53 results about "Visual reconstruction" patented technology

A multimodal intelligent agent system for dynamic environmental monitoring and user-centered support, consisting of: a multimodal sensor module configured to continuously acquire environmental and behavioral data from multiple input modalities, including at least one visual sensor, at least one acoustic sensor, at least one environmental conditions sensor, and at least one proximity or motion detection sensor, each generating modality-specific data streams representing visual images, audio waveforms, physical environmental parameters, and motion signatures within a monitored environment; a data preprocessing and fusion subsystem that is operationally coupled with the multimodal sensor module and configured to normalize, temporally align, and transform the modality-specific data streams into high-dimensional feature embeddings using a variety of encoders, wherein the visual encoder uses convolutional or vision transformer architectures, the audio encoder uses a spectral-temporal feature extractor, and the sensor encoder transforms raw analog data into context vectors suitable for multimodal alignment; a multimodal processing unit consisting of a transformer-based large language model (LLM) trained on paired multimodal datasets and configured to perform semantic fusion, context abstraction, and inference across the aforementioned aligned multimodal feature embeddings to generate a contextual understanding of environmental and behavioral states; an adaptive agent controller coupled to the multimodal inference processing unit and configured to instantiate, manage, and terminate a variety of task-specific intelligent agents, each agent being a software unit configured to perform a specialized function selected from meeting summarization, behavioral analysis, misplaced object detection, or environmental anomaly identification, with the agents dynamically interacting with the inference engine to retrieve contextually relevant multimodal embeddings for task execution; a personalization and adaptive learning subsystem consisting of a user preference database and a neural memory structure configured to update and refine model parameters based on user-specific interaction history, thereby enabling personalized output generation, prioritization of recommendations, and long-term behavioral adaptation; and An output generation interface is operationally connected to the adaptive agent controller and configured to produce multimodal output in textual, visual, and auditory form. The interface is capable of displaying human-readable summaries, notifications, and visual reconstructions of identified entities or environmental states.

The invention provides an oil and gas pipeline defect three-dimensional contour determination method and device. Acquiring a three-axis magnetic flux leakage detection signal of a to-be-detected target oil and gas pipeline and corresponding space coordinate information of the three-axis magnetic flux leakage detection signal; constructing multi-channel input data according to the three-axis magnetic flux leakage detection signal and the space coordinate information; determining a defect contour prediction result of the target oil and gas pipeline according to the multi-channel input data by using a pre-trained defect contour inversion model; wherein the pre-trained defect contour inversion model comprises a multi-axis feature extraction and fusion module, a feature coding module and a multi-task decoding module; the defect contour prediction result is used for indicating whether the target oil and gas pipeline has defects or not, and the defect contour prediction result is further used for indicating the three-dimensional contour shape of the defects under the condition that the target oil and gas pipeline has the defects. Therefore, high-precision visual reconstruction of the defect position and the three-dimensional form of the oil and gas pipeline is realized, and the accuracy of defect identification and the reliability of evaluation are improved.

The invention relates to the technical field of robot visual perception and motion control, and discloses a motion control method and device based on physical reference, equipment and a medium, and the method comprises the steps: obtaining instruction information, a multi-view image and movable assembly pose information; processing the multi-view image according to the instruction information to generate target segmentation information; generating a scale normalization point cloud and a model estimation baseline; determining a physical reference baseline and generating a scale calibration factor; converting the scale normalization point cloud into a physical space point cloud by using a scale calibration factor; extracting a three-dimensional relative position of the target object relative to the movable component in combination with the target segmentation information; an action instruction is generated based on the multi-modal input. According to the method, physical scale alignment of the point cloud is realized through physical reference baseline calibration, so that a visual reconstruction result has real space significance, an accurate action instruction is generated, and the robot space understanding and operation precision is improved.

