125 results about "Gauss point" patented technology

Disclosed in the present invention is a three-dimensional Gaussian splatting optimization method for an unposed input. The method comprises: for an input image, predicting a ray bundle distribution by using a ray prediction model, to obtain distribution features in the form of ray bundles; on the basis of the distribution features in the form of ray bundles, calculating a camera pose; for the ray bundle distribution, performing sampling on the basis of the volumetric density of light rays, to obtain an initial spatial distribution of a three-dimensional Gaussian point cloud focused on a visual center area; for the input image, obtaining a visible shell by means of view frustum projection and object-mask computation; on the basis of the initial spatial distribution of the three-dimensional Gaussian point cloud and the visible shell, performing three-dimensional Gaussian splatting scene training, to obtain a three-dimensional scene reconstruction model satisfying a preset loss function criterion, the loss function comprising a training regularization term for a camera pose parameter. The present invention provides important scene initialization information for three-dimensional Gaussian splatting training, and significantly improves the quality and the richness of detail of the final three-dimensional structure.

The invention relates to a three-dimensional reconstruction-oriented Gaussian point cloud lossless coding and decoding method, a product and a storage medium. The method comprises the following steps: extracting a first Gaussian point cloud corresponding to anhor data in a numpy array format at a coding side; converting the first Gaussian point cloud in the numpy array format into a first Gaussian point cloud in a ply format; meanwhile, through a Morton code-based spatial sorting method, performing Morton code sorting on geometric data and attribute data of the first Gaussian point cloud in the numpy array format, and generating a Morton sequence index table; and carrying out AVS coding on the first Gaussian point cloud in the ply format to obtain compressed data in a binary code stream form. On the decoding side, AVS decoding is carried out when the compressed data in the binary code stream form and the Morton sequence index table are received, and a second Gaussian point cloud in the ply format is obtained through reduction; performing format conversion on the second Gaussian point cloud in the ply format to obtain a second Gaussian point cloud in a numpy array format; and rearranging the attribute data and the geometric data based on the Morton sequence index table to realize one-to-one correspondence of the geometric data and the attribute data between the first Gaussian point cloud and the second Gaussian point cloud before and after coding and decoding. Compared with an existing method, the data consistency before and after coding is improved.

The invention relates to an object attitude estimation method based on scene-level semantic three-dimensional Gaussian spatter, which comprises the following steps of: firstly, constructing scene-level three-dimensional representation according to a multi-view image, and associating semantic embedding of three-dimensional Gaussian points of each target object; positioning a target mask area of a target object described by a language instruction according to the query image, and extracting each target three-dimensional Gaussian point subset through semantic matching and three-dimensional space clustering; secondly, an ICP registration method is guided through two-stage learning, and the initial 6D pose of the target object is obtained; and finally, performing cascade fine optimization by combining pose rendering and similarity comparison under disturbance to obtain a 6D pose of the target object in the target scene. According to the design scheme, target retrieval and instance extraction under open vocabularies are achieved through three-dimensional Gaussian reconstruction of semantic enhancement, and the attitude estimation precision under the conditions of shielding and low texture in a complex scene is effectively improved by combining language prompt and registration and rendering fine optimization guided by two-stage learning.

