53 results about "Visual matching" patented technology

The invention discloses an AI-based creative thinking auxiliary generation method and system, relates to the technical field of artificial intelligence, and solves the problem of inaccurate sound and picture matching in a traditional method by performing timestamp alignment and feature extraction on audio data and a visual image frame and calculating the correlation between the audio data and the visual image frame by using a cross-modal attention mechanism. According to the method, user eye movement track data is introduced, real attention points of a user are mapped into a visual sequence, and an optimized weight matrix is generated by constructing attention masks and fusing model attention weights, so that a generation result is more in line with perception key points of the user. And meanwhile, a feedback mechanism is established based on the synchronization error score, and when the sound and the picture are detected to be asynchronous, the visual frame timestamp can be dynamically adjusted, so that the self-adaptive correction of the content is realized. On the whole, the method has remarkable advantages in the aspects of improving modal alignment precision, enhancing user perception consistency and optimizing generation result naturalness.

The invention relates to the technical field of stereoscopic vision, and discloses a synchronous speed visual matching method and system, electronic equipment and a storage medium, and the method comprises the steps: synchronously collecting a stereoscopic video sequence with a predefined frame rate; executing multi-target hybrid tracking and motion induction detection, and outputting target motion information including position and velocity vectors; extracting hierarchical motion features of the target from continuous multiple frames of the stereoscopic video sequence, and performing unified space-time coding; under geometric constraints of stereoscopic vision, scale cosine similarity, direction similarity and trajectory consistency measurement are calculated and serve as observation evidences to be input into the probabilistic reasoning model for fusion, and a posterior probability representing matching reliability is output; the weight distribution of the speed similarity and the direction similarity is adjusted according to the motion characteristics of the targets in the scene, and the stable corresponding matching relation between the left view target and the right view target is established. According to the method, high-time-resolution information can be utilized, motion features and geometric constraints can be effectively fused, and the method has self-adaptive capacity.

The invention relates to the technical field of target detection, in particular to a low-slow-small flight target detection method based on visual matching, which comprises the following steps: running a real-time target detection thread and a periodic salient target detection thread in parallel, and performing low-slow-small target detection on each frame of visual image by the real-time target detection thread; the periodic salient target detection thread is executed once every a preset period, sky segmentation and saliency detection are carried out on the current frame of visual image, salient candidate targets in a sky area are extracted, the salient candidate targets are added into a candidate target queue, and priority ranking is carried out on the salient candidate targets; and if the significant candidate target with the highest priority enters a preset countering distance, detecting the confidence coefficient of the significant candidate target through a real-time target detection thread, and performing countering decision. According to the method, the problems of difficult identification, easy tracking loss, insufficient tail end precision and the like of the low-slow small flight target in a complex background are effectively solved.

The invention provides an unmanned aerial vehicle navigation deception detection method based on cross-view visual matching. The method comprises the following steps: acquiring a first aerial image collected by an unmanned aerial vehicle at the current moment and unmanned aerial vehicle positioning information; acquiring a first satellite image of the corresponding area based on the unmanned aerial vehicle positioning information; respectively inputting the first aerial image and the first satellite image into a pre-trained first twin network branch and a pre-trained second twin network branch, generating a first aerial image cross-view matching feature and a first satellite image cross-view matching feature, and sharing parameters of the first twin network branch and the second twin network branch, each of the ViT and the ViT comprises a ViT encoder and a cross-view feature mapping module; and determining whether a navigation spoofing attack exists based on the first aerial image cross-view matching feature and the first satellite image cross-view matching feature. By implementing the method, the accuracy and generalization ability of navigation deception detection can be effectively improved.

The invention relates to an inertial vision fusion positioning method and system based on graph optimization in an indoor cross-floor environment, and the method comprises the steps: obtaining the short-distance relative pose estimation data of a pedestrian based on the multi-source motion parameters of the pedestrian; based on the priori three-dimensional feature map and the image of the current scene of the pedestrian, acquiring visual absolute pose data; and based on a pre-constructed factor graph model, fusing the relative pose estimation data and the visual absolute pose data, and obtaining the global optimal three-dimensional position and pose of the pedestrian through nonlinear optimization solution. According to the method, continuity of deep coupling inertial navigation and absolute precision of visual positioning are optimized through the factor graph, inertial navigation accumulated drift is effectively inhibited, the problem of visual matching failure is solved, decimeter-level positioning precision (the average error is 0.24 m) and 100% continuous positioning are realized in an indoor cross-floor complex scene, and positioning robustness is remarkably improved.

