23 results about "Visual flow" patented technology

This invention relates to the field of construction safety monitoring and intelligent visual analysis technology, specifically a wiring-free construction safety monitoring system based on visual algorithm analysis. The system includes: an independent power supply module, a visual perception module, a scene risk topology assessment module, a computing-risk coupling scheduling module, a multi-level wake-up execution module, and a context-related early warning output module. It collects initial visual flow data, remaining power, and predicted charging power of the target monitoring area; divides the topology regions into different risk levels through semantic segmentation and assigns risk weights; dynamically generates computing power scheduling instructions containing the target sampling frame rate and target model complexity; outputs the target violation event identification results; and generates the final early warning result by combining the spatial inclusion relationship between the occurrence location and the topology region boundary. This achieves priority protection for high-risk spatiotemporal nodes and reduces ineffective energy consumption and invalid alarms in low-risk scenarios.

Approaches are disclosed for enhancing frame rate and visual smoothness in real-time video streams through multi-frame interpolation. A classification neural network analyzes two sequential frames, outputting confidence scores that indicate the reliability of motion data for each pixel. These scores determine whether a pixel's motion is accurately described by motion vectors or should be treated as static. The classification results are reused to generate intermediate frames by warping the original frames based on the motion characteristics. Blending weights are calculated by combining warped motion vector confidence values with static values, and a second neural network refines the alignment and blending of candidate frames. This second network predicts intermediate flows and generates new blending weights, which are used to warp and blend the candidate frames, ultimately producing a final interpolated frame that enhances visual smoothness and consistency in the video stream.

Various implementations include methods for processing images from a device camera to identify potential malicious attacks on the camera. These implementations may include using multiple different trained image processing models or vision pipelines to process images received from the device's camera to obtain multiple different image processing outputs, and performing multiple consistency checks on the multiple different image processing outputs. Such consistency checks compare two or more selected outputs from the multiple different outputs to detect inconsistencies that may be associated with or caused by an attack on the camera. Indications of an attack on the camera may be reported to and considered by the device's autonomous driving system, or otherwise addressed in one or more mitigation actions.

The application provides a kind of machine vision collaborative control method and system applied to intelligent processing, it is related to computer vision technical field.First, the association rule set of the visual feedback of processing equipment and control instruction is established, covering the mapping relationship of visual feature sequence and control instruction template and collaborative trigger condition, then based on the association rule set, the real-time collected processing area visual flow is carried out feature sequence extraction and matching processing, obtains control instruction template and collaborative trigger condition satisfaction parameter, then according to the above parameter generates dynamic control parameter adjustment sequence, the sequence is input into the collaborative control module of processing equipment, executes parameter loading and action calibration processing, finally, the mapping weight in association rule set is updated based on the processing result, and the collaborative control log is output, so that the real-time perception and accurate collaborative control of processing equipment in precision manufacturing are realized, and the quality and efficiency of part processing and docking are improved.

The application discloses a visual water flow speed measurement method in a non-uniform light field, comprising the following steps: S1, collecting video frames and converting the collected video frames into single-channel images; S2, performing mask operation based on the single-channel images, shielding irrelevant background interference and extracting ROI; S3, performing smooth filtering on the images after the mask operation and removing abnormal noise points; S4, performing local Gaussian threshold segmentation by using initial values to obtain binary images, wherein the initial parameters include neighborhood size and threshold bias; and S5, detecting feature points by using an ORB feature point detection algorithm on the binary images. Through the local Gaussian threshold segmentation technology, the threshold is dynamically calculated based on the local neighborhood gray scale characteristics of pixels instead of relying on a global fixed threshold, so that the non-uniform illumination space and time fluctuation caused by water surface reflection, building shadow, weather change and the like are effectively offset, the core assumption of the brightness constancy of the optical flow method is met, and tracking failure caused by the distortion of the imaging features of the target is avoided.

The present disclosure belongs to the technical field of nuclear power, and particularly relates to a nuclear power digital maintenance procedure structuring method and device. The present disclosure converts the natural language description of the maintenance procedure into a structured procedure flow chart through a visual flow chart editing mode, dynamically associates each step with related data in a multi-type data storage system, forms a seamless set of multi-element data, and finally generates a digital procedure object that can be parsed and executed by a computer, which is friendly in interface and convenient in operation, and lays a foundation for subsequent intelligent application.

