219 results about "Factor graph" patented technology

According to the VLA-based body robot SLAM method and device and the storage medium, a VLA large model is introduced on the basis of multi-modal fusion SLAM of traditional point cloud geometry, vision and the like, perception is improved from a geometric layer to semantic concept alignment, and the SLAM is more accurate. A VLA large model is used for carrying out dynamic prediction updating on dynamic interference filtering, key frame screening, factor graph relation construction, noise estimation, loopback detection and the like, the real-time requirement is met in the modes of incremental optimization and the like, and the dislocation problem of geometric constraints is corrected through global semantic constraints. The stability of the robot SLAM in extreme scenes such as excessive environmental dynamic interference, loud sensor noise and environmental degradation is improved, and the constructed hierarchical situation map can meet the requirement of a high-order navigation task while the geometric accuracy is met, so that the robot SLAM can be deployed to carriers such as a body-equipped intelligent carrier for subsequent application.

The invention discloses a PPP / INS / vision / LIDAR tight coupling navigation method based on multi-system real-time precision service, and belongs to the technical field of navigation and positioning. In order to solve the problem of insufficient real-time positioning precision and reliability in a complex urban environment, the invention provides a hierarchical fusion architecture: firstly, by using original data of an inertial navigation system, a visual sensor and a laser radar, high-precision and high-frequency local pose estimation is generated through tight coupling factor graph optimization; the accumulative error of inertial navigation is effectively inhibited; then, taking the local attitude as observation information, and performing tight coupling factor graph optimization on the observation information, a precise orbit obtained by PPP enhanced service decoding and a GNSS original observation value after clock correction, namely a pseudo range and a carrier phase; multi-source information is deeply fused on the observation value level, the complementary advantages of the sensors are fully played, and finally continuous and reliable centimeter-level high-precision positioning in the urban complex environment is achieved.

The invention discloses an infant airway management system and method based on navigation and high-precision dynamic registration, and the method comprises the steps: constructing a personalized airway deformation model through a physical constraint-based few-sample transfer learning network, generating an optimal instrument matching scheme, and providing a semantic weighted tight coupling visual inertial odometer; the confidence coefficient of feature points is dynamically adjusted by segmenting airway anatomy semantics in real time, data are fused under a factor graph optimization framework, and sub-millimeter robust positioning is achieved. Based on the high-precision trajectory, the system introduces a space-time attention fusion network, captures a long-range time sequence dependency relationship between operation micro-actions and physiological parameters, calculates a real-time dynamic risk score, and compensates airway soft tissue deformation in real time through a non-rigid deformation field model. The technical problems that an existing navigation system is prone to being lost in a microcosmic dynamic environment and risk early warning lags behind are effectively solved, and the safety and the success rate of the infant airway management operation are remarkably improved.

The invention discloses a quadruped robot space precise positioning method and system based on deep learning. The quadruped robot space precise positioning method specifically comprises the steps that S1, a laser radar data stream and an inertial measurement unit data stream are aligned to generate a synchronous sensing data sequence in a unified mode; s2, determining a mapping pose sequence through consistency evaluation of poses output by the offline point cloud registration algorithm pool; s3, performing parallel loopback detection to generate a unified loopback constraint set; s4, the mapping pose sequence, the loopback constraint set and the pre-integration constraint are written into the factor graph model to generate a three-dimensional point cloud reference map; s5, the local point cloud frame and the three-dimensional point cloud reference map are registered, and a map matching pose is output; s6, the map matching pose and the pre-integration result are written into a factor graph model to continuously solve and output a continuous-time six-degree-of-freedom pose result; and S7, outputting a six-degree-of-freedom pose result in continuous time. According to the invention, direction constraint consistency point cloud registration and dual-channel loopback optimization are introduced, and stable and high-precision positioning of the closed space is realized.

The invention belongs to the technical field of distributed multi-sensor passive detection multi-target tracking. The invention provides a factor graph optimization passive detection multi-target tracking method based on a Gaussian mixture model. According to the embodiment of the invention, the multi-target batch number is distributed by constructing the distributed passive sensor cooperative coordinate system and combining the multi-target identity judgment result between the two sensors; calculating direction finding lines based on two-dimensional observation of a sensor, combining the direction finding lines of the same batch number, obtaining a multi-target position estimation point set through a least square method, and obtaining a multi-target coarse positioning point through weighted fusion; modeling by adopting a Gaussian mixture model, fusing measurement distribution characteristics, solving parameters through an expectation maximization algorithm, and completing solvable conversion of an optimization problem; a factor graph optimization model containing multiple factors is constructed, state estimation is achieved through sliding window optimization, and track association and state updating are completed in combination with the Mahalanobis distance and the Hungary algorithm; and the passive detection multi-target tracking performance of the distributed sensor is effectively improved.