The invention relates to the technical field of nondestructive testing, and discloses a bimetal composite pipe three-dimensional reconstruction method and system based on multi-source data fusion. The method comprises the following steps: acquiring magnetic flux leakage signals and thickness data of the bimetal composite pipe in a high-pressure environment, and preprocessing the magnetic flux leakage signals and the thickness data to obtain a clean multi-source signal data set; performing time domain and frequency domain feature alignment on the data set to generate a fusion data matrix; extracting a preliminary defect feature set through multi-layer convolution processing; performing classification training on the defect features to obtain a defect type classification result containing confidence scores; if the crack exists, depth fitting is carried out to quantify the crack depth; if the preset risk threshold value is exceeded, generating a three-dimensional defect distribution model and evaluating connectivity; and finally, outputting a quantitative evaluation report of the pipeline risk level. According to the method, efficient fusion of multi-source data, intelligent identification of defect types and three-dimensional visual reconstruction are realized, and the accuracy and evaluation efficiency of pipeline defect detection are remarkably improved.

The invention relates to the crossing field of mineralogy and geochemistry, in particular to a mineral and element relation analysis method for gold-bearing minerals based on image matching and data fusion and application of the mineral and element relation analysis method. Comprising the steps of ore sample processing, gold-bearing mineral image calibration through AMICS, trace element scanning through fixed-point EPMA and EDS, image-level matching and spatial data fusion and gold-bearing mineral-element relation analysis. According to the method, the AMICS mineral map and the EPMA element information are fused, relation analysis of trace elements in the gold ore and visual reconstruction of a mineral-element corresponding mechanism can be achieved, and effective data and image models are provided for microcosmic metallogenic information requirements in digital mine construction.

The invention relates to the technical field of robot automatic polishing, in particular to a vision-guided robot polishing path planning method and system, and the method comprises the steps: obtaining a workpiece surface point cloud in real time through a three-dimensional vision sensor, generating a trajectory seed point with adaptive density in combination with curvature, a normal vector and a defect semantic tag, and carrying out the positioning of a workpiece surface point; and a polishing quality evaluation function is constructed based on force sense feedback and a visual reconstruction result, so that path online correction is realized. According to the method, the polishing path and the tool posture can be dynamically adjusted, high-precision differentiation processing is achieved on the free-form surface, the edge and the defect area, and the polishing consistency and efficiency are improved.

The invention provides a rockburst intelligent early warning method based on cross-modal fusion of while-drilling sensing data and micro-seismic monitoring data, and belongs to the technical field of tunnel and underground engineering safety and disaster prevention and control. The rockburst intelligent early warning method comprises the steps that while-drilling parameters are acquired in real time, and drilling three-dimensional track coordinates are recorded; microseismic event signals induced by drilling disturbance are collected in real time, and seismic source parameters are inverted; preprocessing and time-space alignment are carried out on the collected while-drilling parameters and the microseismic event signals, and an aligned multi-modal feature sequence is generated; inputting the aligned multi-modal feature sequence into a trained rockburst risk prediction model, and outputting a rockburst risk grade probability distributed along the axis of each drilling track; and performing spatial interpolation on the rockburst risk level probability distributed along the axis of each drilling track, generating a three-dimensional rockburst risk probability field in front of the tunnel face, performing three-dimensional visual reconstruction on the three-dimensional rockburst risk probability field, performing risk level judgment according to a three-dimensional risk cloud picture in front of the tunnel face, and automatically issuing a corresponding early warning signal.

The invention discloses a visual dense reconstruction method based on pure geometric Gaussian splashing, and relates to the field of visual reconstruction, and the method comprises the steps: distributing an initial Gaussian element for each obtained initial point cloud, and forming an initial Gaussian model; generating a training set based on the acquired RGB images; determining single-view, multi-view and key point consistency loss on the basis of the training set and camera parameters in combination with a pure geometric Gaussian splash technology which only retains geometric parameters to represent a scene, and determining training loss on the basis of the loss; training the initial Gaussian model based on the training loss to obtain a pure geometric Gaussian model; and obtaining rendering depths under all views by adopting a pure geometric Gaussian model, and obtaining a visual dense reconstruction result through TSDF fusion in combination with RGB images and camera parameters corresponding to the views. According to the method, the convergence speed of the reconstruction geometry can be accelerated under the condition of reducing the computing resource and hardware performance requirements, so that the higher geometric accuracy can be achieved in a shorter time.