The application discloses a method for cluttered scene target object grasping based on visual-language-action joint modeling. The application uses object-centered representation to realize a method for cluttered scene target object grasping based on visual-language-action joint modeling, processes the object-centered representation through a pre-trained visual-language model and a grasping model, obtains visual-language features and grasping features of each bounding box, and uses a transformer to implement cross-attention mechanisms among visual-language-action multimodality, generates visual-language-action cross-attention features, and then generates decisions and executes, so that higher sample utilization is realized, and additional data collection and training in the simulation-physical migration process are avoided; compared with a two-stage strategy, visual attributes and planner screening rules for language-visual matching need not be artificially designed, so that more flexible language instructions can be adapted, and better task generalization is achieved.

The invention provides a same-view-field cross-lens real-time target tracking system and method based on visual matching, and relates to the technical field of multi-target tracking, and the system comprises an acquisition module, a target detection module, a feature extraction module, a trajectory tracking module, a database management module and a cross-lens matching module. Wherein the trajectory tracking module can track a detected target in combination with target feature information, generate an identity label and output trajectory information, so as to ensure identity continuity and trajectory integrity of the target under the same lens; the database management module is used for uniformly storing the identity label, the feature information and the track information of the target and providing reliable data support for cross-lens tracking; and the cross-lens matching module is used for matching targets under different lenses based on the information in the database management module so as to realize cross-lens association of multiple lenses in the same view field. Through cooperation of multiple modules, multi-lens cross-lens real-time tracking under the same field of view can be effectively realized, and the accuracy and real-time performance of target matching are improved.

This invention discloses a low-altitude visual matching navigation method, device, system, and storage medium, comprising: in an offline phase, optimizing an aerial image sequence into a high-fidelity 3DGS map model; in an online phase, rapidly fusing coarse poses for rendering using inertial pre-integration and global descriptor retrieval; based on the coarse pose, using the 3DGS differentiable rasterization pipeline to synthesize a high-fidelity new perspective reference image in real time by jointly optimizing the pose increment and reference view fusion weights; obtaining a 2D-2D correspondence between the real-time image and the new perspective reference image through deep learning matching, and converting the depth map synchronously generated by the 3DGS model into a 2D-3D association; and calculating the high-precision visual pose of the aircraft through the PnP algorithm and nonlinear optimization. This invention can improve the problems of low quality of new perspective reference images, insufficient multi-sensor information fusion, and limited positioning accuracy in complex low-altitude scenarios.

The invention discloses a multi-scene-oriented security robot intelligent inspection method and system, and relates to the field of intelligent security, and the method comprises the steps: collecting machine body vibration data generated by interaction between a robot and the ground in an environment with good visibility, and building a standard fingerprint band and a historical image; in a low-visibility environment, the robot collects a fuzzy visual image and vibration data in real time, generates a current inspection fingerprint and matches the current inspection fingerprint with a standard fingerprint band; and when deviation is detected, calling a visual memory library and determining the closest reference landmark by utilizing a deep learning visual matching network, so that the steering angle of the robot is adjusted, and the robot is enabled to be close to the central path of the fingerprint zone again. The method can be applied to industrial control software, is deployed on security robots of multiple models to realize process control of inspection, solves the problem that the security robots are unstable in inspection positioning in a low-visibility environment, and realizes path self-correction and intelligent inspection control based on vibration fingerprints and depth vision matching.

The invention relates to an automatic driving vehicle-mounted sensor fusion positioning method, device and equipment and a medium, and the method comprises the steps: fusing a coarse positioning result corresponding to I MU data and a coarse positioning result corresponding to wheel speed meter data based on a preset Kalman filtering algorithm, recalculating the position deviation, and obtaining the position deviation of the I MU data; carrying out constraint calculation on the relative displacement so as to determine fusion positioning information of the autonomous vehicle; and determining a vehicle moving track of the autonomous vehicle according to the fused positioning information, and performing map matching or visual matching based on the vehicle moving track to complete map matching positioning of the autonomous vehicle. According to the invention, high-precision positioning of data fusion of the automatic driving vehicle-mounted sensor can be realized without the help of high-cost and high-precision inertial navigation equipment under the condition that the GNSS signal is unavailable.