The invention is suitable for the technical field of data processing, and particularly relates to a sensor data time synchronization error compensation method and device, and the method comprises the steps: obtaining the multi-sensor original data of a current frame, and extracting the multi-sensor feature data; obtaining an error estimation vector based on the error estimation network; performing error compensation on the multi-sensor original data to obtain multi-sensor compensation data; constructing a factor graph, and performing optimization solution on the factor graph to obtain a residual set; calculating a comprehensive quality score according to the residual set; according to the comprehensive quality score and the comprehensive quality scores of the first M frames, determining that a triggering condition is met, and triggering fine adjustment; and based on the multi-sensor original data, the multi-sensor feature data and the comprehensive quality score in the time window, performing fine adjustment on the model parameters of the error estimation network to obtain the error estimation network after fine adjustment. According to the method, more accurate compensation of time sequence errors and data distortion among multiple sensors is realized, and continuous self-adaptive optimization is realized.

The invention relates to the technical field of unmanned aerial vehicles, in particular to an unmanned aerial vehicle positioning and orienting method, which comprises the steps of data acquisition and space-time alignment, environment feature fingerprint extraction, adaptive robust factor graph construction and dynamic covariance reweighted optimization. According to the method, the environment feature fingerprints are introduced as a prior diagnosis mechanism, GNSS non-line-of-sight propagation and visual feature depletion areas can be identified in real time, dynamic reweighting is carried out by utilizing the environment feature fingerprints on the factor graph optimization level, GNSS observation values polluted by a multipath effect or visual features with poor quality are automatically inhibited, and the accuracy and the reliability of the GNSS non-line-of-sight propagation and visual feature depletion areas are improved. Therefore, smooth seamless switching from loose coupling to tight coupling is achieved in urban canyons or indoor and outdoor transition areas, posture sudden change caused by traditional fixed threshold switching is avoided, and the positioning reliability of the system in complex electromagnetic and light environments is remarkably improved.

System, methods, and other embodiments described herein relate to implementing surface bias estimation strategies. In one embodiment, a method includes processing probe trace data with a factor graph having nodes and factors that describe an estimate of surface bias; and correcting the probe trace data based on the estimate of surface bias.

The invention relates to the technical field of robot positioning and map construction, and particularly discloses a self-adaptive LiDAR-IMU SLAM method fusing intensity features and Riemannian manifold ground constraints, and the LiDAR-IMU SLAM method comprises the following steps: a, original measurement; b, data preprocessing; c, feature extraction; d, carrying out ground manifold constraint; e, optimizing the factor graph; and f, outputting data. The method can solve the problems that in the prior art, matching fails in a low-texture environment due to dependence on geometric features, positioning drifting is caused by point cloud distortion and inertial accumulative errors, and Z-axis errors are continuously expanded due to lack of dynamic adaptation to complex terrains. Experimental results show that according to the self-adaptive LiDAR-IMU SLAM method, the Z-axis drift error is remarkably reduced by 67.78%, the minimum absolute pose error is 0.254 m, the environment sensing and mapping capacity of the mobile robot in the complex underground space is remarkably improved, and reliable technical support is provided for intelligent inspection and infrastructure monitoring of the underground space.

The invention belongs to the technical field of automated guided vehicles, and particularly relates to a multi-source positioning fusion and navigation correction method and system for an AGV, and the method comprises the steps: carrying out the pose estimation through the bidirectional coupling of extended Kalman filtering and particle filtering, and the estimation uncertainty is fed back to the extended Kalman filter to dynamically adjust the noise parameter. And further introducing sliding window factor graph optimization to carry out back-end smoothing. In the navigation stage, a dynamic decision maker based on deep reinforcement learning is adopted, and optimal control parameters are generated according to real-time positioning, map semantics and historical performance. The system can also enable the filtering model and the decision strategy to adapt to the current environment through periodic online fine tuning. Through innovatively and deeply fusing a classical state estimation method and a leading-edge machine learning method, the positioning precision, the environment understanding capability and the navigation intelligence of the AGV in a complex dynamic scene are remarkably improved.