The invention relates to the technical field of coal mine safety production monitoring and intelligent early warning, and discloses a coal mine hidden danger recognition and intelligent early warning system based on multi-modal feature fusion, which comprises a multi-modal sensing data acquisition module, an edge calculation preprocessing module, a central analysis server and a linkage alarm execution module, the central analysis server integrates a three-dimensional flow field construction unit, a feature extraction unit, a cross-modal fusion unit and a hidden danger decision-making unit. The method comprises the following steps: constructing a three-dimensional limited flow field by using monocular depth estimation and an optical flow field, and inverting gas concentration into a virtual gas concentration distribution diagram; performing cross-modal depth correlation on visible light, infrared, audio and gas distribution characteristics by using a Transform network; and the hidden danger decision-making unit identifies categories based on the fused features and generates a visual heat map. According to the method, visual reconstruction of invisible gas and deep complementation of heterogeneous data are realized, the problem of high false alarm rate of single-mode monitoring is solved, and the accuracy and response speed of coal mine hidden danger early warning are improved.

The invention discloses a semantic-guided lightweight three-dimensional reconstruction method based on three-dimensional Gaussian, and the method comprises the steps: inputting a multi-view two-dimensional image, and extracting two-dimensional semantic features; projecting the two-dimensional semantic features to the three-dimensional Gaussian primitives; learning three-dimensional semantic distribution in a knowledge distillation mode, and predicting three-dimensional semantic embedding features; dividing a large scene into a plurality of sub-regions, and performing divide-and-conquer three-dimensional Gaussian optimization and semantic learning; performing semantic-guided lightweight modeling on the three-dimensional Gaussian primitives in each sub-region; executing sub-scene fusion, and performing quantitative compression on the three-dimensional data in the fused scene; and a rasterization renderer is used to carry out efficient real-time rendering under any visual angle, and a lightweight large-scene three-dimensional reconstruction result is output. According to the method provided by the invention, the real-time rendering performance of a complex scene can be remarkably improved, the model volume is effectively compressed, meanwhile, the model training process is accelerated, and the method is suitable for efficient three-dimensional visual reconstruction tasks.

The invention relates to the technical field of power supply areas, in particular to a dense power supply area visual reconstruction method based on laser radar guidance, and the method comprises the steps: combining laser radar equipment and binocular image collection equipment, and carrying out the synchronous data collection of a target area, so as to obtain laser radar point cloud data and binocular image data; performing external parameter calibration on the laser radar and the binocular camera, solving a rotation matrix and a translation vector, and establishing a unified world coordinate system; projecting the laser radar point cloud to a binocular image coordinate system to obtain an image sparse reference depth; carrying out depth correction and restoration on a binocular reconstruction result by using the laser radar point cloud; constructing a joint optimization model comprising a binocular constraint term and a laser radar constraint term, and optimizing the depth map; according to the method, a three-dimensional point cloud or model with high precision and density is generated, the three-dimensional point cloud or model is used for measuring and identifying the power supply area equipment, and reliable technical support is provided for accurate measurement and identification of the power supply area equipment by guiding a repair and joint optimization scheme.

Disclosed are a model training method based on visual reconstruction technology, a device, and a cluster, relating to the technical field of visual reconstruction. The method comprises: acquiring a residual between a first image and a second image generated by an initial model on the basis of first photographing parameters; and using the residual to train the initial model to obtain a trained initial model (e.g., an image generation model). The residual between the first image and the second image can faithfully reflect information of first content and first interference information in the first image, thereby increasing the possibility of the image generation model accurately identifying interference information in images, and thus ensuring the production of high-quality images.

