizes accurate prediction of the motion state and reasonable distribution of the weight, and provides reliable fusion support for video processing tasks. Especially in processing high dynamic scenes, this weight generation method based on physical constraints significantly improves the adaptability and accuracy of the system. For example, in processing sports videos, the system can accurately predict the acceleration change of the athletes, and generate reasonable feature fusion weights accordingly, improving the accuracy of the motion trajectory reconstruction.

[0108] In an embodiment of the video stream image optimization method of the present application, the following contents can also be specifically included:

[0109] Step S701: input the spatio-temporal decoupling feature and the dynamic weight matrix into a feature correction network, the feature correction network includes an adaptive fusion module and a physical constraint module, the adaptive fusion module performs weighted combination on the spatio-temporal decoupling feature based on the dynamic weight matrix, the physical constraint module performs regular correction on the fused feature according to the motion trajectory constraint, and generates a corrected feature map;

[0110] Step S702: constructing an optimized enhancement network, the optimized enhancement network comprising a residual learning module and a space-time transformation module, the residual learning module performing detail compensation on the corrected feature map, the space-time transformation module performing spatial alignment and temporal smoothing on the compensated feature based on deformable convolution, generating an optimized video feature through multi-level receptive field fusion and feature re-labeling, and performing deconvolution reconstruction on the optimized video feature to obtain a video output eliminating ghosting and ghosting.

[0111] Optionally, the embodiment is directed to the problems of inaccurate feature correction, insufficient detail compensation, and incomplete visual artifact elimination in video processing, and innovatively designs a set of feature optimization scheme based on physical constraints. In terms of feature optimization evaluation, the embodiment designs a correction quality score formula: Correction_Score = a x (Fusion_Accuracy) + b x (Physical_Consistency) + g x (Detail_Enhancement), wherein a, b, and g are weight coefficients, respectively representing fusion accuracy, physical consistency, and detail enhancement degree. The optimization effect score is Enhancement_Score = (Residual_Learning x Spatial_Alignment) / (Artifact_Level + Temporal_Discontinuity), which is used to evaluate the quality of feature optimization.

[0112] The embodiment first realizes accurate correction of features through a deep optimized feature correction network. The adaptive fusion module uses a dynamic weight matrix to combine features when processing space-time decoupled features. The weight allocation process considers the space-time correlation of features, and describes the correlation between different regions by establishing a feature dependency graph. For regions with intense motion, the system highlights the expression of motion features by increasing the weight value. An adaptive threshold mechanism is introduced in the fusion process, which dynamically adjusts the fusion parameters according to the importance of the features. The physical constraint module corrects the features based on motion trajectory constraints, and ensures that the correction results conform to the actual motion characteristics by analyzing the motion law of the target. The correction process uses an iterative optimization strategy, and adjusts the features based on physical laws at each iteration. For complex motion scenes, the system realizes hierarchical correction of features by establishing a multi-level constraint model. This optimized correction mechanism significantly improves the physical rationality of the features.

[0113] The embodiment innovatively realizes a feature enhancement mechanism. When processing the correction features, the optimization enhancement network adopts a strategy combining residual learning and space-time transformation. The residual learning module realizes effective compensation of detailed information by establishing a residual mapping relationship. For image edges and texture regions, the system improves the quality of detail reconstruction by enhancing the weight of residual learning. An edge preservation constraint is introduced in the compensation process to ensure that detail enhancement does not introduce new artifacts. The space-time transformation module adopts a deformable convolution design to realize accurate alignment of features by dynamically adjusting the shape and position of the convolution kernel. For moving targets, the system adaptively adjusts the transformation parameters by analyzing the motion state. This optimized enhancement mechanism provides reliable feature support for video output.