The invention discloses an enhanced 3D (three-dimensional) surface reconstruction method based on 3D (three-dimensional) Gaussian Splitting. Global consistent depth priori is obtained through virtual stereo pair rendering; constructing a factor graph and introducing a cross-view geometry / luminosity consistency constraint to form local beam adjustment loss; the prior is used as a learnable parameter to be combined with 3DGS to be optimized, and meanwhile, the Pull loss is assisted to pull low-credibility pixels; and finally, multi-loss function end-to-end training is adopted. The method comprises the following steps of: in Tanksamp; the method has the advantages that F1 is equal to 0.58 and DTU Chamfer is equal to 0.48 mm on a Temples data set, the training time is only 20 min, compared with the prior art, geometric accuracy SOTA and speed magnitude improvement are achieved at the same time, and the method is suitable for VR / AR, robot and industrial measurement scenes.

The invention provides an offshore area navigation safety control and optimization method, which comprises the following steps of: 1, acquiring multi-source observation data of an offshore navigation target, and preprocessing the multi-source observation data to obtain a space-time observation association diagram; 2, evaluating the dynamic credibility of each observation node by using a multi-modal consistency graph neural network; constructing an adaptive weighted factor graph model according to the obtained dynamic credibility, and solving a system state to obtain an optimal fusion track, a corrected environmental parameter and a posterior error covariance matrix; 3, constructing a regional risk field based on the optimal fusion track, the corrected environmental parameters and a posterior error covariance matrix; 4, on the basis of the regional risk field, a multi-target collaborative optimization model is constructed and solved, and navigation control parameters are obtained; and 5, executing the navigation control parameters, and performing adjustment and optimization by adopting multi-level closed-loop feedback to complete the navigation safety control and optimization of the offshore area.

The invention discloses a multi-modal body navigation method and system and a storage medium. The method comprises the following steps: acquiring multi-modal sensor data and generating multi-modal sensing features; constructing a factor graph, and configuring a cross-modal consistency factor which takes the modal credibility and the local alignment parameter as a to-be-optimized state variable in the factor graph; joint nonlinear optimization is carried out on the robot pose in the factor graph and the state variables based on the multi-modal perception features, and the optimal pose and pose covariance are obtained; and a target reachable probability grid is updated based on the optimal pose and the pose covariance, the grid stores the success probability of the grid nodes reaching the target, and a navigation trajectory with the maximum success rate is planned accordingly. According to the method, the modal weight is varied, so that the problem that the fixed weight cannot adapt to sensor failure is solved; positioning uncertainty is coupled to navigation decision, the problem that positioning risks are ignored in traditional planning is solved, and robustness and safety of the robot in a dynamic complex environment are improved.

The invention relates to the technical field of human body kinematics parameter measurement, in particular to an adult joint motion range evaluation method and system based on a multi-view video, and the method comprises the steps: obtaining a multi-view video sequence of a testee, and generating an observation data set containing two-dimensional key point observation, a human body segmentation mask and camera parameter information; constructing an individualized joint geometric model containing a bone segment length, a joint center, a joint principal axis and a joint angle coordinate system based on the observation data set, and generating a three-dimensional symbol distance field; constructing a factor graph and performing incremental optimization solution, and outputting a joint angle sequence and an abnormal frame set; and determining an activity range interval and carrying out anti-fact consistency check, and if the validity is not satisfied, calculating an information matrix by a Jacobian matrix to generate a supplementary collection instruction to update a result, thereby improving robustness and verifiability.

The invention discloses a 5G and Beidou fused vehicle positioning navigation system. The system comprises a multi-source heterogeneous positioning module; a high-precision map and a dynamic traffic information base; a self-adaptive multi-layer map matching module; the dynamic path planning module considers steering limitation and real-time position feedback; and a central information fusion and decision unit. According to the invention, multi-source heterogeneous positioning data such as Beidou / GPS, 5G positioning, high-precision IMU, a vehicle-mounted atomic clock and a barometric altimeter are fused, and a tight combination algorithm based on ESKF or a factor graph is adopted, so that the system can realize accurate positioning in an environment with good satellite signals, partial shielding and even serious loss. Continuous, smooth and high-precision vehicle state estimation can be output, particularly when the number of visible satellites is insufficient, double-satellite positioning can be achieved by using the vehicle-mounted atomic clock to predict the clock error and combining the elevation constraint of the barometric altimeter, and the usability and robustness of positioning are improved.