The embodiment of the invention provides a three-dimensional Gaussian model updating method and device and computing equipment, and the three-dimensional Gaussian model updating method comprises the steps: obtaining a three-dimensional Gaussian model constructed based on a three-dimensional scene, and a reconstructed image and an original image of the three-dimensional Gaussian model at a target visual angle, the three-dimensional Gaussian model is constructed based on three-dimensional Gaussian points of the three-dimensional scene; based on the original image and the reconstructed image, determining a first loss of the current updating period, and based on the number of Gaussian points in the three-dimensional Gaussian model, determining a second loss of the current updating period, the first loss representing the fitting degree of the reconstructed image and the three-dimensional scene, and the second loss representing the fitting degree of the reconstructed image and the three-dimensional scene; the number of the Gaussian points is determined based on pruning of the three-dimensional Gaussian points of the three-dimensional Gaussian model in the previous updating period; and updating the three-dimensional Gaussian model based on the first loss and the second loss. The accuracy of visual reconstruction in the updating process of the three-dimensional Gaussian model is ensured, and performance loss and computing resource consumption caused by three-dimensional Gaussian point redundancy are avoided.

The invention relates to the technical field of wind power plant monitoring and management, and particularly discloses a virtual-real fusion real-time reconstruction method and system for a wind power plant profile. Comprising five stages: firstly, realizing task adaptive allocation through a heterogeneous edge-cloud cooperative computing node cluster; secondly, performing semantic paragraph division and value marking on the wind power plant based on terrain, wake flow and power gradient features; thirdly, calculating a semantic fracture index to judge the profile stability, and generating a reconstruction instruction when needed; then, differential visual reconstruction is executed according to the value label, and dual-path robustness verification is carried out in parallel; and finally, performing closed-loop optimization on system parameters in combination with interactive feedback and a verification result. According to the method, the whole process from data acquisition, semantic understanding, intelligent reconstruction to continuous optimization is realized, and the real-time performance, the accuracy and the self-adaptive capability of visual analysis of the wind power plant are improved.

This invention discloses a method and system for three-dimensional phenotypic analysis of potatoes based on computer vision. The method includes: first, constructing an acquisition system with multiple high-resolution RGB cameras in a triangular layout and a uniform light source, including a motion turntable; after distortion correction via checkerboard calibration, simultaneously acquiring multi-view images of potatoes; next, generating a dense depth map through sparse point cloud reconstruction combined with an improved multi-view stereo vision reconstruction mechanism including dynamic adaptive aggregation range and curvature-weighted matching cost calculation; registering and fusing to obtain three-dimensional point cloud data containing color and texture; then extracting basic parameters such as length, volume, and RGB features; and using an improved point cloud segmentation model including curvature-sensitive sampling, geometric attention modules, and a hybrid loss function to segment buds and extract parameters such as number and depth. This provides technical support for potato breeding and intelligent sorting.

This invention discloses a visual reconstruction method and system for remote sensing interpretation based on a generative diffusion model. The method includes: acquiring high-resolution and low-resolution remote sensing image data; adding different levels of Gaussian noise to the training data using a forward stochastic differential equation until pure Gaussian noise data is obtained; training a noise conditional scoring network to predict the scores corresponding to these noisy data; adding noise to the low-resolution image using a forward stochastic differential equation to finally obtain pure Gaussian noise; using a trained neural network to guide the random noise to gradually converge and generate a super-resolution remote sensing image; rapidly identifying land cover types on the generated remote sensing image; and delineating land cover patches on the original remote sensing image and assigning patch information based on the identified land cover categories. This invention achieves a super-resolution effect from low resolution without changing the land cover types and patch boundaries, thereby reducing interpretation costs and improving interpretation efficiency.