[0114] The embodiment realizes high-quality output of the video through an optimized feature reconstruction mechanism. The multi-layer receptive field fusion adopts a hierarchical feature combination strategy, and receptive fields of different scales are responsible for capturing image information at different levels. A spatial attention mechanism is introduced in the fusion process to dynamically adjust the fusion weight by calculating the feature response map. The feature re-calibration process considers the space-time consistency to ensure the continuity of the reconstruction result by establishing a feature correspondence relationship. The deconvolution reconstruction adopts a progressive strategy, focusing on restoring the main structure in the initial stage, and gradually optimizing the details in the subsequent stage. The ghosting and ghosting elimination in the reconstruction process is realized through physical constraints and temporal smoothing to ensure the visual quality of the output video. This optimized reconstruction mechanism significantly improves the quality of the video output.

[0115] The embodiment establishes a complete video optimization framework. The generation of the video output adopts a multi-stage optimization strategy, first correcting the features through the feature correction network, and then realizing feature enhancement and reconstruction through the optimization enhancement network. The system establishes a video quality evaluation mechanism to monitor the processing effect in real time and supports dynamic adjustment of the optimization strategy. This optimized processing mechanism provides reliable technical support for video quality improvement. The embodiment provides an innovative solution for video optimization, and through the optimized feature correction and enhancement mechanism, the system realizes full-process optimization from feature processing to video output. This reliable processing mechanism significantly improves the video quality, especially in processing complex motion scenes, the output video exhibits clearer details and fewer visual artifacts.

[0116] The innovation of the embodiment mainly lies in the physical constraint-based feature correction and multi-level feature enhancement. By constructing an optimized feature processing network and a perfect reconstruction mechanism, the system realizes accurate optimization and high-quality reconstruction of video features, providing reliable quality assurance for video processing tasks. Especially in processing high-speed motion scenes, this optimization method based on physical constraints significantly improves the processing capacity and output quality of the system. For example, when processing motion target tracking video, the system can effectively eliminate the motion-induced ghosting and ghosting, providing clearer target trajectory display.

[0117] In order to effectively solve the deficiencies of the prior art in blur processing, feature decoupling and image optimization, etc., an embodiment of a video stream image optimization device for implementing all or part of the video stream image optimization method is provided, which is shown in Figure 2 , and the video stream image optimization device specifically includes the following content:

[0118] A preprocessing module 10 is configured to generate a dynamic blur kernel according to the motion speed and optical flow parameters of a video image and a target object, model the optical flow and detect occlusions of the video image by using a plurality of convolution units, input the video image and the blur kernel into a physical perception module for data preprocessing, and generate standardized features with motion trajectory constraints.

[0119] A space-time decoupling module 20 is configured to input the standardized features into a space-time decoupling network, the space-time decoupling network includes a content branch and a motion branch, the content branch reconstructs static region texture features, the motion branch reconstructs dynamic region motion features, and the reconstruction results of the content branch and the motion branch are fused based on a cross-layer feature connection mechanism to generate space-time decoupled features.

[0120] An image optimization module 30 is configured to input the space-time decoupled features into a physical constraint corrector, the physical constraint corrector generates a dynamic weight matrix according to the motion trajectory constraints, adaptively fuses the space-time decoupled features, and corrects the physical laws, optimizes the corrected features through a residual connection and a space-time transformation network, and generates a video output that eliminates ghosting and ghosting.

[0121] As can be seen from the above description, the video stream image optimization device provided by the embodiment of the present application can ensure the safety of data processing by innovatively designing a dynamic blur kernel generation mechanism, space-time decoupling and physical constraints. A content-motion dual-branch network is constructed, combined with residual connection and space-time transformation, to establish a reliable feature reconstruction system. A physical constraint corrector is introduced to provide high-quality video output while protecting user privacy through adaptive fusion and rule correction. This method effectively solves the deficiencies of the prior art in blur processing, feature decoupling and image optimization, etc.