The invention discloses an object level motion blur three-dimensional scene synthesis method and an object level motion blur three-dimensional scene synthesis device based on three-dimensional Gaussian splash (3D Gaussian splash, 3DGS). According to the method, a multi-view image and corresponding camera parameters are used as input, and three-dimensional Gaussian point scene representation including spatial position, scale, rotation, opacity and appearance characteristics is constructed; generating a binary mask of a moving foreground object and a static background by using an image segmentation model; on the basis, camera exposure time is dispersed into a plurality of time steps, space pose parameters changing along with time are introduced only for three-dimensional Gaussian points corresponding to the foreground object, instantaneous three-dimensional scene representation under each time step is obtained, integration or weighted averaging is carried out on rendering results of the plurality of time steps, and the real-time three-dimensional scene representation of the foreground object is obtained. Generating a motion blur rendering result with real physical significance; meanwhile, a mask-guided foreground-background separation rendering strategy and a partition loss function are adopted, clear rendering supervision is applied to a static background area, fuzzy rendering supervision is applied to a moving foreground area, and joint optimization of three-dimensional Gaussian scene parameters is achieved; in the reasoning stage, the clear background and the fuzzy foreground are fused and output according to the mask, and a three-dimensional scene reconstruction result with the clear background and the reasonable motion fuzzy effect of the foreground is obtained. According to the method, the exposure integral process and the object movement track are explicitly modeled in the three-dimensional geometric space, and object-level controllable fuzzy three-dimensional scene generation is realized.

The invention discloses a head reconstruction method based on Gaussian sputtering, and belongs to the field of computer vision. Performing alignment and clone splitting on the initial point cloud model by using the supervision image to obtain a head accurate point cloud of the supervision image, and taking the head accurate point cloud position as a Gaussian point cloud position; projecting the Gaussian point cloud to an image coordinate system, and performing feature extraction by taking the supervised image as convolutional neural network input to obtain a convolutional feature map; sampling the head precise point cloud and the convolution feature map of the image coordinate system to obtain a feature vector, and decoding the feature vector to obtain a Gaussian point cloud parameter so as to obtain a three-dimensional head Gaussian model; a head image of any viewpoint is quickly rendered through the three-dimensional head Gaussian model, pixel-by-pixel loss calculation is performed on the rendered image and a real image, and the three-dimensional head Gaussian model is optimized. According to the method, under the condition of sparse input, the convolutional neural network and Gaussian sputtering are fitted, and the obtained high-fidelity three-dimensional head Gaussian model can be rendered quickly in real time and has generalization.

The invention discloses a digital human reconstruction method and system based on grid and Gaussian point mixed representation, and the method comprises the steps: carrying out the analysis and processing of a human body image in a video data set, and obtaining a human body grid and posture parameters; subdividing the human body grid, binding a vertex of the refined grid with a three-dimensional Gaussian point, and initializing a standard space Gaussian position; performing feature coding on the Gaussian position, and learning geometric attributes and color attributes of Gaussian points in combination with the feature codes and the attitude parameters to obtain a complete Gaussian attribute set; and according to the complete Gaussian attribute set, generating a digital human image by adopting a linear hybrid skin method in combination with optimization constraints. According to the invention, stable reconstruction and high-quality rendering of different character images under different postures can be realized.

The invention discloses a 3D Gaussian splash rendering optimization method based on a Unity engine in VR, and relates to the technical field of computer graphics. PLY point cloud files are converted into SOG high-compression formats containing multi-level LOD information, after low-transparency Gaussian points are removed, Morton coding sorting is carried out on the center coordinates of the Gaussian points, and then the 3D Gaussian splash rendering optimization method based on the Unity engine in VR is realized. The method comprises the following steps: constructing an octree spatial index containing a spatial bounding box and multiple LOD data, removing Gaussian points in a grading manner according to a camera distance, selecting an LOD hierarchy, carrying out deep barrel sorting on visible Gaussian points, executing drawing by taking a barrel as a unit, and combining near-to-far sorting, an improved mixed equation and screen space edge removal and amplification operation to obtain a target image. According to the method, through multi-dimensional optimization such as loading, space elimination and rendering, the loading efficiency of the VR equipment is improved, the rendering burden is reduced, Tile-Based hardware is adapted, and smooth operation of 3D Gaussian splashing on the VR all-in-one machine is realized.