A visual recreation system of visually generating a recreation of a medical incident associated with a patient. The system includes a memory and a processor. The memory includes instructions that, when carried out by the processor, cause the processor to receive, from one or more sensors, a location information and a time information of the medical incident. The instructions further cause the processor to generate a visual simulation of the medical incident including an avatar of the patient and a visual reconstruction of a medical environment based on the location information and the time information of the medical incident.

The invention provides an architectural space information model strong generalization generation method and system integrating a feature matrix. The system comprises a data extraction module, a design task translation module, a visual reconstruction module and a question and answer interaction module. According to the method and the system, a representation normal form of building space parameters can be constructed in combination with an architect perspective, building geometric parameters are translated into a mathematical matrix, automatic and rapid coding of the parameters is realized, and the understanding ability of a large model for building space is improved; and as a data level front-end technology, the method supports the application of a large model in the aspects of building simulation model automatic construction, building design knowledge questions and answers, building effect picture rendering and the like.

This invention relates to the field of intelligent inspection and 3D visual reconstruction technology for automotive parts, specifically a method for 3D image reconstruction of lightweight internal high-pressure molded parts for automobiles. The method includes: acquiring a sequence of structured light images, multispectral reflectance texture images, and a priori simulation model of the target object, and completing spatial coordinate registration; dynamically allocating edge extraction weights based on the local curvature gradient of the 2D image to generate a non-uniform density 3D point cloud; constructing a surface reconstruction energy function using surface reflectance variation data as a penalty term, fitting the point cloud to the 3D surface, and generating a target 3D mesh model; fusing 3D geometric features, 2D texture features, and the priori simulation model, and outputting the wall thickness reduction rate and residual stress distribution via a graph neural network; generating and overlaying a risk heat map, and outputting the evaluation decision results. This invention extends from geometric reconstruction to risk semantic reconstruction, improving the comprehensiveness and reliability of online inspection of internal high-pressure molded parts.

The invention discloses a visual reconstruction implantation system based on optical communication and a control method. The implantation system comprises an external device and an implantation device, the external device is used for generating a stimulation instruction according to the visual information, embedding a near-infrared light signal, then sending the near-infrared light signal to the implantation device, and receiving a feedback light signal from the implantation device; the implantation device comprises a nerve interface module which is used for carrying out electrical stimulation on optic nerve tissue according to a stimulation instruction in the near-infrared light signal and collecting an electrophysiological signal; and the optical communication and energy supply module is used for receiving the near-infrared light signal, supplying power to the implanted device, modulating the electrophysiological signal into a feedback light signal and then transmitting the feedback light signal to an external device. Reliable issuing of a high-channel stimulation instruction and real-time feedback of a neural electrophysiological signal are realized through a near-infrared light bidirectional time division multiplexing link, and the resolution and stability of visual reconstruction are remarkably improved.

This application discloses a model training method based on visual reconstruction supervision, comprising: constructing a video anomaly detection model to be trained, the model including a global-local temporal modeling module, a feature reconstruction module, and one or more visual classification branch modules; obtaining an anomaly video training set, and performing weakly supervised training on the video anomaly detection model to be trained using the anomaly video training set to obtain a trained target detection model; wherein, the global-local temporal modeling module is used to obtain target visual features corresponding to each target input video in the anomaly video training set, and the feature reconstruction module and the visual classification branch module are used to perform visual reconstruction supervision based on each target visual feature during the weakly supervised training process. Implementing this application embodiment can fully utilize the original visual signals of the video training set during the model training process, which is beneficial to improving the weakly supervised model's ability to understand fine-grained visual features and its temporal modeling ability.