[0122] From the hardware level, in order to effectively solve the deficiencies of the prior art in blur processing, feature decoupling and image optimization, etc., an embodiment of an electronic device for implementing all or part of the video stream image optimization method is provided, and the electronic device specifically includes the following content:

[0123] The processor, the memory, the communications interface, and the bus; wherein the processor, the memory, the communications interface complete the communication among each other through the bus; the communications interface is used for realizing the information transmission between the video stream image optimization device and the core service system, the user terminal, and the related database and other related devices; the logic controller can be a desktop computer, a tablet computer, a mobile terminal, and the like, and the embodiment is not limited thereto. In the embodiment, the logic controller can be implemented by referring to the embodiments of the video stream image optimization method and the embodiments of the video stream image optimization device, the contents of which are incorporated herein, and the repeated parts are not described herein.

[0124] It can be understood that the user terminal can include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, and the like. The smart wearable device can include smart glasses, a smart watch, a smart bracelet, and the like.

[0125] In actual application, part of the video stream image optimization method can be executed on the electronic device as described above, or all operations can be completed in the client device. Specifically, the selection can be made according to the processing capability of the client device and the use scene limitation of the user, and the like. The application is not limited thereto. If all operations are completed in the client device, the client device can further include a processor.

[0126] The client device described above can have a communication module (i.e., a communication unit) and can be communicatively connected with a remote server to realize the data transmission with the server. The server can include a server of the task scheduling center side, and the server of the intermediate platform can also be included in other implementation scenarios, for example, a server of a third-party server platform communicatively connected with the server of the task scheduling center. The server can include a single computer device, a server cluster composed of multiple servers, or a distributed server structure.

[0127] Figure 3 A schematic block diagram of a system configuration of the electronic device 9600 of the embodiment of the application is shown in FIG. 9. As shown in FIG. 9, the electronic device 9600 can include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that the structure shown in FIG. 9 is exemplary; other types of structures can also be used to supplement or replace the structure to realize the telecommunication function or other functions. Figure 3 Figure 3 The structure shown in FIG. 9 is exemplary; other types of structures can also be used to supplement or replace the structure to realize the telecommunication function or other functions.

[0128] ​In an embodiment, the video stream image optimization method function can be integrated into the central processor 9100. The central processor 9100 can be configured to control as follows:

[0129] Step S101: generating a dynamic blur kernel according to a video image and a motion speed of a target object and an optical flow parameter, using a plurality of convolution units to model an optical flow and detect occlusion of the video image, inputting the video image and the blur kernel into a physical perception module for data preprocessing, and generating a standardized feature with a motion trajectory constraint;

[0130] Step S102: inputting the standardized feature into a space-time decoupling network, the space-time decoupling network including a content branch and a motion branch, the content branch reconstructing a static area texture feature, the motion branch reconstructing a dynamic area motion feature, and fusing reconstruction results of the content branch and the motion branch based on a cross-layer feature connection mechanism to generate a space-time decoupled feature;

[0131] Step S103: inputting the space-time decoupled feature into a physical constraint modifier, the physical constraint modifier generating a dynamic weight matrix according to the motion trajectory constraint, adaptively fusing the space-time decoupled feature, and correcting a physical law to generate a corrected feature, and optimizing the corrected feature through a residual connection and a space-time transformation network to generate a video output eliminating ghosting and ghosting.

[0132] As can be seen from the above description, the electronic device provided by the embodiments of the present application ensures data processing security through the innovative design of the dynamic blur kernel generation mechanism, through space-time decoupling and physical constraints. The content-movement dual-branch network is constructed, the residual connection and the space-time transformation are combined, and a reliable feature reconstruction system is established. The physical constraint modifier is introduced, and through adaptive fusion and law correction, high-quality video output is provided while protecting user privacy. The method effectively solves the deficiencies of traditional technologies in blur processing, feature decoupling, and image optimization.

[0133] In another embodiment, the video stream image optimization device can be configured separately from the central processor 9100, for example, the video stream image optimization device can be configured as a chip connected with the central processor 9100, and the video stream image optimization method function is realized through the control of the central processor.