The invention belongs to the technical field of intelligent traffic, and particularly discloses an unlisted non-motor vehicle driver identification method and system and a program product, and the method comprises the steps: carrying out the time-space correlation matching of a front image and a back image of a driving non-motor vehicle and a driver, and an electronic license plate number of the non-motor vehicle; the method comprises the following steps: acquiring a front image and a back image of a non-motor vehicle, performing visual matching on the corresponding front image and back image, performing space-time and visual double-base judgment based on results of the two matching modes, determining a target front image corresponding to a target back image, finally performing face recognition on the target front image, and determining identity information of a corresponding non-motor vehicle driver without listing a tag. According to the method, the identity of the non-motor vehicle driver can be accurately and efficiently traced and identified, so that a complete evidence chain is formed, and an effective basis is provided for traffic management personnel to treat and punish non-motor vehicle non-listed behaviors.

The invention discloses a ship robot double-wire welding process and welding equipment based on model driving and visual matching fusion. The ship robot double-wire welding process comprises the following steps that an adaptive robot double-wire welding process is selected; visual scanning, model introduction and fusion are completed by the structured light shooting camera; clicking the welding plan to generate a welding operation file; a welding seam information json file is imported, and a robot welding operation list is generated; gun cleaning operation is executed, and welding is conducted after gun cleaning is completed; a welding task is automatically executed according to the welding operation list, gun cleaning is automatically completed according to the length of the welding seam, and welding of the next welding seam is continued; through the model driving and visual matching fusion algorithm, collaborative operation of model importing and visual scanning is achieved, the success rate of one-time workpiece recognition is high, the recognition precision is high, the welding seam positioning time is shortened, the arcing rate of the robot is increased, and the overall welding capacity and the unit area output efficiency are improved.

Method (200) for recognizing visually matching tones, wherein the method comprises: Receiving visual training data (105) at a visual coder (110) that has an initial machine learning (ML) model; Identifying data corresponding to a visual object in the visual training data (105) using the first ML model; Receiving audio training data (107) synchronized with the visual training data at an audio forwarding regulator (115) which has a second ML model, wherein the audio training data (107) has a visually matching tone and a visually mismatched tone, both of which are synchronized with one and the same frame in the visual training data (105) which contains the visual object, wherein the visually matching tone corresponds to the visual object, whereas the visually mismatched tone is generated by a sound source which is not visible in the same frame; Filtering data matching the visually appropriate tone from an output of the second ML model using an information bottleneck (120); and Training a third ML model following the first and second ML models (235) using the data corresponding to the visual object and data corresponding to the visually inappropriate tone.

The invention provides a spatial positioning method and device, equipment and a storage medium. The method comprises the following steps: constructing a neural radiation field based on three-dimensional point cloud data and a multi-view discrete image of a target scene; obtaining user behavior data which comprises a basic pose image shot by a user and / or editing data when a virtual object is edited, extracting an effective pose from the user behavior data, and generating a virtual shooting pose set; generating a rendered image set bound with a pose through the neural radiation field; and finally, receiving a positioning request, and determining a target pose in combination with the to-be-positioned image, the rendering image set and the three-dimensional point cloud data. The target pose is determined in combination with the multi-modal data, the precision limitation of a single-modal sensor in a complex scene is broken through, visual matching accumulative errors and dynamic environment interference are eliminated, the computing resource consumption is reduced while the positioning precision is improved, meanwhile, the effective pose is extracted in combination with the user behavior data for rendering, and the user experience is improved. And the problem of complex or inaccurate calculation caused by redundant rendering is avoided.

The system receives an indication of a sensitive operation. The system obtains a unique ID of a user's UE. Based on the unique ID of the UE, the system retrieves a visual authentication method including a visual ID. The system records the visual ID, and retrieves a corresponding stored visual ID. The system performs a liveness check associated with the visual ID, to determine whether the visual ID is a recording or a live version of the visual ID. Upon determining that the visual ID is the recording, the system refuses to authenticate the user. Upon determining that the visual ID is the live version of the visual ID, the system compares the visual ID and the corresponding stored visual ID to determine whether the visual ID and the corresponding stored visual ID match. Upon determining that the visual ID and the corresponding stored visual ID match, the system authenticates the user.

