1271 results about "Real time video" patented technology

The invention relates to the technical field of video behavior recognition, and discloses a behavior recognition and early warning system for a real-time video stream. The system comprises a spatio-temporal feature modeling module, a behavior fragment extraction module, an anomaly propagation modeling module, a risk area positioning module and an early warning strategy generation module. The spatial-temporal feature modeling module builds a dynamic model based on historical data, captures a skeleton key point three-dimensional coordinate sequence, a motion optical flow vector field and a micro-expression intensity spectrum, and outputs a theoretical behavior mode vector; the behavior fragment extraction module generates a multi-modal difference feature tensor through cross-modal difference analysis; the exception propagation modeling module generates an exception propagation path risk probability distribution cloud picture in combination with spatial constraint and trajectory information; the risk area positioning module identifies a high-risk area and marks a boundary; and the early warning strategy generation module dynamically configures monitoring parameters, starts high-frame-rate micro-expression capture for a high-risk area, and performs a track disturbance test on an adjacent area.

The invention relates to the technical field of computer vision, and discloses a real-time video analysis method based on deep learning. The method comprises the following steps: acquiring a real-time video stream through image acquisition equipment, and performing frame segmentation processing to generate a continuous video frame sequence; and extracting features of the video frame sequence by using a pre-trained convolutional neural network to obtain a multi-dimensional feature vector, inputting the multi-dimensional feature vector into the time sequence analysis model to calculate dynamic relevance, and outputting an inter-frame movement track and object behavior features. And constructing a scene understanding map containing a spatial position and a time evolution relationship according to the above-mentioned data, and carrying out abnormal event detection and generating event marking data based on the map. And performing semantic analysis on the event marking data, determining an abnormal event type and a confidence score, triggering a real-time alarm signal according to a result, and updating a historical event database. In the analysis process, the resource occupancy rate of the system is continuously monitored, the calculation precision is dynamically adjusted, a degradation processing mechanism is started when a preset threshold value is exceeded, and key area analysis is preferentially guaranteed.

The invention discloses a video analysis-based multi-scene operator violation behavior identification method and system, and belongs to the technical field of intelligent operation safety monitoring and artificial intelligence identification, and the method comprises the steps: collecting a real-time video stream of a multi-scene operation site; recognizing a continuous action time sequence in the real-time video stream by using an action recognition depth model; constructing the continuous action time sequence into an action behavior sequence; the action behavior sequence is constructed into a directed behavior graph with time, space and action labels, the directed behavior graph is compared with a directed behavior graph corresponding to the standard action behavior sequence, and illegal behaviors are recognized; and carrying out multi-mode early warning on the identified illegal behaviors. According to the method, the bottleneck that the traditional image recognition technology is weak in action sequence semantic understanding and poor in environmental adaptability is broken through, and accurate recognition and real-time early warning of illegal behaviors in multi-scene operation are achieved.

The invention relates to the field of data processing, and provides a 5G-based video call information processing method and system, and the method comprises the steps: continuously obtaining a real-time video frame sequence and 5G network environment perception data in a video call scene, carrying out the multi-dimensional state mapping processing of the 5G network environment perception data, constructing a network transmission adaption model, and carrying out the real-time video frame sequence and 5G network environment perception data. Generating a video coding control instruction based on the network transmission adaptation model, performing content-aware coding conversion on the real-time video frame sequence, and outputting a coding optimization stream; in the transmission process of the coding optimization stream, link state fluctuation information is obtained through a 5G network feedback channel, transmission strategy dynamic calibration is performed on the coding optimization stream according to the link state fluctuation information, and a calibration transmission stream is obtained; and carrying out decoding time sequence alignment processing on the calibration transport stream, generating a visual call output sequence which is synchronous with the time of the original video stream unit, and pushing the visual call output sequence to a receiving end presentation device.