The invention discloses an integrated three-dimensional space modeling method based on Gaussian splashing and a three-dimensional point cloud. The method comprises the following steps: step 1, converting a scene multi-view image and depth data into the three-dimensional point cloud; 2, constructing a point sequence input set; step 3, inputting the point sequence input set into a PointMamba model; 4, performing Gaussian parameter initialization, performing scale expansion on a local sparse region, and performing direction constraint on a global key region; 5, obtaining a prediction depth map through a binocular parallax matching algorithm, calculating a rendering depth map of the initial Gaussian point set, and adjusting the optimization intensity; and step 6, executing an adaptive encryption operation, and outputting a three-dimensional space model. According to the method, high-precision three-dimensional reconstruction of a complex scene is realized, and the method is suitable for scenes needing high-density point cloud modeling and semantic understanding, such as building scanning, industrial detection and virtual reality.

The invention discloses an SLAM system optimization method and device based on a layered anchor point diagram structure and a storage medium, and the method comprises the steps: collecting and preprocessing sensor data, constructing a layered anchor point architecture composed of a structure anchor point, an appearance anchor point, a global anchor point and a dynamic anchor point, gaussian point distribution is restrained through anchor points, camera poses are mapped to appearance codes, absolute coordinates are associated, dynamic interference is processed, and efficient sharing is achieved through an anchor point diagram. According to the method, the dynamic environment adaptability is enhanced, the resource efficiency is optimized, the fidelity and robustness of the map are improved, the multi-machine cooperation bandwidth pressure is reduced, and the method is suitable for automatic driving, robot navigation and other scenes.

The invention provides a substation intelligent operation and maintenance method and cockpit system based on Gaussian model and Internet of Things data fusion, and the method comprises the steps: carrying out Gaussian point cloud modeling based on substation information, and obtaining an initial three-dimensional Gaussian point cloud model, associating the equipment attribute information with the initial three-dimensional Gaussian point cloud model based on equipment monomer identification to obtain a target three-dimensional Gaussian point cloud model; based on the parking area and the vehicle parameters of the transformer substation, simulating initial operation and maintenance parameters of the operation vehicle in the transformer substation, and based on the target three-dimensional Gaussian point cloud model and the initial operation and maintenance parameters, carrying out operation risk detection to obtain a risk detection result; optimizing the initial operation and maintenance parameters based on the risk detection result to obtain optimized operation and maintenance parameters, and generating an electronic operation and maintenance fence based on the optimized operation and maintenance parameters; and loading the optimized operation and maintenance parameters and the electronic operation and maintenance fence to the target three-dimensional Gaussian point cloud model for operation and maintenance safety early warning. The safety and efficiency of the maintenance operation of the transformer substation are improved.

The invention discloses a map construction method of a three-dimensional Gaussian SLAM (Simultaneous Localization and Mapping) algorithm based on depth information fusion, which comprises the following steps of: firstly, adopting a dynamic point cloud downsampling method based on depth information, so that the defect that structural information is easy to lose due to the fact that point cloud spatial distribution and geometric characteristics caused by the depth information are not considered in traditional random downsampling can be overcome; the sampling rate is dynamically adjusted through depth information, a key geometric structure is reserved while data redundancy is reduced, good geometric priori is provided for generation of Gaussian point clouds, and the generated Gaussian point clouds are more fit with the surface of an object; moreover, according to the method, the optimization efficiency in the map reconstruction process is improved through the perspective principle of depth information adaptive point size and simulation of human eye observation, and the geometric error of luminosity rendering of the reconstructed map is smaller; in addition, the depth information is optimized by utilizing multi-frame depth fusion, so that the geometric position generated by the Gaussian point cloud is more accurate, and the subsequent optimization time is reduced.