Audio matching is performed between a first video content item and a second video content item to identify a matching audio segment. First temporal boundaries within the first video content item and second temporal boundaries within the second video content item corresponding to the identified matching audio segment are identified. A visual matching between the first video content item within the first temporal boundaries and the second video content item within the second temporal boundaries is performed using a modified visual similarity threshold that is lower than a baseline visual similarity threshold. Whether a match exists between the first and second video content items is determined based on the visual matching.

The invention relates to a multi-stage cognitive modeling and parameter optimization method based on a visual pairing comparison task, and aims to solve the technical problems of incomplete cognitive process modeling, weak crowd distinguishing ability, task parameter empirical and the like in the existing VPC task evaluation technology. Precise evaluation of the cognitive function and optimization design of task parameters are achieved through a five-step method, wherein VPC task parameterization design and eye movement data collection are carried out; constructing a'familiarity accumulation-memory attenuation-novelty attention and utility 'three-stage cognitive model; performing backstepping on individual free parameters based on maximum likelihood estimation; carrying out multi-population cognitive parameter distribution modeling and subtype division; and task parameter closed-loop optimization based on simulation and efficiency discrimination. According to the method, quantitative distinguishing of learning efficiency, memory stability and novel preferences is achieved, cognitive parameter recovery precision and multi-crowd distinguishing efficiency are improved, a standardized and transplantable technical path is provided for early screening of cognitive impairment and crowd typing, and the method is suitable for cognitive neuroscience research and clinical cognitive evaluation scenes.

A method and system for automated image annotation based on a landscape element knowledge graph and visual matching is disclosed. The method includes: constructing a landscape element knowledge graph; performing structured association modeling of landscape nodes, landscape element nodes, image nodes, and cultural knowledge nodes; extracting relevant landscape nodes from the graph based on user-inputted images and shooting locations to limit the range of elements to be identified; extracting features from the input image using a zero-shot visual matching model and comparing them with image samples of candidate landscape elements to obtain identification results; extracting attributes and associated cultural knowledge content from the graph based on the identified landscape element nodes; semantically fusing the identification results with cultural knowledge according to preset rules to generate structured or natural language annotation text; and outputting the annotation text to a terminal for display. This invention combines geographic location, knowledge graph, and visual matching to achieve accurate, information-rich, and culturally profound automated annotation of landscape images.

The present disclosure provides a webpage data processing method and device, equipment and medium, relates to the technical field of artificial intelligence, in particular to the technical field of webpage development and deep learning. The method comprises: determining an interactive operation for a target webpage; obtaining a screenshot of the target webpage and attributes of a plurality of webpage elements in the target webpage, the attributes indicating functions possessed by the webpage elements; based on the attributes of the plurality of webpage elements, screening a plurality of candidate webpage elements related to the interactive operation from the plurality of webpage elements; determining the positions and sizes of the plurality of candidate webpage elements, and cutting the screenshot to obtain visual segments of the plurality of candidate webpage elements; determining the intention matching degrees of the attributes of the plurality of candidate webpage elements and the interactive operation; determining the visual matching degrees of the visual segments of the plurality of candidate webpage elements and the interactive operation; and based on the intention matching degrees and the visual matching degrees, determining a target webpage element from the plurality of candidate webpage elements and executing the interactive operation.

The application relates to the technical field of robot control, and discloses a robot task planning method based on a multimodal large model. The application collects multimodal data of a robot motion scene, including voice information and visual information, pre-processes the information to obtain a task instruction sequence required to be executed by the robot and an entity data set in a scene where the robot is located, inputs the pre-processed data into a multimodal large model based on a large language model, calculates, based on the pre-processed data, a semantic evaluation coefficient, a visual matching coefficient, a feasibility coefficient and a priority coefficient corresponding to each task instruction in the task instruction sequence, and obtains a task comprehensive execution coefficient corresponding to each task instruction by weighted summation, reorders the task instructions of the task instruction sequence based on the task comprehensive execution coefficient, and obtains a final task planning sequence, thereby improving the rationality and efficiency of task execution of the robot.