[0134] As shown in FIG. 9B, the electronic device 9600 can further include a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily include all the components shown in FIG. 9B; in addition, the electronic device 9600 can further include components not shown in FIG. 9B. Figure 3 Figure 3 As shown in FIG. 9B, the electronic device 9600 can further include a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily include all the components shown in FIG. 9B; in addition, the electronic device 9600 can further include components not shown in FIG. 9B. Figure 3 ​Components not shown in the figure can be referred to the prior art.

[0135] As shown in FIG. 9, the central processing unit 9100, which is also sometimes referred to as a controller or operation control, can include a microprocessor or other processor device and / or logic device, which receives input and controls the operation of the various components of the electronic device 9600. Figure 3

[0136] The memory 9140, for example, can be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, or other suitable device. Information relating to failures described above can be stored, in addition to programs for executing the information. The central processing unit 9100 can execute the programs stored in the memory 9140 to perform information storage or processing, etc.

[0137] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and text. The display can be, for example, an LCD display, but is not limited thereto.

[0138] The memory 9140 can be a solid state memory such as a read only memory (ROM), a random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROM, etc. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 can include an application / function storage section 9142 for storing application programs and function programs or for storing a flow for performing the operation of the electronic device 9600 by the central processing unit 9100.

[0139] The memory 9140 can also include a data storage section 9143 for storing data such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. A driver storage section 9144 of the memory 9140 can include various drivers of the electronic device for communication functions and / or for performing other functions of the electronic device such as a messaging application, a contact application, etc.

[0140] ​The communication module 9110 is a transmitter / receiver that transmits and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, as in the case of a conventional mobile communication terminal.

[0141] Based on different communication technologies, a plurality of communication modules 9110, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, can be provided in the same electronic device. The communication module 9110 (transmitter / receiver) is also coupled to the speaker 9131 and the microphone 9132 via the audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby implementing a conventional telecommunication function. The audio processor 9130 can include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is coupled to the central processor 9100, thereby enabling recording on the local device via the microphone 9132 and enabling playing of a sound stored on the local device via the speaker 9131.

[0142] The embodiments of the present application further provide a computer readable storage medium capable of implementing all steps of the video stream image optimization method whose execution subject is a server or a client in the above-mentioned embodiments, and a computer program stored on the computer readable storage medium, wherein the computer program is executed by a processor to implement all steps of the video stream image optimization method whose execution subject is a server or a client in the above-mentioned embodiments, for example, the steps of:

[0143] Step S101: generating a dynamic blur kernel according to a video image and a motion speed and an optical flow parameter of a target object, performing optical flow modeling and occlusion detection on the video image by using a plurality of convolution units, inputting the video image and the blur kernel into a physical perception module to perform data preprocessing, and generating a standardized feature with a motion trajectory constraint;

[0144] Step S102: inputting the standardized feature into a space-time decoupling network, wherein the space-time decoupling network includes a content branch and a motion branch, the content branch reconstructs a static area texture feature, the motion branch reconstructs a dynamic area motion feature, and a reconstruction result of the content branch and the motion branch is fused based on a cross-layer feature connection mechanism to generate a space-time decoupling feature;

[0145] Step S103: inputting the space-time decoupling feature into a physical constraint corrector, generating a dynamic weight matrix according to the motion trajectory constraint, adaptively fusing the space-time decoupling feature, and correcting a physical law, optimizing the corrected feature through a residual connection and a space-time transformation network, and generating a video output eliminating trailing and ghosting.

[0146] As can be seen from the above description, the computer readable storage medium provided by the embodiments of the application ensures data processing security by innovatively designing a dynamic blur kernel generation mechanism, through space-time decoupling and physical constraints. A content-motion double-branch network is constructed, residual connection and space-time transformation are combined, and a reliable feature reconstruction system is established. A physical constraint modifier is introduced, adaptive fusion and regular correction are performed, high-quality video output is provided while protecting user privacy. The method effectively solves the deficiencies of traditional technologies in blur processing, feature decoupling and image optimization.