The invention discloses an augmented reality inspection method and device based on a building information model, and a medium. The method comprises the following steps: converting the building information model into a BIM conditional Gaussian point field and a semantic signed distance field, and collecting multi-source heterogeneous sensor data; constructing a joint optimization factor graph, calculating a current camera pose estimation value and a covariance matrix, and monitoring numerical values of a projection edge consistency residual error and an SDF distance residual error in real time; according to the uncertainty degree of pose estimation, the transparency and the shielding effect of augmented reality rendering content are dynamically adjusted, and when it is monitored that the alignment error of a local area exceeds a threshold value and the pose covariance is lower than the threshold value, local non-rigid deformation adjustment based on engineering tolerance constraint is triggered; generating a signature visual residual proof and carrying out digital signature and timestamp authentication; and mapping the signature visual residual proof to a pre-constructed hash tree to calculate a root hash value, and storing the root hash value and the digital signature to the block chain.

A plant three-dimensional reconstruction method for a weak texture image and scale distortion comprises the following steps: step 1, recovering an initial Gaussian point cloud and camera parameters through an SfM method by using a multi-view RGB image; 2, differential rendering is executed in training iteration, a predicted image is compared with a real image, and a loss function containing various constraints is calculated to serve as a basis for gradient solving and parameter updating; 3, calculating a pixel weighted average gradient and an artifact suppression coefficient of each Gaussian in a visible view angle; 4, updating Gaussian parameters through back propagation, and adaptively triggering Gaussian cloning, splitting or deleting operation based on the pixel gradient, the covariance matrix and transparency; and 5, dynamically adjusting the Gaussian projection matrix according to the focal length and the resolution of the camera during rendering so as to maintain scale consistency, and performing anti-aliasing rendering in combination with pixel-level super-sampling. According to the method, the reconstruction precision and the structure reduction capability of the virtual plant in the weak texture region are remarkably improved.

The invention relates to the field of computer vision and navigation, and provides an SLAM acceleration method based on three-dimensional Gaussian rendering, and the method comprises the steps: initializing a three-dimensional Gaussian point cloud and a camera pose parameter based on input data, executing the camera tracking of a target frame in the input data, optimized camera pose parameters are output based on a staged switching strategy and an iteration mechanism, and camera tracking utilizes a floating point format with reduced precision; and executing a hierarchical screening mechanism in the optimized camera pose parameter to determine a key frame from the color image and the depth image, and performing mapping through a renderer supporting multichannel data by using the key frame and the optimized camera pose parameter to update the three-dimensional Gaussian point cloud. According to the method, through staged switching and iteration, pose optimization can be rapidly converged, a precision reduction floating point format is used, the numerical processing burden is relieved, a layered screening mechanism can be used for reducing the calculation cost lower than that of frame-by-frame fine comparison, and then the problem that the efficiency of an SLAM process is low is solved.

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 embodiment of the invention provides a three-dimensional Gaussian sputtering scene rendering method and device based on an unreal engine. The method comprises the following steps: analyzing point cloud data of a three-dimensional Gaussian sputtering scene based on a graphics processor, and obtaining structured point cloud attribute data; mapping each Gaussian point in the point cloud attribute data into a quadrilateral instance, and determining a mapping result; according to the point cloud attribute data and a preset double-tone sorting algorithm, determining a sorting result of Gaussian point depths corresponding to the quadrilateral instance; according to the point cloud attribute data, the mapping result, the sorting result and a preset physical rendering model, processing the obtained light source data in the three-dimensional Gaussian sputtering scene, and determining an illumination calculation precision standard; and according to the distance between the preset camera and the point cloud data in the three-dimensional Gaussian sputtering scene, performing scene rendering by using the point cloud attribute data, the mapping result and the illumination calculation precision standard. According to the scheme, the engine compatibility problem is solved, and the rendering performance is optimized.