This application discloses a data transmission system and method for operation guidance in remote interventional surgery, relating to the field of intelligent sensor technology. The method includes: acquiring a control dataset; calculating the video end-to-end latency, visual matching ambiguity, and operator hand tremor components based on the control dataset; processing the optimal pairing set corresponding to the control dataset and the visual matching ambiguity using a preset filter to obtain motion state estimates; performing temporal cross-correlation analysis on the video end-to-end latency and operator hand tremor components to calculate the delay-tremor coupling strength; constructing a future position prediction distribution model based on the motion state estimates and the delay-tremor coupling strength; superimposing the future position prediction distribution model with a preset anatomical structure model, and displaying the superimposed result on the current surgical video screen to guide the surgeon in remotely controlling the surgical operation from the control terminal. This application improves the safety of remote precision operations.

This invention discloses a fast relocalization method based on tight coupling between sparse semantic anchors and inertial sensing, belonging to the fields of augmented reality and computer vision. It constructs a lightweight sparse semantic map by extracting sparse feature points with stable geometric and semantic attributes from the environment as sparse semantic anchors. When device tracking is unstable or lost, inertial navigation and visual matching threads based on sparse anchors are activated in parallel. The core PnP algorithm is used to quickly recover the visual pose, and the method is tightly coupled with inertial data for optimization, ultimately outputting a high-precision, smooth six-DOF pose. This solves the problem of excessively long relocalization time and experience interruption in traditional visual SLAM scenarios with fast motion and weak textures, achieving millisecond-level, user-unnoticed tracking recovery, significantly improving the robustness and user experience of augmented reality systems.

The present application relates to the technical field of natural language processing, and more particularly to a script visual asset structured generation method based on a large model, which comprises the following steps: obtaining a script original text and dividing it into scene text units; extracting deep semantic vectors by using a large-scale pre-training language model; mapping literary descriptions to a primary visual feature set based on a visual semantic feature library; encapsulating each component by using a structured mapping engine, and constructing a structured visual asset description model including global scene parameters, entity object parameters and dynamic interaction parameters; and detecting and outputting consistent structured data through a cross-scene logic checking mechanism. The present application realizes the automatic conversion of script text into parameterized data, and improves the semantic analysis depth and visual matching accuracy.

The invention belongs to the technical field of electric digital data processing, and particularly relates to an automatic matching method of an oil and gas production decline curve, electronic equipment and a storage medium. According to the method, mathematical characteristics of a double logarithmic space are utilized, a complex curve matching problem is converted into a linear translation optimization problem, automatic alignment of data points is achieved in combination with an interpolation algorithm, the single well analysis time is shortened to be within 1 minute from 1-2 hours of a traditional method, and efficiency is improved by two orders of magnitude; a numerical optimization algorithm is adopted to replace manual visual matching, and by establishing a centroid initialization-based least square optimization model, an optimal translation parameter is automatically calculated, and deviation caused by human factors is eliminated, so that the parameter inversion precision is greatly improved; an overlap ratio quantitative evaluation mechanism is introduced, a matching result is objectively evaluated by setting a scientific threshold value, the problem that a traditional method lacks a quality judgment standard is solved, and high-precision and high-efficiency automatic matching of oil and gas well production data and a theoretical curve is achieved.

The application provides a material tray automatic identification and deviation correction method for an ASRS system, which comprises the following steps: triggering a multi-modal sensor to collect a tray gray scale image, a three-dimensional point cloud, a label material code and an ultrasonic wave front distance value to form a multi-modal data cache after a stacker is in place; extracting a tray four-corner point sub-pixel coordinate from the multi-modal data cache through equalization and edge detection, and obtaining a normal vector and a distance through point cloud filtering and plane fitting; obtaining a fusion confidence through weighted fusion of an RFID confidence, a visual matching and a geometric verification score, and outputting a tray pose parameter set; solving a six-degree-of-freedom pose based on EPnP and LM optimization, comparing the six-degree-of-freedom pose with a standard picking pose to obtain an offset and a deflection angle, selecting a deviation correction strategy according to a comprehensive offset amplitude, and performing differentiated compensation on different deviation levels. The application improves identification reliability through multi-modal weighted fusion, balances between response time and success rate through deviation grading self-adaptive correction and iterative closed-loop verification.

