227 results about "Graph optimization" patented technology

The invention provides an unmanned aerial vehicle high-precision autonomous navigation method and system based on GNSS and IMU deep fusion, and aims to solve the problems of insufficient navigation precision and poor robustness in a complex electromagnetic environment. Through a tight coupling architecture, GNSS original observed quantity and IMU pre-integration results are jointly modeled in an observation layer, and multi-source constraints are introduced in combination with factor graph optimization, so that the positioning precision and consistency in weak signal and shielding scenes are remarkably improved. For abnormal observation, a robust kernel function is adopted to dynamically adjust the weight, and the influence of electromagnetic interference and a multipath effect is effectively inhibited. Meanwhile, navigation calculation and model prediction control MPC are combined, and sub-meter hovering and high-precision trajectory tracking are achieved. According to the method, in high-voltage transmission line inspection, dependence on a high-cost sensor is reduced, the engineering application value is high, the method can be widely applied to the fields of electric power inspection, disaster emergency, infrastructure monitoring and the like, and reliable technical support is provided for high-precision autonomous navigation of the unmanned aerial vehicle in a complex environment.

The invention discloses a large language model low-delay reasoning method based on dynamic reasoning graph optimization, and provides a low-delay reasoning method based on dynamic reasoning graph optimization. Constructing a template inference graph which can be rewritten and replayed, and establishing a template library according to input shape vectors; during reasoning, a template is matched with a distance threshold value, and only the attention / feedforward sub-graph is locally recaptured when the distance threshold value exceeds the threshold value; executing a forward execution graph, injecting a key value cache page pointer, numerical value precision and an adapter identifier, and performing playback; dividing the pre-filling and decoding sub-graphs to implement graph-level scheduling; switching the key operator when the key operator operates between a standard / fast kernel and different precisions; page-level backspacing of speculative branches is achieved through a shadow page table and reference counting, and batch and template selection is adjusted in a self-adaptive mode based on online indexes; compared with the prior art, the method has the advantages that the recapture and start overhead is reduced, tail delay and jitter are inhibited, and the hardware utilization rate and the service stability are improved.

The invention relates to the technical field of CAD drawing optimization, and discloses a CAD drawing optimization method based on parameterized design. The method comprises the steps that a design constraint condition set is received, and geometric constraint parameters and engineering constraint parameters in the design constraint condition set are extracted; generating an initial geometric topological structure according to the geometric constraint parameters, and performing parameterized mapping on the structure in combination with the engineering constraint parameters to form a parameterized geometric model; identifying key feature nodes in the model, dynamically associating the key feature nodes with engineering constraint parameters, and constructing a parameter-driven relation network; and iteratively optimizing the parameterized geometric model through the network to generate an optimized geometric model meeting a design constraint condition set. According to the method, close association of geometry and engineering constraints is achieved, repeated design modification is reduced, efficiency and accuracy are improved, the design standardization degree can be improved, collaborative design is facilitated, and the method is suitable for CAD drawing design of complex products.

The invention discloses a high-precision point cloud map construction method based on pose map optimization, and the method comprises the steps: preliminarily detecting the observation reliability of each frame of GNSS, and screening credible data; a priori factor is constructed for initial position GNSS signal quality, and map coordinate anchoring is completed; estimating a relative pose between adjacent frames, and constructing an inter-frame constraint edge; scanning context global descriptors are introduced to realize efficient screening and accurate matching of loopback candidate frames; further evaluating the quality of the GNSS by using a loopback detection result and the pose of the laser odometer, setting a GNSS factor weight, and constructing a GNSS factor edge; constructing a complete pose graph model, performing overall optimization on all pose nodes by using a nonlinear optimization method, and correcting accumulative errors; and converting the laser point cloud into a global coordinate system, and splicing to generate a global consistent high-precision point cloud map. According to the invention, the global positioning capability of the high-quality GNSS data and the local geometric precision of the laser point cloud can be effectively fused, and the high-precision point cloud map can be continuously and stably constructed.