This invention discloses a 3D scene segmentation and interaction method based on 3D Gaussian, comprising: Step 1, instance discovery: inputting 3D Gaussian scene data, dividing the 3D Gaussian points into structurally coherent instance-level Gaussian groups to obtain refined instances; Step 2, instance and scene semantic assignment: selecting representative viewpoint images of refined instances and inputting them into a visual-language model to generate semantic description labels for the instances; filtering instance pairs in the scene through spatial geometric relationships, and clarifying the spatial relationship description of instance pairs through a large model, constructing a static scene graph that integrates geometric proximity relationships and instance semantic relationships, forming a structured description of all instances; Step 3, natural language-driven instance localization and interaction: receiving user input commands, combining the instance semantic description, the static scene graph, and the real-time viewpoint direction relationship during the query to perform multi-dimensional matching, determining the target instance ID, and executing interactive operations.

This invention relates to a method for Gaussian point cloud scene model extraction and 3D semantic segmentation. It solves the problems of low quality, poor accuracy, and lack of detail in existing image 3D processing techniques. The method includes: S1, reading the image and obtaining initial Gaussian point cloud data; S2, inputting an RGB image and outputting it as a data annotation module; S3, calculating the distance from each pixel to the nearest background point; S4, randomly selecting a viewpoint in the differentiable rendering module, projecting the Gaussian point cloud from the world coordinate system to the image coordinate system using a 2DGS projector, and rendering the image from that viewpoint using the 2DGS renderer; S5, training the target using boundary loss; and S6, testing on a public dataset and a self-made dataset, presenting and explaining the results in two parts. The advantages of this invention are: fast training speed, high segmentation accuracy, convenient operation, and effective improvement of image rendering quality.

This application provides a method, system, medium, and device for human-centered video scene reconstruction and separation. The method includes: for a first-person perspective video sequence, initializing a 3D Gaussian set covering the background, hands, and objects based on a priori knowledge of structure recovery from motion; assigning a learnable dynamic category probability vector to each Gaussian point in the 3D Gaussian set; constructing a dedicated deformation branch; according to the learnable dynamic category probability vector of each Gaussian point, assigning each Gaussian point to the dedicated deformation branch for processing through a preset soft-hard two-stage routing mechanism, determining the Gaussian points processed by the dedicated deformation branch; rendering the Gaussian points processed by the dedicated deformation branch to determine the 4D scene reconstruction image and the decomposed reconstruction images of the background, hands, and objects. This application achieves 4D scene reconstruction of human-centered video and explicit, fine-grained separation of the background, hands, and objects.

The embodiment of the invention provides a three-dimensional Gaussian model anomaly recognition method and device and computing equipment, and the method comprises the steps: obtaining a target three-dimensional Gaussian model which comprises a plurality of three-dimensional Gaussian points and a plurality of verification regions, the verification area is divided based on a preset spatial range; performing distribution statistics on the visual parameters of the three-dimensional Gaussian points in the plurality of verification areas to obtain visual feature distribution of the plurality of verification areas; and judging whether the target three-dimensional Gaussian model is abnormal or not based on the visual feature distribution of the plurality of verification regions and a preset visual feature distribution threshold. According to the method, local accurate division of the space of the three-dimensional Gaussian model is realized, the statistical law of the local features of the three-dimensional Gaussian model is effectively captured, and automatic and refined recognition of the anomaly of the three-dimensional Gaussian model is realized, so that the accuracy of model anomaly recognition is improved, and an objective basis is provided for quality evaluation of the three-dimensional Gaussian model.

The application provides a picture rendering method and device, computer equipment and a storage medium. Based on the spatial geometric parameters and appearance parameters of a Gaussian point, the equivalent maximum radius of each Gaussian point in a screen space is determined, and accurate bounding box information is generated accordingly. Compared with the approximate method of a traditional axis-aligned bounding box, the application can more accurately represent the actual coverage range of the Gaussian point on the screen, improve the accuracy and picture detail fidelity of the rasterization process, reduce the number of irrelevant tiles divided into the bounding box, and effectively improve the rasterization efficiency.