The application discloses a kind of multi-modal named entity recognition methods for multi-graph scene, belong to computer vision and multi-modal information extraction technical field, including: constructing entity cross-graph candidate visual region set and determining initial visual support weight;Based on entity type label, entity condition multi-instance supervision is carried out, and false matching is inhibited by non-entity negative sample constraint;Based on the entity type prediction result of set, visual region disturbance is carried out, and contribution degree and discriminativeness are determined according to target type prediction score change;Based on contribution degree and discriminativeness, visual reconstruction is guided by contribution, and support maintenance ability is determined according to target type prediction score retention degree;Further, visual contribution calibration score is generated to calibrate aggregation, and calibrated cross-graph entity level visual representation is directed to text representation, to obtain multi-modal named entity recognition result.The method can improve the reliability and accuracy of entity recognition in multi-graph scene.

The application discloses a kind of virtual interaction training method and device of visual reconstruction brain-computer interface, the method comprises: generating virtual reality scene, and extracting the importance representation relevant to current task in virtual reality scene;Extract the key visual information therein, and be mapped to the spatial position of the cortical electrode array by the preset visual field-cortical topological mapping model;Based on the individualized perception function constraint obtained in advance, the key visual information after mapping is converted into the space-time stimulation parameter of cortical electrode;Collect the behavior data of user under space-time stimulation parameter execution task, to calculate behavior error comprehensive score;According to behavior error comprehensive score, the difficulty parameter of task is dynamically adjusted, to complete closed-loop adaptive training.The problems of single training scene, subjective parameter adjustment, system closed-loop loss and insufficient evaluation standard are solved, and personalized, quantitative evaluation and safe and efficient of visual reconstruction brain-computer interface postoperative rehabilitation training are realized.

The invention discloses a wireless photovoltaic visual cortex prosthesis system and method for total-blindness visual reconstruction, and the system comprises an external image processing unit which is used for collecting an environment image and generating a corresponding optical stimulation pattern; the near-infrared projection unit is connected with the external image processing unit and is used for converting the optical stimulation pattern into a near-infrared light stimulation sequence; the optical transparent cranial window assembly is used for transmitting the near-infrared light stimulation sequence to the visual cortex; and the flexible photovoltaic stimulation array is used for receiving the near-infrared light stimulation sequence through the optical transparent cranial window assembly and directly converting the near-infrared light stimulation sequence into an electrical stimulation signal for stimulating the primary visual cortex V1. By combining the flexible photovoltaic stimulation array with the optical transparent cranial window assembly, a wireless photoelectric link from the outside to the primary visual cortex is constructed, and a solution for reconstructing functional vision by directly stimulating the cortex without a leather-covered cable is provided for all-blind patients.

The present application relates to the technical field of wind farm monitoring and management, and specifically discloses a wind farm profile virtual-real fusion real-time reconstruction method and system. It includes five stages: first, the task is adaptively distributed through the heterogeneous edge-cloud collaborative computing node cluster; second, the wind farm is divided into semantic paragraphs and marked with value based on the characteristics of terrain, wake and power gradient; third, the semantic fracture index is calculated to judge the profile stability, and the reconstruction instruction is generated when needed; fourth, the differentiated visual reconstruction is performed according to the value label, and the double-path robustness check is carried out in parallel; finally, the system parameters are optimized in a closed loop combined with the interactive feedback and the check results. This method realizes the whole process from data acquisition, semantic understanding, intelligent reconstruction to continuous optimization, and improves the real-time performance, accuracy and adaptive ability of the wind farm visualization analysis.

The invention discloses an image-based three-dimensional reconstruction method for electromagnetic simulation. Collecting and receiving a multi-view image structure by using a camera, and recovering to obtain sparse point cloud and camera parameters; inputting the data into a computer for processing, and performing initialization processing on each point of the sparse point cloud by using a ball initialization method to obtain respective quadric surface primitives; performing joint optimization on each quadric surface primitive through a simulated annealing algorithm; and extracting a triangular mesh for a symbol distance field formed by all the quadric surface primitives to obtain a three-dimensional reconstruction result, and inputting the three-dimensional reconstruction result into electromagnetic simulation. The method is clear in process and high in automation degree, geometric holes in a reconstruction model can be effectively reduced, surface continuity and geometric consistency are improved, the problem that an existing visual reconstruction method is difficult to meet the requirements for electromagnetic simulation geometric precision and continuity is solved, and the method is worthy of popularization and application. The usability and reliability of a reconstruction model in high-frequency electromagnetic simulation are remarkably improved, and the method has a good engineering application prospect.