The invention relates to the technical field of smart park integrated management, and discloses a smart park comprehensive management and control system and method based on multi-modal data analysis, and the method comprises the steps: obtaining a park multi-dimensional real-time data flow, and extracting a current situation feature vector and a personnel distribution thermodynamic diagram; analyzing the historical emergency event data sequence by using a space-time convolutional neural network, and outputting an event type identification vector; and generating a dynamic evacuation path planning and resource scheduling scheme based on a graph optimization algorithm, constructing a minimum cost flow optimization problem taking the weighted sum of time cost and waiting time as a target function, and solving to obtain an optimal people flow allocation scheme. The problem that emergency response cannot adapt to dynamic environment changes is solved. Experiential knowledge cannot be effectively inherited and utilized. And the emergency response system is slow in response and improper in resource allocation.

The invention discloses a high-precision robust robot detection method and system based on constraint pose graph optimization, and the method comprises the steps: collecting point cloud data of a detection object in multiple poses, and introducing a standard sphere and a standard ball rod in the collection process for constructing a graph constraint; a robust constraint pose map optimization algorithm framework SYMM-RCPGO is constructed; substituting the point cloud data into SYMM-RCPGO to construct an attitude graph and a constraint attitude graph to obtain a target function; iteratively solving the SYMM-RCPGO in the SE3N by combining the objective function according to the graph constraint, and using a dynamic constraint boundary algorithm to relax the tightness of the constraint to obtain the three-dimensional point cloud data of the detection object with a complete appearance; and registering the measured three-dimensional point cloud of the detection object with the designed CAD model, and accurately calculating the pose matrix of the detection object to complete the positioning of the detection object. According to the method, the SYMM distance function with a higher D-optimal measurement upper limit, the robust Welsch function and the constraint graph are utilized, so that the method is superior to a traditional PGO in the aspects of accuracy and robustness.

The invention discloses a building plane element recognition method based on a graph neural network and a convolutional neural network, and belongs to the technical field of building plane element recognition. The identification method comprises the following steps: inputting a building plane pixel image, and carrying out preprocessing through image layer screening and information extraction to obtain a binary pixel image; vectorizing the binarized pixel image to obtain a vectorized image; performing image segmentation on the vectorized image; using the segmented vectorized image to construct a regional adjacency graph; performing regional adjacency graph optimization on the regional adjacency graph through graph analysis and regional merging by using a graph neural network, and identifying a room edge; calculating the area of the room; using the convolutional neural network image recognition model to recognize articles in the room; and performing room function prediction according to an identification result. According to the method, the problem of shape blurring caused by convolution in a traditional method is avoided. The graph neural network model training effect is improved, and the classification accuracy is improved. The method is suitable for various planar graph styles and is high in universality.

The invention discloses a high-altitude track AGV accurate positioning and map building method and system. The method comprises the following steps: collecting data through multiple sensors; constructing a fusion model, and dynamically distributing fusion weights of laser and visual SLAM based on visual image information entropy and laser point cloud geometric feature distribution; semantic features such as track planes and edge lines are extracted and matched with the prior model; and jointly inputting the weight fusion pose, the UWB constraint and the semantic constraint into a back-end graph for optimization, solving an optimal pose and updating the map. The system comprises corresponding modules. Through cooperation of dynamic weight distribution and semantic auxiliary optimization, the problems of low AGV positioning precision and poor robustness in a high-altitude orbit environment are solved, dynamic closed-loop updating of a map can be realized, and the automation and intelligence level of an intelligent detection system is remarkably improved.

The invention discloses an electroencephalogram epilepsy detection method and system based on a node adaptive graph neural network, relates to a computer system based on a biological model, and provides the scheme for solving the problems of graph structure immobilization and the like in the prior art. The method comprises the following steps: an electroencephalogram signal acquisition and preprocessing step; constructing a hybrid EEG graph; optimizing a self-adaptive residual image; node specific diffusion convolution is carried out; modeling time sequence characteristics; and performing classified output. The system comprises a data acquisition module, a mixed graph construction module, a self-adaptive mapping module, a node specific convolution module, a time sequence modeling module and a classification output module. When the system runs, the steps of the method are executed, so that the electroencephalogram epilepsy detection function based on the node adaptive graph neural network is realized. The method has the technical advantages that (1) graph structure self-learning is carried out; (2) carrying out brain region personalized modeling, and strengthening region feature expression; and (3) combining space-time dependence modeling, and completely depicting the epilepsy dynamic process.

Quantum optimization with Rydberg atom arrays is provided. In particular, methods are provided for solving combinatorial graph optimization problems, constraint satisfaction problems, maximum independent set problems, algebraic problems, and factoring.