[0147] The embodiments of the application also provide a computer program product capable of implementing all steps of the video stream image optimization method in which the execution subject is a server or a client in the above embodiments. The computer program / instruction is executed by a processor to implement the steps of the video stream image optimization method, for example, the computer program / instruction implements the following steps:

[0148] Step S101: generating a dynamic blur kernel according to a video image and a motion speed and an optical flow parameter of a target object, performing optical flow modeling and occlusion detection on the video image by using a plurality of convolution units, inputting the video image and the blur kernel into a physical perception module for data preprocessing, and generating standardized features with motion trajectory constraints;

[0149] Step S102: inputting the standardized features into a space-time decoupling network, the space-time decoupling network including a content branch and a motion branch, the content branch reconstructing static region texture features, the motion branch reconstructing dynamic region motion features, and fusing reconstruction results of the content branch and the motion branch based on a cross-layer feature connection mechanism to generate space-time decoupled features;

[0150] Step S103: inputting the space-time decoupled features into a physical constraint modifier, the physical constraint modifier generating a dynamic weight matrix according to the motion trajectory constraints, performing adaptive fusion and physical law correction on the space-time decoupled features, and optimizing the corrected features through a residual connection and a space-time transformation network to generate video output eliminating ghosting and ghosting.

[0151] As can be seen from the above description, the computer program product provided by the embodiments of the application ensures data processing security by innovatively designing a dynamic blur kernel generation mechanism, through space-time decoupling and physical constraints. A content-motion double-branch network is constructed, residual connection and space-time transformation are combined, and a reliable feature reconstruction system is established. A physical constraint modifier is introduced, adaptive fusion and regular correction are performed, high-quality video output is provided while protecting user privacy. The method effectively solves the deficiencies of traditional technologies in blur processing, feature decoupling and image optimization.

[0152] Those skilled in the art will appreciate that embodiments of the present application can be readily used as software, hardware, or a combination of software and hardware. In a software embodiment, the methods can be tangibly embodied in a machine-readable storage medium having stored thereon instructions that can be used to program a computing device to perform the methods. The program instructions can be software supplied within hardware such as a computer or a dedicated machine. The software can be distributed on a computer program product, such as a compact disc, flash memory, or any other computer readable medium, to one or more systems or devices.

[0153] The present application is described in relation to flow diagrams and / or block diagrams of methods, apparatus (devices) and computer program products according to embodiments of the application. It is understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing unit or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more functions specified in one or more of the flow diagrams and / or block diagrams. Figure 1 one or more functions specified in one or more of the flow diagrams and / or block diagrams.

[0154] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagrams and / or block diagrams flow or flows and / or block or blocks. Figure 1 one or more functions specified in one or more of the flow diagrams and / or block diagrams. Figure 1 one or more functions specified in one or more of the flow diagrams and / or block diagrams.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagrams and / or block diagrams flow or flows and / or block or blocks. Figure 1 one or more functions specified in one or more of the flow diagrams and / or block diagrams. Figure 1 one or more functions specified in one or more of the flow diagrams and / or block diagrams.

[0156] The principles and implementation modes of the present application are described in the specific embodiments. The above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation modes and application ranges can be changed; in conclusion, the content of the specification should not be understood as the limitation of the present application.