The present application belongs to the technical field of electric digital data processing, and particularly relates to an automatic matching method of oil and gas production decline curve, an electronic device and a storage medium. The method utilizes the mathematical characteristics of double logarithmic space to convert the complex curve matching problem into a linear translation optimization problem, and realizes the automatic alignment of data points in combination with an interpolation algorithm, so that the single well analysis time is shortened from 1-2 hours of the traditional method to within 1 minute, and the efficiency is improved by two orders of magnitude. A numerical optimization algorithm is adopted to replace artificial visual matching, an optimal translation parameter is automatically calculated by establishing a least square optimization model based on centroid initialization, and the deviation caused by human factors is eliminated, so that the parameter inversion accuracy is greatly improved. A coincidence quantitative evaluation mechanism is introduced, the matching result is objectively evaluated by setting a scientific threshold, the problem of lacking quality evaluation standard in the traditional method is solved, and the high-precision and high-efficiency automatic matching of oil and gas well production data and theoretical curve is realized.

This invention provides a multi-source fusion reliable navigation and positioning method for complex urban nighttime scenarios, including: GNSS, INS, and visual data acquisition and preprocessing to achieve quality control of input data; visual feature extraction and initial matching using deep learning technology; identification and removal of dynamic object interference by combining optical flow residuals and IMU motion parameters; estimation of the confidence of visual feature matching point pairs based on LSTM; construction of a GNSS / INS / Vision tightly coupled navigation and positioning model and a filter optimizer by fusing the matching confidence; and output of the final navigation result. This invention improves the robustness of visual matching in complex environments with changing urban nighttime lighting and dynamic object interference, significantly enhancing the continuity and reliability of autonomous perception and navigation and positioning of the vehicle.

The invention relates to a tactile feedback glove capable of generating skin stretching and squeezing effects and a control method, and belongs to the technical field of man-machine interaction. Comprising a glove assembly and a fingerstall assembly, position and posture synchronization of a virtual hand and a real hand is achieved through a calibration system in a Unity virtual environment, in the interaction process, a computer conducts collision detection on virtual environment interaction actions and communicates with a control system to generate a driving signal, a servo motor is controlled to drive a plastic belt to be combined with a specific mechanical structure, and therefore the mechanical performance of the robot is improved. And rendering tactile feedback under the gripping action in real time. According to the method, the visual matching effect in the interaction process is optimized in the Unity interaction scene, the interaction reality sense of a user is enhanced, and the device can be effectively applied to interaction operation of a VR virtual environment, tactile feedback training of a special scene and the like in an ideal state.

The invention belongs to the technical field of road monitoring, and particularly relates to a multi-modal BEV remote sensing method and system suitable for a high-speed trunk scene, and the method comprises the steps: collecting multi-modal data through a camera, a laser radar and a map unit, and carrying out the time synchronization; visual BEV features, LiDAR BEV features and map BEV features are extracted from the collected multi-modal data; and calculating confidence weights corresponding to the three BEV features, and carrying out weighted fusion on the BEV features based on the confidence weights to obtain fused BEV features. According to the method, three types of heterogeneous modal data of the camera, the laser radar and the high-precision map are fused, and a dynamic confidence coefficient weight calculation mechanism based on the image entropy, the point cloud density and the map-visual matching degree is introduced, so that adaptive evaluation and weighted fusion of the reliability of each modal feature under different environmental conditions are realized; and the perception robustness and the remote detection precision in a complex scene are obviously improved.

Embodiments of the present application provide a mobile robot human tracking method and system based on visual matching, and a medium. The method comprises: collecting a region video in real time, pre-processing the region video to obtain a plurality of single-frame images; performing background compensation on the plurality of single-frame images and performing cross-frame difference processing to obtain difference images; setting search conditions to traverse the difference images to obtain difference image characteristic values; comparing the difference image characteristic values with preset characteristic threshold values to obtain characteristic deviation rates; determining whether the characteristic deviation rates are greater than or equal to a preset deviation rate threshold value; if yes, eliminating image pixel points corresponding to the difference image characteristic values; and if no, storing the corresponding difference image characteristic values to a data set, inputting the data set into a preset position prediction model, outputting human position information, inputting the human position information into a preset following model to output following parameters, and dynamically moving according to the following parameters.