The invention relates to the technical field of retail supply chain and store resource management, and discloses a service reservation data-based store resource dynamic allocation method, which comprises the following steps of: constructing a coupling association graph of a service item and commodity combination by adopting a graph optimization algorithm fused with sparse perception constraint; according to a core service item determined by the coupling correlation graph, obtaining a reservation fluctuation mode of the core service item by using a time sequence decomposition method of an embedded state feedback mechanism; constructing and training an attention convolutional neural network model combined with local feature self-correction, and predicting service resource demand distribution of the store by using the model; generating a store resource collaborative configuration strategy based on the service resource demand distribution and the associated commodity combination real-time inventory data; according to the invention, accurate coupling prediction and collaborative dynamic allocation of the service reservation demand and the commodity inventory demand are realized, so that the efficiency and accuracy of store resource allocation are remarkably improved.

The invention relates to the field of robot autonomous navigation and environmental perception, and discloses a robot SLAM method and system fusing dual dynamic point cloud segmentation and multiple sensors, and the method comprises the steps: obtaining a three-dimensional point cloud frame, an inertial measurement unit and foot end odometer data, and obtaining a three-dimensional point cloud frame through a low-dynamic point cloud segmentation module and a high-dynamic point cloud segmentation module; the high-dynamic point cloud segmentation module identifies and filters dynamic point sets generated by low-speed and high-speed moving targets, static point cloud data are generated, the high-dynamic point cloud segmentation module detects free space state mutation, and motion state information of the dynamic targets is output. Data of the static point cloud, an inertial measurement unit and a foot end odometer are sent into a graph optimization SLAM fusion module for global pose optimization, in order to compensate segmentation delay, the graph optimization module also predicts a dynamic position and constructs an envelope frame, and the point cloud is temporarily shielded during point cloud registration. According to the method, various dynamic targets can be comprehensively filtered out, calculation delay can be compensated, and the positioning precision and mapping quality of the robot in a dynamic environment are improved.

The embodiment of the invention provides a method and equipment for assisting graph optimization and a storage medium. The method comprises the following steps: acquiring an initial mask pattern, wherein the initial mask pattern comprises a main pattern and at least one initial auxiliary pattern; obtaining a plurality of candidate auxiliary graph groups, wherein one candidate auxiliary graph group in the plurality of candidate auxiliary graph groups comprises at least one candidate auxiliary graph corresponding to the main graph; determining respective imaging costs of the plurality of candidate auxiliary graph groups; and determining at least one target auxiliary pattern for optimizing the initial mask pattern based on the respective imaging costs of the plurality of candidate auxiliary pattern groups. In this way, the imaging performance of the mask pattern under different process conditions can be improved.

The invention relates to the technical field of medical instruments, in particular to a capsule endoscopy focus automatic identification and positioning system. The system comprises an image acquisition module, an intelligent analysis module and a space positioning module. The image acquisition module receives an alimentary canal image sequence captured by the capsule endoscopy; the intelligent analysis module is integrated with a pre-trained deep convolutional neural network, analyzes the image frame by frame, and outputs a pixel-level segmentation mask identifying the contour and type of the focus; and the spatial positioning module estimates the pose of the capsule endoscopy and constructs a topological map of the inner wall of the digestive tract by analyzing the motion of feature points in the image sequence and combining a visual odometer and a graph optimization SLAM algorithm, and finally maps the recognized focus to specific coordinates of the map. According to the invention, automatic identification and accurate positioning of the nidus of the digestive tract are realized.

The invention relates to the technical field of indoor positioning, in particular to a millimeter wave CSI (Channel State Information) indoor positioning method and system based on a generative adversarial network, and millimeter waves are large in bandwidth, narrow in beam, high in time resolution and space angle resolution and capable of reducing the influence of a multipath effect. Position estimation is directly obtained from CSI data directly obtained in the millimeter wave communication process, complex signal processing and filtering steps of obtaining other parameters from CSI are avoided, and the system calculation complexity is greatly reduced. The information of the AP position is extracted from the high-dimensional observation vector by using the generative adversarial network, so that complex positioning algorithms such as Kalman filtering or graph optimization are avoided, and the calculation complexity is reduced; according to the method, AP position estimation is obtained through confrontation training of the generator and the discriminator, position coordinates of reference points do not need to serve as labels, input data are all label-free data, dependence on marked data is avoided, and prior knowledge such as the AP or the reference points is not needed.