The application discloses a virtual camera simulation data generation method and system based on a Gaussian point cloud model, relates to the technical field of three-dimensional point cloud processing and modeling, and comprises the following steps: after completing multi-source data interaction, a point cloud dataset is established according to the data interaction result; after completing space-time registration on the dataset, a fusion space dataset is constructed, and a Gaussian point cloud model is obtained through modeling; then, virtual camera parameters are configured, a frustum is constructed to identify invalid points, and point labels are established; center perception fusion is carried out by using the point labels to obtain a perspective simulation image; and after target perception feedback correction, simulation data is established. The application solves the technical problem that the prior art cannot accurately fuse multi-source information, the virtual camera simulation data generated has a large deviation from a real scene, and the data processing result is inaccurate, and achieves the technical effect that the virtual camera simulation data accurately fits the real scene and the accuracy of the data processing result is improved.

This invention relates to the fields of computer vision and 3D scene reconstruction technology, specifically to a method, apparatus, and storage medium for 3D scene relighting and self-illumination editing based on Gaussian splashing. The method includes acquiring images of non-illuminating and illuminating objects from multiple perspectives of a target scene, and constructing illuminating and non-illuminating image sets; constructing a 3D Gaussian model of the target scene, training the 3D Gaussian model based on the non-illuminating image set to obtain the geometric and material information of the 3D Gaussian model; optimizing the 3D Gaussian model based on the illuminating and non-illuminating image sets to obtain a set of Gaussian points with self-illumination attributes; rendering a self-illuminating map based on the geometric and material information and the self-illuminating Gaussian point set, and modifying the lighting effect of the self-illuminating map based on the self-illuminating Gaussian point set. This invention has the beneficial effects of freely adjustable light source color and intensity, and highly realistic light source editing.

The invention relates to the technical field of virtual scene generation, and discloses a dynamic scene generation method and device, and the method comprises the steps: carrying out the modeling of a multi-view scene image based on three-dimensional Gaussian splashing, and building an object geometric model of a target scene; performing surface Gaussian point cloud analysis on the object geometric model, and sampling to obtain a simulation driving point for simulating the target scene; constructing a driving skeleton of the target scene according to the simulation driving points, and mapping deformation of the driving skeleton to deformation of a three-dimensional Gaussian kernel through deformation physical simulation to obtain a dynamic skeleton of the target scene; and rendering the dynamic skeleton to generate a corresponding dynamic scene.

The application discloses a kind of high-efficiency Gaussian splashing reconstruction method and device for dynamic endoscope scene in the technical field of medical image processing, comprising: the endoscope video sequence obtained is preprocessed, and initial Gaussian point cloud is generated;The motion trajectory of each Gaussian point in initial Gaussian point cloud is modeled using Gaussian point cloud morphing model based on discrete cosine transform, and dynamic Gaussian scene is generated;The dynamic and static attributes of Gaussian point cloud morphing model based on discrete cosine transform are compressed and stored using residual-aware hybrid precision quantization strategy;Dynamic Gaussian scene is rendered using hardware-aware dense inference strategy, and three-dimensional reconstruction image of endoscope video is generated in real time.The application can effectively improve the expression efficiency of endoscope scene dynamic modeling, while reducing storage / deployment overhead.

The application provides a three-dimensional point cloud segmentation method, system, device and medium, the method comprising: constructing an initial three-dimensional Gaussian point cloud according to multi-view image data of a target scene; obtaining first base elements located in a boundary blur region in all three-dimensional Gaussian base elements of the initial three-dimensional Gaussian point cloud according to initial semantic masks of each frame of image in the multi-view image data, and performing an adaptive splitting operation on each first base element to obtain a plurality of second base elements; performing multi-scale semantic feature training on a target three-dimensional Gaussian point cloud with the initial semantic mask as a supervision signal to obtain a semantic feature vector of each base element in the target three-dimensional Gaussian point cloud; and assigning a corresponding semantic label to each base element in the target three-dimensional Gaussian point cloud according to the semantic feature vector to obtain a three-dimensional Gaussian point cloud after semantic segmentation. The application realizes optimization of a boundary splitting process of Gaussian splashing and a semantic feature matching process, and improves the accuracy, robustness and flexibility of three-dimensional Gaussian scene segmentation.