The invention provides a radiotherapy skin dose visualization system and method based on multi-view two-dimensional photos, and the system comprises an image collection device which is used for shooting the two-dimensional photos of the skin of a patient from a plurality of different views and maintaining the consistency of collection distances; the three-dimensional reconstruction module is used for receiving the two-dimensional picture and reconstructing a first three-dimensional model of the skin surface of the patient based on a motion recovery structure and multiple stereoscopic vision; the data acquisition module is used for extracting a second three-dimensional model based on the CT image and three-dimensional dose distribution data from an external treatment planning system; the deformable registration module is used for performing non-rigid registration on the first three-dimensional model and the second three-dimensional model and outputting a non-rigid space deformation field required by registration and alignment; and the dose projection and visualization module is used for projecting the three-dimensional dose distribution data to the first three-dimensional model according to the deformation field and performing visual rendering output. According to the application, non-rigid deformation of the skin can be effectively compensated, and meanwhile, low cost and high-standardization dose visualization are achieved.

The invention discloses an atmospheric pollutant detection method, relates to the technical field of unmanned aerial vehicles, and aims to solve the problem that errors are caused by disturbance factors such as wind due to heterogeneity of data detected by an unmanned aerial vehicle in the flight process due to the influence of factors such as wind speed and flight time in the prior art. According to the method for correcting the wind disturbance error and interpolating, the limitation that traditional unmanned aerial vehicle environment detection equipment can only obtain discrete point location or linear path data is broken through, and wind disturbance error correction and spatial interpolation caused by time difference are carried out by using data of multiple height planes; multi-dimensional visual reconstruction of a vertical profile, a height plane and a spatial three-dimensional concentration field of a pollution field domain is completed by using a single machine, so that the problem of measured data error caused by heterogeneity of data detected by an unmanned aerial vehicle in a flight process due to influence of factors such as wind speed and flight time in the prior art is avoided.

The application relates to a method, a system, an electronic device and a storage medium for generating a scene map, wherein the method comprises the following steps: acquiring a repetitive texture image, determining a repetitive texture feature in a scene image set according to the repetitive texture image; for each image in the scene image set, determining whether the repetitive texture feature exists in the image, if yes, determining a region of the repetitive texture feature on the image; removing image content in the region to obtain a target scene image set, and performing visual reconstruction according to the target scene image set to obtain a scene map, or acquiring image features outside the region to obtain a target feature set, and performing visual reconstruction according to the target feature set to obtain a scene map. Through the application, the problem that a scene map reconstructed from a scene with repetitive texture has spatial errors is solved, and a correct three-dimensional scene map can be obtained.

This invention discloses a human action recognition method based on generative visual reconstruction and novel perspective synthesis, comprising: acquiring an original human skeleton sequence; reconstructing the original human skeleton sequence into a realistic human image sequence using a generative diffusion model; performing novel perspective synthesis on the realistic human image sequence, generating an enhanced human image sequence from a target perspective through three-dimensional spatial projection, the sequence including limb self-occlusion relationships and visual deformation features generated by perspective switching; extracting posture feedback from the enhanced human image sequence to obtain an enhanced skeleton sequence with physical constraint information from the target perspective; constructing an action recognition network including an anatomical partitioning sequencer module and a cascaded spatiotemporal Mamba encoder, constructing multi-view comparison learning sample pairs with the original human skeleton sequence and the enhanced skeleton sequence, and training the action recognition network; and using the trained action recognition network to determine the action category of the skeleton data to be recognized.