The invention provides a pose map optimization method and device and XR equipment, and relates to the technical field of image processing, and the method comprises the following steps: calculating speedometer constraints between adjacent key frames; calculating a speedometer covariance matrix corresponding to speedometer constraints through an uncertainty propagation model; wherein the uncertainty propagation model is obtained by deducing an odometer constraint formula through an analytical method; calculating a loopback constraint between the current frame and the closed-loop candidate frame and a loopback covariance matrix corresponding to the loopback constraint; according to the speedometer constraint, the speedometer covariance matrix, the loopback constraint and the loopback covariance matrix, constructing a pose map optimization problem; and establishing a global information matrix, carrying out optimization solution on the global information matrix, and updating the pose of the key frame. The method can be operated in real time at an XR equipment end with limited computing power, low-delay and high-precision pose estimation is realized, and stronger robustness can be shown in a complex environment.

The invention belongs to the field of three-dimensional model construction, and particularly relates to a road management method of a smart city based on digital twinning. The method aims at solving the problems that traditional modeling depends on two-dimensional or limited section data, geometric accuracy is insufficient, road association is difficult to dynamically reflect, and large deviation is shown in traffic management strategy optimization. The method comprises the following steps: arranging a high-precision laser scanner on a road to obtain a point cloud, and carrying out gross error removal and curvature self-adaptive thinning pretreatment; carrying out point cloud integration registration (screening reference features, calculating descriptor similarity matching, and eliminating accumulative errors through least squares, graph optimization and a Levenberg-Marquardt algorithm), and carrying out point cloud integration registration (screening reference features, calculating descriptor similarity matching, and eliminating accumulative errors through least squares, graph optimization and a Levenberg-Marquardt algorithm); constructing an urban road digital twinborn model; and finally, simulating signal lamp timing and lane management and control based on the model, and outputting an optimal strategy through multi-objective optimization. The model fidelity is improved, congestion is effectively relieved, and the traffic efficiency is improved.

The invention discloses an event causal identification method and device based on an iterative graph neural network, equipment and a medium, and relates to the technical field of artificial intelligence and machine learning, and the method comprises the steps: carrying out the sentence coding and event extraction of an input text, and obtaining a sentence embedding and event mention result; then, sentence-level embedding and document-level embedding of the event are generated using a multi-granularity context awareness mechanism. Then, constructing an initial event causal graph structure, and encoding the initial event causal graph structure to obtain graph embedding; and finally, dynamically updating an event causal graph structure by combining sentence-level embedding, document-level embedding and graph embedding through an iterative graph optimization mechanism, and realizing accurate identification of the event causal relationship. Through a multi-granularity context perception mechanism and an iterative graph optimization mechanism, local and global context information is effectively integrated, the accuracy and robustness of document-level event causal relationship recognition are improved, and the method is particularly excellent in performance when processing long texts and cross-sentence causal relationships and can better adapt to complex document structures.

The invention discloses an unsupervised point cloud segmentation method and system based on multi-feature fusion and graph optimization. The method comprises the following steps: training a point cloud instance segmentation model by using a 3D indoor scene data set; firstly, scene point cloud in a data set is extracted to segment a foreground and a background, downsampling is carried out on the foreground, and features are extracted; then constructing an undirected graph segmentation point cloud distribution pseudo tag, and performing up-sampling to a complete point cloud; inputting a Mask3D model, and training a segmentation model in combination with a weak supervision loss function; and finally, deploying a depth camera to capture a target scene point cloud, and inputting the trained model to output an instance segmentation result. According to the method, the traditional point features and the pre-training features are considered, the performance dependence on the pre-training model is reduced, and the reliability of the method is improved. Meanwhile, aiming at the characteristics that foreground objects in an indoor scene are diverse in distribution and complex in structure, a foreground separation method is innovatively adopted, so that the model can be adjusted and optimized aiming at a foreground effect, and a better segmentation effect is achieved.