The invention relates to the technical field of transformer substation intelligent patrol, and provides a transformer substation three-dimensional fusion patrol method and system based on digital twinborn and autonomous identification. According to the method, a fused three-dimensional model is constructed through multi-source data acquisition and a three-dimensional Gaussian splash algorithm, and in combination with deep learning-based point cloud semantic segmentation and clustering, an equipment-patrol means coverage relationship is generated. Creating a virtual inspection proxy object based on a three-dimensional virtual environment, and controlling terminals such as an unmanned aerial vehicle to collect real-time video image data; the system carries out automatic identification on pictures, automatically completes equipment level alignment and standard point location identification, generates fine control of camera zooming, horizontal rotation, pitching and the like, and realizes standardized view finding and acquisition. By combining an enhanced recognition algorithm, traditional image processing and a deep learning model are fused, model self-evolution is realized through incremental learning, flexible expansion and collaboration of various patrol terminals are supported through a unified interface, and refined, real-time and intelligent patrol operation and maintenance requirements of an intelligent substation are met.

The invention discloses a multi-view child motion coordination ability evaluation system and method based on deep learning, belongs to the technical field of motion evaluation, and solves the problems that child motion evaluation in the prior art mainly depends on manual observation and simple physical testing, and multiple angles and key details of child actions are difficult to synchronously track. The method comprises the following steps: training to obtain a skeleton point detection model and an athletic ability evaluation model, acquiring real-time video streams of personnel entering a field based on an acquisition camera, identifying skeleton key points in a preprocessing data set by the skeleton point detection model, performing multi-person detection on a personnel matching result based on an athletic area division method, and evaluating the athletic ability of the personnel. The exercise ability evaluation model carries out quantitative analysis on multi-person detection results; according to the invention, visual identification and motion state detection technologies are combined, a front view angle and side view angle dual-camera layout is adopted, and a deep learning algorithm is matched, so that automatic children dynamic motion evaluation is realized. The action process can be completely captured, and the detection accuracy and efficiency are improved.

The invention discloses a video stream adaptive low-delay real-time transmission method and system based on edge calculation. The method comprises the following steps: receiving a real-time video stream from a network camera, creating a pipeline queue, adding timestamp information for each video frame, and setting a queue protection mechanism; the coded video frames are taken out from the input queue, and the frames in the video are processed through hardware acceleration decoding; the resource use condition of the system is monitored in real time; executing a self-adaptive frame skipping decision according to a performance monitoring result; timestamp generation: dynamically calculating a timestamp interval according to an actual processing frame rate; receiving the decoded original video frame and the corresponding timestamp information, accelerating decoding by using hardware, and executing a video coding operation; and packaging and transmitting the coded video data, and providing a standard protocol interface to be connected with a client for playing. According to the scheme, stable low delay and relatively low resource occupation can be kept, and meanwhile, the video quality is remarkably improved.

The invention relates to the technical field of image processing, in particular to an unmanned aerial vehicle positioning method and system based on machine vision, and the method comprises the steps: obtaining a current frame image and a reference image in a real-time video stream of an unmanned aerial vehicle, generating an initial matching pair set, and calculating the structural consistency of each matching pair, the method comprises the steps of adaptively determining a screening threshold value of a current frame image based on structural consistency, determining a screening matching pair set by utilizing the screening threshold value, evaluating a positioning contribution weight of each matching pair of the screening matching pair set, executing weighted pose calculation based on the positioning contribution weights, and obtaining an instantaneous pose of an unmanned aerial vehicle in the current frame image. And inputting the instantaneous pose as an observation value into a time sequence filtering model, performing time sequence fusion in combination with a motion model of the unmanned aerial vehicle, and outputting the final pose estimation of the unmanned aerial vehicle in the current frame image so as to complete the accurate positioning of the unmanned aerial vehicle. The method improves the accuracy of unmanned aerial vehicle positioning.