Claims

1. A video stream image optimization method, characterized in that, The method includes: A dynamic blur kernel is generated based on the motion speed of the video image and the target object, as well as optical flow parameters. Multiple convolutional units are used to perform optical flow modeling and occlusion detection on the video image. The video image and the blur kernel are then input into the physical perception module for data preprocessing to generate standardized features with motion trajectory constraints. The standardized features are input into a spatiotemporal decoupling network, which includes a content branch and a motion branch. The content branch reconstructs the texture features of static regions, and the motion branch reconstructs the motion features of dynamic regions. The reconstruction results of the content branch and the motion branch are fused based on a cross-layer feature connection mechanism to generate spatiotemporal decoupling features. The spatiotemporal decoupling features are input into a physical constraint corrector. The physical constraint corrector generates a dynamic weight matrix based on the motion trajectory constraints, performs adaptive fusion and physical law correction on the spatiotemporal decoupling features, and optimizes the corrected features through residual connections and a spatiotemporal transformation network to generate a video output that eliminates motion blur and ghosting.

2. The video stream image optimization method according to claim 1, characterized in that, The process of generating a dynamic blur kernel based on the motion speed of the video image and the target object, as well as optical flow parameters, and then using multiple convolutional units to perform optical flow modeling and occlusion detection on the video image includes: Multi-scale feature matching is performed on continuous video frames based on the pyramid optical flow algorithm to construct the motion trajectory matrix of the target object. The motion trajectory matrix is ​​then input into a multi-layer convolutional neural network for feature extraction. The multi-layer convolutional neural network includes a spatial attention module and a channel attention module. Dynamic blur kernel parameters are calculated based on the spatiotemporal correlation of the motion trajectory matrix. An optical flow model is constructed using multiple dilated convolutional units, each containing a residual connection layer and a normalization layer. Frame-by-frame optical flow estimation is performed on the video image sequence. An occlusion detection map is generated based on the consistency analysis of the optical flow vector field and the motion trajectory matrix. The occlusion detection map is then weighted and combined with the dynamic blur kernel parameters to obtain a dynamic blur kernel that considers the occlusion relationship.

3. The video stream image optimization method according to claim 1, characterized in that, The step of inputting the video image and the blur kernel into the physical perception module for data preprocessing to generate standardized features with motion trajectory constraints includes: The video image and the dynamic blur kernel are input into the physical perception module, which includes multiple deconvolution layers and upsampling layers. The video image is deblurred based on the dynamic blur kernel. The deconvolution layers iteratively reconstruct the image features, and the upsampling layers improve the resolution of the reconstructed features to generate an initial feature map. A motion trajectory constraint model is constructed using a bidirectional long short-term memory network. The motion trajectory constraint model includes a forward propagation layer and a backward propagation layer. The initial feature map is input into the motion trajectory constraint model. The forward propagation layer extracts forward temporal features, and the backward propagation layer extracts backward temporal features. Based on the correspondence between the forward temporal features and the backward temporal features, trajectory constraint conditions are constructed. The initial feature map is batch normalized to generate standardized features with motion trajectory constraints.

4. The video stream image optimization method according to claim 1, characterized in that, The standardized features are input into a spatiotemporal decoupling network, which includes a content branch and a motion branch. The content branch reconstructs the texture features of static regions, including: Static region segmentation is performed on the standardized features to construct a non-local attention module. The non-local attention module includes a feature transformation sub-network and a similarity calculation sub-network. The feature transformation sub-network maps the standardized features to a high-dimensional feature space, and the similarity calculation sub-network calculates the similarity matrix of each pixel in the feature map based on cosine distance to generate a static region feature map. The static region feature map is input into a texture reconstruction network, which includes multiple dense connection modules and skip connection modules. The dense connection modules use multi-scale convolutional kernels to extract multi-level texture information, and the skip connection modules fuse texture information at different levels. Based on a self-attention mechanism, the fused features are enhanced in detail to generate reconstructed static region texture features.