The present application provides a kind of based on multi-source vision sensor fusion's blast furnace slag iron flow detection method and system, belongs to blast furnace slag iron flow detection technical field, including: hyperspectral image data input slag iron ratio estimation model, the composition of molten iron and slag is identified and separated, extract the spatial distribution information of iron phase and slag phase, calculate slag iron ratio;Slag iron flow time series data input time series visual flow velocity perception model, the dynamic analysis of slag iron flow flow process, extract time evolution characteristics and flow trajectory information, calculate slag iron flow velocity;Three-dimensional point cloud data input three-dimensional point cloud reconstruction flow section analysis model, the spatial form of slag iron flow is modeled and analyzed, extract three-dimensional geometric structure and cross-sectional shape characteristics, calculate slag iron flow cross-sectional area.Slag iron ratio, slag iron flow velocity, slag iron flow cross-sectional area are jointly modeled, the adaptive calculation of slag iron flow mass flow is carried out, and slag iron flow detection result is output.The present application can effectively carry out slag iron flow detection.

Approaches to improving frame rate and visual smoothness in real-time video streams through multi-frame interpolation are revealed. A neural classification network analyzes two consecutive images and outputs confidence scores indicating the reliability of motion data for each pixel. These scores determine whether the motion of a pixel is accurately described by the motion vectors or should be treated as static. The classification results are reused to generate intermediate frames by warping the original images based on their motion properties. Blended weights are calculated by combining the confidence scores of the warped motion vector with static values, and a second neural network refines the alignment and blending of the candidate images.This second network predicts intermediate streams and generates new blend weights that are used to warp and blend the candidate images, ultimately producing a final interpolated image that improves the visual smoothness and consistency in the video stream.

This invention relates to the field of visual analytics technology, and discloses a method and apparatus for flow field visual analysis driven by a large language model. The method includes: acquiring a flow field dataset to be analyzed and user commands; performing semantic analysis on the user commands using a large language model to obtain structured commands; based on the structured commands, retrieving corresponding target data from the flow field dataset using the large language model, and generating descriptive text corresponding to the target data; running a visualization agent based on the target data and the descriptive text to obtain visualization results; and displaying the target data according to the data display method indicated by the user commands. By using a large language model to achieve natural language-based interaction, user operation commands or text commands are converted into structured commands, thereby presenting visualization results and explanations in real time according to user needs, greatly improving the user experience.

This invention discloses a naked-eye 3D augmented reality interactive display system, relating to the field of naked-eye 3D display and real-world interaction technology. The invention acquires and fuses user limb depth images through a multi-source depth data acquisition module, completes the mapping of depth and parallax and viewpoint correction through a parallax physical depth mapping module, divides regions to determine occlusion types through an occlusion relationship determination module, generates a dynamically smooth occlusion layer through an adaptive occlusion layer generation module, and displays the occlusion effect fusion display module by superimposing the occlusion layer with multi-viewpoint virtual images. This effectively solves the problems of inaccurate occlusion relationships between user limbs and virtual objects and visual abruptness in naked-eye 3D interaction, significantly improving the spatial consistency and visual smoothness of the interaction, and achieving a realistic and natural naked-eye 3D augmented reality interactive experience.

The utility model relates to water conservancy exploration technical field, concretely is a kind of four-legged robot of three-dimensional topographic survey and visual flow test function, including organism, the lower part of organism is provided with walking foot, and walking foot is provided with four groups in the lower part of organism, and the upper part of organism is provided with controller, and walking foot provides intelligent control, the upper part of organism is installed with measuring mechanism, and the outside of measuring mechanism is provided with first protective cover. The four-legged robot of three-dimensional topographic survey and visual flow test function is simulated machine dog walking by the setting of four groups walking foot, compared with wheeled driver applicability stronger, it is convenient to adapt the measurement of complex water conservancy environment, cooperate measuring mechanism and measure water conservancy environment, and the outside of measuring mechanism is provided with first protective cover and second protective cover, and it provides protection for measuring mechanism.