The invention provides a safety production behavior monitoring method and system based on AI video analysis. The method comprises the steps of collecting a real-time video data stream of a production area; inputting each frame of video image in the real-time video data stream into a target detection model for target detection to obtain a personnel target output by the target detection model and a target position coordinate of the personnel target in each frame of video image; cutting out a local image area of the personnel target in each frame of video image based on the target position coordinate, and performing feature recognition based on the local image area to obtain personnel features; performing comparison on the basis of the personnel characteristics and the personnel standard behavior characteristics to obtain personnel behavior states, and performing track association on the basis of the personnel behavior states corresponding to the continuous multi-frame video images to obtain personnel behavior tracks; and performing safety production behavior monitoring based on the personnel behavior state and the personnel behavior track, and generating an abnormal behavior early warning signal. According to the method and the device, the real-time performance and the accuracy of safety monitoring in a production scene are improved.

The invention relates to the technical field of photovoltaic power station operation and maintenance, in particular to a photovoltaic power station intelligent inspection system based on AI vision, which comprises a video acquisition module, a geometric reference construction module, a tremor offset resolving module, a coordinate inverse correction module and a defect fine calibration module, the video acquisition module is used for acquiring a real-time video stream of an unmanned aerial vehicle polling photovoltaic array, and performing time-space synchronization calibration on the video stream to generate an original image sequence. According to the invention, the computer vision technology is utilized to calculate a current frame blanking point in a video picture as a tiny offset of a reference object relative to a reference position in real time, the displacement is deducted from a GPS coordinate, and a coordinate inverse correction module is utilized to inversely calculate a visual axis offset into a ground projection error. The shake amount and the shake direction of the camera at each moment can be accurately calculated, and then the GPS coordinates are corrected in turn, so that the geographic accuracy of defect positioning is improved, and the operation and maintenance personnel can accurately find a fault component.

The invention provides a real-time video image compression method based on deep learning, and relates to the technical field of video image compression, and the method comprises the steps: carrying out the key feature recognition through employing an attention mechanism; performing convolution training optimization on the video image sample data set by using a deep learning network structure; a self-encoder structure is designed to carry out feature map encoding compression; a video image compression adaptive network is generated through series fusion; a real-time video image frame is collected for preprocessing, and feature compression processing is performed on a standard video image frame based on a video image compression adaptive network. According to the method and the device, the technical problem that the video compression quality is reduced due to the fact that the generalization ability is insufficient in the face of various scenes and the video compression strategy is difficult to adaptively adjust according to different scenes in the prior art can be solved, the adaptive network is constructed through the combination of deep learning and the auto-encoder, and the video compression quality is improved. And the video compression strategy is dynamically adjusted according to the contents of different video images, so that the video compression quality is improved.

A computer-vision system or engine that (a) generates from a pixel stream a digital representation of a person and (b) determines attributes or characteristics of the person from that digital representation and (c) based on those attributes or characteristics, outputs data to a cloud-based analytics system that enables that analytics system to identify and also to authenticate the person. The attributes or characteristics of the person include their pose, and the system or engine analyses that pose to extract a facial image from a video stream that is the best facial image for use by the cloud-based analytics system to identify and authenticate the person. The computer-vision system or engine outputs the facial image to the cloud-based analytics system, but does not output the full-frame real-time video to the cloud-based analytics system.

The invention relates to the technical field of parcel sorting, in particular to a parcel detecting and sorting method and system based on machine vision. According to the method, historical parcel motion trajectory data and real-time video information are fused, a category and weight differentiation trajectory model is constructed, sudden offset is captured in combination with computer vision, the trajectory is accurately predicted, and the attitude offset is obtained. The initial fixed parameter of the actuator is dynamically adjusted to be a first action parameter based on the offset, a compensation coefficient is generated through trajectory tracking error verification, and a second action parameter is obtained; meanwhile, friction coefficient changes are calculated through the real-time track and pressure data, a dynamic response coefficient is determined, and finally a multi-sorting-port cooperation instruction is generated to achieve path dynamic distribution. According to the process, through historical and real-time data fusion, dynamic parameter adjustment, closed-loop feedback and multi-system cooperation, the problems of traditional sorting static presetting, parameter blind adjustment, resource allocation imbalance and the like are solved, and the sorting precision, adaptability and efficiency are improved.