The invention discloses an SLAM (Simultaneous Localization and Mapping) optimization mapping method and device combining geometric verification and constraint of a reflector, relates to the technical field of industrial-grade mobile robot navigation, and solves the problem that in the prior art, a continuous constraint mechanism cannot be established in a mapping process to correct accumulative errors in motion. The method comprises the following steps: acquiring an original laser radar point cloud containing a reflector, screening out reflection points with the distance smaller than a preset distance, and carrying out clustering and quintuple geometric verification on the screened points by adopting a density clustering algorithm, so as to obtain a multi-dimensional geometric verification mechanism based on PCA (Principal Component Analysis); local coordinates of the center of the reflector are calculated and converted into global coordinates, the reflector is used as a stable Landmark depth to be fused into an SLAM back-end graph optimization framework, and optimization nodes including a timestamp, a robot pose, a reflector ID and an observation pose are constructed based on a global coordinate system to participate in graph optimization; and the matching problem caused by insufficient natural characteristics or environment change in a long corridor scene can be solved in a targeted manner.

The invention relates to the field of unmanned driving, and particularly provides a mining area grid map generation method and system for an unmanned mine card, a terminal and a medium, and the method comprises the steps: collecting the static environment data of a loading area through a multi-source sensor disposed on the mine card, and generating an initial point cloud map through denoising, time synchronization and SLAM fusion processing based on graph optimization; and an initial grid map is exported through automatic point cloud rasterization classification and boundary extraction. On the basis, establishing a real-time updating mechanism: continuously collecting dynamic operation data, extracting current loading boundary characteristics, and carrying out geometric comparison and area change judgment on the current loading boundary characteristics and an initial boundary; and if the change exceeds a threshold value, triggering an incremental updating process, only performing local recalculation and map coverage on a change area, and finally outputting an updated grid map. According to the invention, high-precision automatic construction and efficient real-time maintenance of the grid map of the mining area are realized, and the safety and efficiency of unmanned mine card operation are improved.

The invention discloses a commodity type selection recommendation method and system based on big data, and relates to the technical field of big data intelligent recommendation, and the method comprises the steps: carrying out the feature screening through CIIF and Shannon entropy, carrying out the optimization through an edge contribution degree and an adaptive pruning threshold value, carrying out the adjustment through an RL-QN strategy, carrying out the aggregation through HGCN, and generating a behavior enhancement rule set through NVR logic regularization. The dynamic weight of the rule is calculated through incremental reasoning, PoNSGA-III is used for optimization, a Combb-K objective function and a particle optimization algorithm are combined for global search, and a virtual center is adjusted to optimize a particle update path. According to the method, the personalized precision of commodity recommendation is improved through a collaborative linkage mechanism among feature screening, heterogeneous graph optimization and HGCN aggregation, and the diversity and stability of commodity recommendation are improved through fusion of a rule logical reasoning engine and a neural network regularization technology in combination with a parallel search mechanism of a multi-objective optimization algorithm.

The invention discloses a laser radar-vision tight coupling positioning method and system based on adaptive uncertainty modeling, and belongs to the technical field of automatic driving or robots. In order to solve the problem that the performance of an existing method is reduced in a dynamic and degraded scene, the method is realized through the following steps: quantitatively evaluating the data quality of visual and laser radars in real time; generating a covariance matrix for each observation dynamic based on the quality index, and adaptively selecting a robust kernel function; and finally, solving an optimal pose in a weighted robust factor graph optimization framework, and outputting a quantized credibility score. According to the invention, through intelligent perception and utilization of observation information, the precision, robustness and security of the positioning method in a complex environment can be significantly improved.

The invention relates to the field of artificial intelligence, and provides a hardware-aware ONNX graph optimization method and a hardware-aware graph optimization and compilation engine, which break through the barrier between hardware-independent optimization and hardware-related compilation, and improve the optimization efficiency through a unified and closed-loop optimization framework. An original ONNX model calculation graph is optimized and compiled based on hardware characteristics of an AI chip, deep collaborative optimization from the ONNX calculation graph to AI chip codes is achieved, and therefore the performance of the AI chip is played to the maximum extent.

The invention relates to the technical field of robots and automation, and discloses a dynamic environment-oriented laser SLAM (simultaneous localization and mapping) system and an efficient loopback detection method thereof, the system comprises a data preprocessing module, a probability persistence map management module, a persistence weighted front-end odometer module, a loopback detection and state reverse correction module and a rear-end pose map optimization module, static and dynamic elements in an environment are distinguished by utilizing a probability persistent map, structural conflicts in the environment are intelligently identified by comparing the similarity of a structural anchoring descriptor synthesized by high-persistent voxels with an error of geometric verification, and when the structural conflicts are identified, a system performs reverse correction on old structural information in the map; and when the geometric verification is passed, generating a loopback constraint. According to the method, transient dynamic object interference and long-term structural change of the environment can be effectively dealt with, and the robustness, loopback detection accuracy and map long-term consistency of the SLAM system in the complex dynamic environment are improved.