5. The video stream image optimization method according to claim 1, characterized in that, The motion branch reconstructs the motion features of the dynamic region, and fuses the reconstruction results of the content branch and the motion branch based on a cross-layer feature connection mechanism to generate spatiotemporal decoupled features, including: A multi-scale motion encoder is constructed, which includes a pyramid pooling module and a dynamic convolution module. The pyramid pooling module performs multi-scale decomposition on the standardized features, and the dynamic convolution module adaptively adjusts the convolution kernel parameters based on the motion amplitude to reconstruct features of the dynamic region and generate dynamic region motion features. The static region texture features and the dynamic region motion features are input into a feature fusion network. The feature fusion network includes a channel recalibration module and a spatial reweighting module. The channel recalibration module adaptively selects feature channels, and the spatial reweighting module adjusts feature weights based on a position sensitivity map. The recalibrated features are cascaded and combined through cross-layer residual connections to generate spatiotemporally decoupled features.

6. The video stream image optimization method according to claim 1, characterized in that, The step of inputting the spatiotemporal decoupling features into a physical constraint corrector, wherein the physical constraint corrector generates a dynamic weight matrix based on the motion trajectory constraints, includes: A motion physical constraint model is constructed, which includes an acceleration analysis module and a velocity prediction module. The acceleration analysis module calculates the acceleration distribution map of the target object based on the motion trajectory constraints, and the velocity prediction module predicts the motion state at the next moment based on the acceleration distribution map, thereby generating physical constraint parameters. The spatiotemporal decoupling features and the physical constraint parameters are input into a weight generation network. The weight generation network includes a feature association module and a weight calculation module. The feature association module performs temporal consistency analysis on the spatiotemporal decoupling features. The weight calculation module evaluates and scores the feature consistency based on the physical constraint parameters. The evaluation scores are converted into weight coefficients through normalization processing to generate a dynamic weight matrix.

7. The video stream image optimization method according to claim 1, characterized in that, The adaptive fusion and physical law correction of the spatiotemporal decoupling features, followed by optimization of the corrected features through residual connections and a spatiotemporal transformation network, to generate video output with eliminated motion blur and ghosting, includes: The spatiotemporal decoupling features and the dynamic weight matrix are input into a feature correction network. The feature correction network includes an adaptive fusion module and a physical constraint module. The adaptive fusion module performs weighted combination of the spatiotemporal decoupling features based on the dynamic weight matrix. The physical constraint module performs regular correction on the fused features according to motion trajectory constraints to generate a corrected feature map. An optimization and enhancement network is constructed, which includes a residual learning module and a spatiotemporal transformation module. The residual learning module performs detail compensation on the correction feature map, and the spatiotemporal transformation module performs spatial alignment and temporal smoothing on the compensated features based on deformable convolution. Optimized video features are generated through multi-layer receptive field fusion and feature recalibration. The optimized video features are then deconvolved to reconstruct the video output with motion blur and ghosting eliminated.

8. A video stream image optimization device, characterized in that, The device includes: The preprocessing module is used to generate a dynamic blur kernel based on the motion speed of the video image and the target object, as well as optical flow parameters. Multiple convolutional units are used to perform optical flow modeling and occlusion detection on the video image. The video image and the blur kernel are then input into the physical perception module for data preprocessing to generate standardized features with motion trajectory constraints. The spatiotemporal decoupling module is used to input the standardized features into the spatiotemporal decoupling network. The spatiotemporal decoupling network includes a content branch and a motion branch. The content branch reconstructs the texture features of static regions, and the motion branch reconstructs the motion features of dynamic regions. The reconstruction results of the content branch and the motion branch are fused based on the cross-layer feature connection mechanism to generate spatiotemporal decoupling features. The image optimization module is used to input the spatiotemporal decoupling features into the physical constraint corrector. The physical constraint corrector generates a dynamic weight matrix based on the motion trajectory constraints, performs adaptive fusion and physical law correction on the spatiotemporal decoupling features, and optimizes the corrected features through residual connections and spatiotemporal transformation networks to generate video output with eliminated motion blur and ghosting.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the video stream image optimization method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the video stream image optimization method according to any one of claims 1 to 7.

Images