A computer-implemented discussion interface system enables structured discourse between hierarchically differentiated user groups through a novel architectural arrangement. The system presents a graphical user interface with a first comment stream region positioned above a second comment stream region, with a shared input element positioned between them. A user classification module assigns users to hierarchical groups based on profile data analysis and classification criteria. A comment routing module receives text input through the central input element and directs display to either the upper or lower comment stream region based on the user's hierarchical group assignment. Comments from first group users move upward from the input element while comments from second group users move downward, creating bidirectional visual flow.

This invention relates to the field of video processing technology and discloses a generative video motion redrawing method and system based on multimodal pose control. The method involves: acquiring the original video frames of the input video and performing pose keypoint detection to obtain a pose keypoint sequence; mapping the pose keypoint sequence to a pose latent space to obtain a pose embedding vector, and extracting semantics from the text description to obtain a text embedding vector, while simultaneously extracting appearance features from a reference image to obtain a visual embedding vector; fusing these to obtain a multimodal control vector; performing multi-stage denoising processing on the multimodal control vector and a random noise image input diffusion model to obtain a generated video frame; and performing temporal constraint enhancement on the generated video frame to output a redrawn video frame. This invention improves the accuracy and robustness of pose keypoint extraction, effectively suppresses pixel abrupt changes and pose shifts between adjacent frames, and improves the temporal coherence and visual smoothness of the generated video.

This invention discloses a distributed acquisition method and system for multimodal grasping data for embodied intelligence. The method includes: achieving high-precision clock synchronization between the Linux master control terminal and the Windows vision slave control terminal through the Chrony protocol, with the deviation controlled within 2ms, and establishing a UDP start / stop control protocol; employing preheating asynchronous writing to ensure zero-frame loss synchronization of the visual stream; acquiring the golden pose and tactile incremental intensity based on manual teaching to generate physically reasonable six-degree-of-freedom pose perturbation and force perturbation samples; integrating dual-sided optical tactile sensors to solve the problem of missing proprioception caused by serial port congestion; aligning tactile, proprioception, and visual data with a unified timestamp in the offline stage, exporting the standard Zarr format, and directly supporting supervised fine-tuning of VLA large models. This invention ensures continuous high-fidelity acquisition of proprioception data under serial port congestion through bus arbitration, and achieves distributed acquisition of multimodal grasping data through tactile tensor entropy, nonlinear collapse detection, and cross-modal cross-correlation analysis.

This invention relates to a method and system for detecting micro-damage in pipelines based on visual flow field features. The method includes: constructing a feature space adaptation layer by extracting features and mapping them across models of pipeline images acquired by different types of UAVs to generate a feature space adaptation layer corresponding to each model; calling the corresponding adaptation layer according to the UAV model identifier to perform feature space transformation and generate a unified feature map; acquiring three-dimensional displacement data and projecting it onto the image coordinate system to obtain sensor displacement projection values; calculating the initial optical flow field based on the unified feature map; using the sensor displacement projection values ​​to perform error compensation on the initial optical flow field to generate a corrected optical flow field; performing multi-scale convolution on the unified feature map to extract spatial structure features; performing temporal difference and convolutional coding on the corrected optical flow field to extract motion mode features; performing cross-modal fusion through a cross-attention mechanism and inputting the data into a classification network; outputting the micro-damage probability at each pixel location and generating a detection result with damage annotation.

This invention proposes an intelligent segmentation method for reticular fiber stained pathological images and its application. Addressing the limitations of existing technologies in identifying reticular fiber structure disruptions to define tumor boundaries and artifacts caused by sliding windows, this invention employs adaptive sliding window segmentation and multi-stage pre-filtering to extract effective image blocks. The input is a dual-stream network, where structural and visual flows are fused through cross-attention based on tissue physical size mapping receptive fields to achieve collaborative verification of visual abnormalities and structural disruptions. Model training utilizes a dynamic boundary-aware loss term with physical scale constraints for joint optimization. Finally, a spatial weight fusion mechanism combined with a graph model incorporating physical priors eliminates breakage artifacts and smooths global boundaries. This invention is primarily used for high-fidelity lesion segmentation of pituitary neuroendocrine tumors.