The invention relates to the technical field of automatic driving, and discloses a vehicle driving scene rendering method and device based on three-dimensional Gaussian sputtering, and the method comprises the steps: obtaining a driving scene image collected in a vehicle driving process and a radar point cloud which is synchronous with the driving scene image in time; and obtaining a road surface area image based on a driving scene image, and rendering the initialized Gaussian point cloud by taking the radar point cloud and the road surface area image as constraint conditions to obtain a vehicle driving scene rendering image. According to the scheme, the Gaussian structure obtained through reconstruction is more fit with the real terrain, and therefore the visual consistency and quality of the vehicle driving scene rendering image which is finally rendered are improved.

The invention discloses a dynamic scene image domain adaptation system and method based on four-dimensional Gaussian sputtering and a computer storage medium, and relates to the field of computer vision and computer graphics, and the method comprises the steps: obtaining dynamic images captured at different time points from a plurality of visual angles; generating a four-dimensional Gaussian model based on the dynamic image; decomposing the four-dimensional Gaussian model into a conditional three-dimensional Gaussian model and an edge one-dimensional time component; extracting a target domain embedding vector; based on the extracted target domain embedding vector and the embedding vector of the three-dimensional Gaussian model, performing affine transformation on each Gaussian point, and mapping the Gaussian representation to the distribution of the target domain; and the multi-view consistency is maintained by predicting the corresponding relationship between different training views. According to the dynamic scene image domain adaptation method based on four-dimensional Gaussian sputtering provided by the embodiment of the invention, the four-dimensional Gaussian is decomposed into the conditional three-dimensional Gaussian and the one-dimensional Gaussian based on time distribution by utilizing extension, so that the multi-view visual consistency can be ensured.

The invention relates to a Gaussian point information redistribution method. The method comprises the following steps: acquiring a visual Gaussian point set covering pixel points and extracting an attribute parameter set of the visual Gaussian point set; randomly rejecting part of Gaussian points as to-be-rejected points, and determining a neighborhood Gaussian point set of the to-be-rejected points; constructing an approximate model of pixel point colors about opacity, calculating gradient terms of the colors to the opacity, and forming a gradient term set; constructing an incidence matrix of the to-be-rejected points and the neighborhood points based on gradient terms, and calculating an information distribution coefficient in combination with the distribution factor; compensating the opacity and color information of the to-be-rejected points to neighborhood points according to the coefficients, and updating attribute parameters of the neighborhood points; and finally, gradually adjusting the shielding rate in training, modifying opacity associated parameters to maintain physical constraints, completing multiple rounds of information redistribution, finally realizing Gaussian point information redistribution, relieving over-fitting and improving reconstruction efficiency and quality.

The invention discloses a dynamic scene reconstruction method and device based on Gaussian point cloud. The method comprises the following steps: firstly, acquiring videos from at least six different camera positions by using at least six synchronous cameras to obtain a multi-view image sequence; according to the multi-view image sequence and a four-dimensional Gaussian model, generating a Gaussian point cloud containing a time dimension and a Gaussian identity identification code; integrating the Gaussian point clouds including the time dimension and the Gaussian identity identification code to generate an enhanced monomer PLY file, the enhanced monomer PLY file including the Gaussian identity identification code, and a parameter sequence and an attribute sequence corresponding to the time sequence key frame in a specified time period; and compressing the parameter sequence and the attribute sequence of the enhanced monomer PLY file to obtain a compressed enhanced monomer PLY file. And then, analyzing and dynamically rendering the compressed enhanced monomer PLY file by using a special GPU renderer to obtain a multi-view dynamic virtual scene. Therefore, high-fidelity dynamic scene reconstruction can be realized.