The invention discloses a calculation graph memory layout automatic optimization method and system based on double-layer intermediate representation, and relates to the technical field of artificial intelligence model memory layout optimization. Aiming at the defect that the memory layout optimization effect in the existing AI compiler is limited, the scheme adopted by the invention comprises the following steps of: constructing a computational graph based on a given deep learning model; the calculation graph is converted into logic intermediate representation, and graph optimization is executed; converting the optimized logic intermediate representation into a physical intermediate representation, and adding a memory layout descriptor to each tensor; candidate execution configuration is generated for the whole computational graph through a memory optimizer, an optimal scheme is decided, and physical intermediate representation is reconstructed; and analyzing the reconstructed physical intermediate representation, automatically inserting a memory release operation after calculating the final use position of the tensor in the graph, and finally compiling the physical intermediate representation to generate an executable code of the target hardware. The method is used for realizing automatic optimization of the memory layout of the computational graph.

A system and method are described which provide a unique obfuscation mechanism for conducting performance optimization of deep neural network (DNN) computational graphs. The method obfuscates performance optimization in three steps. First, an obfuscation step where the original computation graph is obfuscated such that an adversary cannot feasibly identify the original model, thus providing confidentiality. Second, the optimization step is carried out flexibly and independently by the optimizer party on the obfuscated computational graph, providing performance speedups. Finally, the de-obfuscation step where the original model is retrieved by the model owner in its optimized form.

The invention discloses an SLAM (Simultaneous Localization and Mapping) method and system based on visual inertial guidance and laser cascade registration, and belongs to the technical field of environmental perception, and the method comprises the following steps: acquiring multi-source data, predicting a pose by adopting an IMU (Inertial Measurement Unit) pre-integration method, taking the pose as an initial value, and outputting a robot visual-inertial pose by combining a camera image and utilizing a visual-inertial odometer technology; constructing a visual inertia factor; taking the vision-inertia pose as an initial value of registration, performing multi-stage registration on the input laser radar point cloud frame, outputting a laser estimation pose, and constructing a laser factor; visual loopback detection and laser loopback detection are respectively carried out, an effective closed loop is determined according to the space-time consistency of the two kinds of loopback detection, loopback factors are constructed, a factor graph is constructed, graph optimization is executed, and globally consistent tracks and map points are obtained. According to the method, the real-time performance is maintained, the registration convergence is improved, and the method is suitable for positioning and mapping in a complex degraded scene.

The invention discloses a robot orchard inter-row positioning method based on multi-modal semantic graph optimization, and aims to solve the problems of inter-row leakage and inter-row jump in positioning in orchard moving operation. Time synchronization and external parameter calibration are carried out on multi-sensor data, semantic analysis and geometric preprocessing are carried out, and the positioning accuracy of the orchard inter-row positioning method is improved. The method comprises the following steps: generating tree trunk, pile body, tree wall and ground boundary observation, constructing a semantic factor graph containing row identifiers and continuous variables, adopting a graph converter to perform adaptive weighting and covariance calibration, combining soft and hard two-stage increment optimization of anti-fact mutual exclusion gating and discrete continuous combination, outputting the row identifiers, robot trajectories, transverse deviation and course angles, and constructing a semantic factor graph containing the row identifiers and the continuous variables. The technical effects of stably inhibiting the inter-row jump and improving the row identification judgment accuracy and the transverse deviation and course angle estimation precision are achieved.

The invention discloses a monocular online reconstruction method based on three-dimensional Gaussian splashing and geometric prior. According to the method, under the condition that only a calibration-free monocular image is input, a camera and scene priori are predicted through a pre-trained visual geometric model, a direct primitive sampling strategy based on a Gaussian difference operator is combined, a three-dimensional Gaussian primitive is generated from the image and the primitive so as to expand a scene map, and the scene map is expanded. A coarse-to-fine rendering-driven joint optimization method is adopted to optimize the camera pose and the scene Gaussian map at the same time, and global consistency is ensured through online loop detection and pose map optimization. Compared with other scene reconstruction methods based on radiation field rendering and feedforward reconstruction technologies, the method has the capabilities of high-fidelity rendering and robust pose estimation in indoor and outdoor scenes.