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 application relates to a SINS large misalignment angle fast alignment method based on a Lie group and a factor graph, and belongs to the technical field of large misalignment angle alignment, and comprises the following steps: S1, defining system states under a Lie group framework; S2, constructing a factor graph model of the SINS / GNSS large misalignment angle fast alignment; S3, constructing an SINS factor cost function under the Lie group framework; S4, constructing a GNSS factor cost function under the Lie group framework; S5, optimizing the system states under the Lie group framework by using the factor graph model, the SINS factor cost function and the GNSS factor cost function, and obtaining optimal estimation of the system states; and S6, obtaining a theoretical value of a Lie group matrix containing elements related to an attitude, a speed and a position according to the optimal estimation of the system states, that is, completing the SINS large misalignment angle fast alignment method. The application has the advantages of strong robustness, good precision, simple operation, fast convergence, low cost, high alignment precision and good practicability.

A power-imbalanced multi-level decoding method for sparse code multiple access includes: encoded bits of all users are mapped to multi-dimensional sparse codewords through predetermined codebooks; a factor graph matrix is constructed using the predetermined codebooks, all users are classified according to the factor graph matrix under predetermined constraints to determine Z levels; based on a predetermined total transmission power, a progressive multi-level power optimization algorithm is employed to perform power-imbalanced allocation for all users according to the Z levels, thereby determining a locally optimal power vector; transmission signals corresponding to the multi-dimensional sparse codewords are transmitted according to the locally optimal power vector; SCMA detection and power-oriented decoding are sequentially performed for users each level, and outputs decoded bit sequences for the L levels.

The application discloses a marine buoy multi-source fusion positioning method and system based on a factor graph optimization, which acquires GNSS observation data, IMU measurement data and marine environment auxiliary data; a factor graph containing state nodes, GNSS position factors, IMU pre-integration factors and marine dynamics constraint factors is constructed, and wave and current theories are used to constrain buoy movement; in view of multipath effects, marine surface reflection geometry and marine root mean square wave height are combined to calculate a multipath weighting factor, and a GNSS covariance matrix is adaptively adjusted; according to IMU data, a sea state level is discriminated, and an edge window length and a trigger interval of incremental smoothing solving are adaptively linked and adjusted; through adjacent buoy ranging information, collaborative constraints are constructed, and based on Mahalanobis distance and chi-square distribution threshold value detection, abnormalities are detected and local reconstruction is performed. The application effectively suppresses marine surface multipath interference, slows down the accumulation of calculation errors during signal interruption, and realizes high-availability continuous positioning under limited computing power.

The invention particularly relates to a method for constructing a Sigma factor atlas structure driven by emotion clustering. According to the emotion clustering driven Sigma factor atlas structure construction method, semantic aggregation is performed on unstructured data in a knowledge base through emotion recognition and topic clustering, a dominant intention and a subtask chain are extracted, graph structure representation is formed, and finally a factor atlas capable of being used for decision reasoning grows. According to the emotion clustering-driven Sigma factor graph structure construction method, emotion information and memory management are innovatively combined, a brand new memory utilization normal form is provided for an artificial intelligence system, the structure organizability and context association degree of a knowledge graph are improved, the automation level is high, and the method is suitable for popularization and application. The method is suitable for multi-scene intelligent analysis tasks such as multi-mode RAG retrieval and fault root cause analysis.

The application provides a low-density parity-check code decoding method and device based on a graph neural network, which comprises the following steps: step 1, constructing a factor graph comprising variable nodes, check nodes and edge relationships, and mapping log-likelihood ratio information of a received signal to initial embedding of the variable nodes; step 2, generating message features according to node embedding, node degree and iteration step length, and constructing attention weights for each edge to measure the importance of the message; step 3, inputting the message features and the attention weights into a gated recurrent unit to update the residual of the edge weight, and feeding back the updated edge weight to the graph structure; step 4, weighting and aggregating the messages from the adjacent nodes according to the edge weight, calculating the updated embedding of the variable nodes and the check nodes, and performing gated modulation combined with the check result; step 5, repeating steps 2 to 4 until a preset iteration number or a decoding convergence condition is reached, and mapping the final variable node embedding to a decoding result to realize LDPC code word recovery. By introducing the attention mechanism and the gated residual update into the message passing process, the application realizes adaptive modeling of the contribution degree of different edge messages, dynamically remembers the historical state, thereby improving the decoding performance and the convergence speed; meanwhile, the application can effectively reduce the bit error rate and is suitable for high-speed reliable data transmission scenarios in a 5G / 6G wireless communication system.