This invention discloses a method and application for analyzing emotion change trends in interactive contexts, relating to the field of artificial intelligence technology. The method includes: Step S1: Real-time collection of multi-source heterogeneous data from the user in the current interactive context, wherein the multi-source heterogeneous data includes at least visual stream data, auditory stream data, text behavior stream data, brain current data, and environmental context data; Step S2: On-device privacy anonymization processing of the multi-source heterogeneous data, extraction of high-dimensional feature vectors, and construction of an adaptive multi-scale time window based on the semantic boundaries of interactive events, temporal alignment and fusion of feature vectors at different time granularities to generate multimodal fusion features for the current moment; Step S3: Loading the user's personalized emotional dynamic baseline. This invention improves the human-computer interaction experience, service success rate, and user emotional health level in application scenarios such as intelligent customer service, online education, intelligent cockpits, and mental health assistance.

This invention provides a method for visual flow data anomaly identification and supplementation, comprising: labeling stations with topological tags based on watershed topology; acquiring and preprocessing basic hydrological data of the stations; constructing candidate supplementation paths based on the preprocessed basic hydrological data, wherein the candidate supplementation paths include at least a path derived based on the water level-flow relationship and a predicted path based on historical water level-flow time series data; constructing an anomaly hierarchical identification mechanism to identify abnormal time series change characteristics of online flow data and scene characteristics of video monitoring images, generating corresponding abnormal time series feature labels and anomaly cause labels; selecting a matching supplementation method from the candidate supplementation paths based on the topological tags, abnormal time series feature labels, and anomaly cause labels, and outputting the corresponding flow supplementation data and anomaly cause. This invention can automate and intelligently identify and supplement visual flow data anomalies, improving the continuity and reliability of flow data in complex scenarios.

This invention discloses a synthetic aperture radar target recognition method based on the fusion of physical perception and vision. This method effectively addresses the problems of insufficient utilization of physical topological information and difficulties in feature semantic alignment in existing methods by constructing a GAT-Former dual-stream heterogeneous network. This invention models the dual-stream input space of the complex domain physical and visual flows, explicitly solves the scattering center, and constructs a dynamic topological graph. During the recognition process, a physical-guided cross-attention fusion mechanism and a heterogeneous integration strategy are used. Geometric topological features are extracted using deep unfolding networks and graph attention networks, visual texture features are extracted using ConvNeXt, and the final decision is optimized through weighted fusion with ResNet-50. Simulation results show that, compared with other benchmark methods, the proposed method significantly improves fine-grained classification accuracy and noise robustness on the ATRNet-STAR dataset, achieving deep complementarity between physical mechanisms and visual texture.

This invention discloses a real-time piano playing posture correction system based on hand shape recognition, specifically in the field of music-assisted teaching technology. It addresses the problems of joint positioning loss and inaccurate dynamic force assessment under occlusion during playing. First, a primary contour is generated by fusing visual flow entities and reflection edges to construct a key mapping mesh. Then, inverse kinematics deduction is performed using skeletal proportion constraints to reconstruct the three-dimensional skeleton sequence under occlusion. Next, focusing on the key sinking time window, the vertical displacement gradient of the metacarpophalangeal joints and the rate of change of interphalangeal joint angles are compared to quantify joint motion values ​​and construct a force deformation feature vector. The extracted feature vectors are input into a standardized database for comparison, and posture error labels are parsed out. Finally, graphic and synchronous speech correction signals are generated at rests or long notes in the electronic score, constructing a closed-loop correction system that does not interrupt the playing rhythm, providing scientific support for piano posture-assisted teaching.

The invention discloses a feature map up-sampling method and system based on high-resolution guided anisotropic Gaussian kernel prediction, a storage medium and a product, belongs to the technical field of computer vision and image processing, and particularly belongs to the technical field of image up-sampling, feature recovery and high-resolution reconstruction. The problems that in the prior art, the calculation overhead is large, the optimization process is unstable, the capability of being suitable for any scene after one-time training is lacked, and the memory and calculation complexity is increased along with resolution are solved. The method comprises the following steps: acquiring a high-resolution natural image, and constructing an up-sampling self-supervised data set; constructing a Gaussian kernel prediction model based on multi-scale characteristic anisotropy, and carrying out multi-task self-supervised training by adopting a data set; and integrating the trained multi-scale feature anisotropy Gaussian kernel prediction model into a visual assembly line, and carrying out up-sampling on a low-resolution feature map. The method is used for feature map up-sampling.