The invention relates to a mobile robot motion estimation method based on pose decoupling. The method comprises the following steps: multi-sensor synchronization and decoupling state modeling; performing attitude-position decoupling pre-integration based on an IMU (Inertial Measurement Unit); visual observation modeling and reprojection constraint construction are carried out; carrying out pose optimization solution based on the decoupling factor graph; pose recovery, motion track output and local re-initialization are carried out; compared with the prior art, the method has the advantages that independent modeling and decoupling estimation are carried out on the pose rotation component and the translation component, the propagation effect of rotation errors to translation solution in traditional visual motion estimation is effectively inhibited, and the stability of translation estimation is remarkably improved. Under visual degradation scenes of weak texture, low parallax, illumination variation, motion blur and the like, estimation drift can be reduced, track continuity and consistency can be improved, and robustness of the system to abnormal matching and noise disturbance can be enhanced; besides, the decoupling mechanism provided by the invention has good universality, can be integrated with various types of visual SLAM or visual-inertial fusion frames, and provides more reliable motion estimation capability for stable autonomous navigation of the mobile robot in a complex environment.

The invention relates to a 2D semantic map-assisted multi-sensor fusion indoor positioning method, which is used for solving the problems of positioning drift and loopback confusion of a mobile robot in an environment with dynamic environment change and large-scale geometric feature degradation. The method comprises the following steps: collecting multi-source data and loading a semantic map; carrying out data distortion removal and real-time geometric feature extraction; dynamically maintaining and updating a global geometric feature map; constructing a laser scanning matching model based on landmark constraints, constructing inter-frame relative motion constraints by using an IMU pre-integration technology, and removing error loops by combining geometric and semantic consistency verification; and a tight coupling factor graph model containing multiple constraints is constructed, solving is carried out through a sparse nonlinear iterative optimization algorithm, and globally consistent robot optimal pose estimation is output. According to the method, the positioning reliability, precision and calculation efficiency of the indoor mobile robot under environment shielding and dynamic change are remarkably improved, the positioning precision can be smaller than or equal to + / -15 mm, and the CPU occupancy rate is reduced by about 61.2% compared with an original positioning algorithm.

The invention discloses a laser three-dimensional point cloud registration and correction method, system, equipment and medium based on a low-weight inertial sensing and diffusion probability model, and belongs to the technical field of point cloud registration and correction, and the method comprises the steps: obtaining original point cloud data containing electric tower structure information, carrying out the point cloud preprocessing, and obtaining a component point cloud; acquiring inertial measurement unit data which is in time synchronization with the original point cloud data, and generating an initial transformation matrix by adopting sliding window pre-integration and combining with Kalman filtering; inputting the initial transformation matrix into a diffusion probability model, and outputting a high-precision transformation matrix; constructing a factor graph model based on the high-precision transformation matrix to obtain a global consistent registration pose; and fusing the component point clouds based on the registration poses to generate a three-dimensional point cloud model with global consistency. According to the method, the accuracy of point cloud registration can be effectively improved, the attitude estimation error is reduced, stable point cloud registration can still be realized under the condition of structure shielding or view angle change, and the overall mapping success rate is improved.

Techniques associated with generating and maintaining sparse geographic and map data. In some cases, the system may maintain a factor graph comprising a plurality of nodes. In some cases, the nodes may comprise pose data and sensor data associated with an autonomous vehicle at the geographic position represented by the node. The nodes may be linked based on shared trajectories and shared sensor data.

The invention relates to the technical field of robot autonomous navigation and environmental perception, in particular to a multi-sensor fusion SLAM and autonomous exploration method in a complex agricultural environment. The method comprises the following steps: correcting motion distortion through IMU forward propagation and laser radar back propagation; fusing the laser point-surface residual error and the visual luminosity residual error by adopting error state iteration Kalman filtering with sequence updating; carrying out object-level segmentation on the local point cloud map, and removing dynamic objects based on a weighted residual mechanism; carrying out loopback detection based on feature matching of deep learning; fusing three factors of a laser-visual odometer, IMU pre-integration and loopback through a factor graph model to obtain a global consistent track and map; autonomous exploration is carried out based on an optimization result, and automatic map construction is realized through terrain analysis and hierarchical path planning. The method effectively improves the positioning precision, robustness and automation level in an agricultural environment with dynamic interference and sparse features, and supports long-time reliable operation.

This invention discloses a vehicle cooperative positioning system for traversing continuous blind zones, comprising a module for acquiring multi-source observation data, a module for constructing a spatiotemporally consistent cooperative factor graph, a cooperative constraint optimization module, an adaptive cooperative strategy adjustment module, and a cooperative information re-fusion module. This invention belongs to the field of data processing technology, specifically a vehicle cooperative positioning system for traversing continuous blind zones. This scheme employs observability-driven cooperative constraint construction and spatiotemporally consistent cooperative factor graph modeling, maintaining high observability and stable positioning even in scenarios without satellite signals, achieving unified fusion of multi-source heterogeneous data. It utilizes communication quality-driven adaptive cooperative switching and predictive prior constraints for delay-compensated re-fusion, achieving autonomous positioning without drift during communication interruptions and unified global trajectory calibration after communication recovery, providing high-precision and continuous positioning support for traversing continuous blind zones.

The application relates to the field of engineering vehicles, in particular to a mapping and positioning method and system based on a laser radar and an engineering vehicle. The method comprises the following steps: acquiring multi-sensor information; performing coordinate transformation on the multi-sensor information to generate an initial value of a pose; determining self-defined point cloud information according to laser point cloud information; calculating a constraint factor according to the self-defined point cloud information and optimizing a factor graph model according to the constraint factor; and outputting trajectory information and a global map according to the point cloud information, the multi-sensor information, the initial value of the pose and the factor graph model. The mapping and positioning method based on the laser radar improves the algorithm for map optimization, so that the mapping and positioning method can use a 6-axis inertial sensor to obtain pose information without affecting the positioning accuracy, and the accuracy and robustness of the LIO-SAM algorithm applied to an automatic driving vehicle and a data set are improved.

The application relates to the technical field of surveying and mapping geographic information, and discloses a surveying and mapping intelligent analysis system based on multi-source data fusion, which comprises a multi-source acquisition and space-time reference alignment module, a factor graph consistency fusion and gate module and the like. The multi-source acquisition and space-time reference alignment module is configured to lock a main clock source through a hardware timing protocol, set a hard time alignment threshold, receive multi-source sensor data, judge a time synchronization deviation fed back by the sensor, and output original observation data flow when the time synchronization deviation is lower than the hard time alignment threshold. The factor graph consistency fusion and gate module is configured to receive the original observation data flow and build a global factor graph containing a full-source data constraint chain. The hardware timing protocol locking and hard time alignment threshold mechanism are adopted to establish a unified space-time reference at a physical level from the source, eliminate spatial misplacement caused by sensor sampling differences, realize high-precision consistency fusion of multi-source data, and solve the problems of difficulty in determining deviation sources and low fusion efficiency in traditional technologies.

The invention discloses an underwater SLAM method and system combining a region-level descriptor and LSH retrieval, belongs to the technical field of autonomous navigation of underwater robots, and solves the problems that track plotting errors can be accumulated along with time due to particularity of an underwater environment, and a traditional SLAM method lacks an effective position sensing mechanism to suppress the drift, so that the navigation accuracy is poor. And meanwhile, an effective preprocessing mechanism for improving the image quality is lacked. The method comprises the following steps: acquiring and preprocessing a polar coordinate image; acquiring a prior pose of a current key frame, inputting the prior pose to a position sensing ICP, performing scanning matching on point clouds of adjacent key frames in sequence, and establishing a sequence constraint relation; extracting an occupied area from the historical key frame point cloud, constructing an area descriptor, carrying out neighbor retrieval, screening candidate key frames, verifying matching, and generating an effective loopback constraint; and constructing a factor graph, performing global graph optimization, accumulating and updating environment point clouds, and generating an underwater environment map. The method is suitable for the underwater robot in the scenes of ocean exploration, environment monitoring, search and rescue operation and the like.