The invention belongs to the technical field of computer vision and deep learning, and relates to a cultured fish intelligent feeding system based on multi-feature fusion and a control method thereof, and the system comprises an image acquisition module, a space-motion feature extraction module, a motion track feature extraction module, a feeding state recognition module, a feeding control module and a feedback adjustment module. Extracting a depth optical flow image reflecting fish school feeding state characteristics through a fish school real-time video stream; classifying the fish school feeding state image samples to form a space-motion optical flow feature map data set; meanwhile, fish swimming trails are extracted, and a swimming trail feature map data set is formed; and training is carried out to obtain an optimal model weight, real-time identification is realized, an identification result is transmitted to the feeding control module, the feeding state of a fish school after feeding is completed is evaluated, the feeding amount is adjusted, or a feeding decision is optimized. According to the invention, real-time, accurate and efficient identification and decision control of the feeding state of the fish school can be realized, and intelligent feeding is realized on a low-cost edge end computing platform.

The invention relates to the technical field of wireless streaming media, in particular to a real-time video transmission method and system based on 5G, and the method comprises the following steps: based on mobile terminal equipment, analyzing the signal strength and service bearing change of a 5G base station, screening an optimal access sequence, comparing channel capability with video partition characteristics, and adjusting coding parameter distribution. And analyzing link capability and packet loss performance, and optimizing uplink and downlink compression parameters to obtain a compression change trend. According to the invention, by dynamically identifying the signal association and network load change between the mobile terminal and the 5G base station, the multi-dimensional access priority sequence is generated in combination with the real-time bandwidth demand, the coding and allocation parameters are flexibly adjusted according to the video partition content characteristics and the wireless channel adaptation condition, and the access priority sequence is optimized in combination with the uplink and downlink performance and the frame type change. Dynamic regulation and control and synchronous updating of compression parameters are automatically completed, continuous transmission of video streams in a complex network environment is achieved, and the end-to-end consistency of video content and the integrity of picture data are improved.

The invention provides a high-altitude operation emergency scheduling method and system, and the method comprises the steps: firstly obtaining a real-time video data set of a high-altitude operation region, carrying out the preprocessing of the real-time video data set, and generating a standardized video data flow, and then calling a video analysis model to extract an operation environment feature set and an operator behavior feature set; and then constructing a graph structure based on the working environment feature set, generating an environment state feature vector through graph convolutional network multi-level feature aggregation, and generating a dynamic emergency scheduling strategy in combination with the environment state feature vector and the working personnel behavior feature set. And finally, generating an emergency scheduling instruction set after strategy verification, and transmitting the emergency scheduling instruction set to the operation terminal equipment to trigger emergency response operation, so that a targeted emergency scheduling strategy can be dynamically generated according to actual conditions, automation and intelligence of emergency scheduling are realized, and the emergency response efficiency is improved.

The invention relates to the field of production safety management and artificial intelligence, and discloses an edge early warning method based on visual identification and semantic fusion, which is composed of the following modules: a data acquisition module, which is used for collecting field real-time videos and images, the visual identification module is constructed based on a PyTorch deep learning framework and is used for receiving the video and image information transmitted by the data acquisition module; and the semantic fusion module is used for receiving the target boundary position and the category label list output by the visual identification module and carrying out cross-model data fusion and semantic understanding. According to the method, two mechanisms of rule-based deterministic analysis and model-based anomaly detection are fused, the rule base can accurately identify typical violation and dangerous behaviors, the anomaly detection can capture unexpected anomaly modes, the limitation of a single method is effectively solved, missing detection and misjudgment are reduced, and the false alarm rate of the system is greatly reduced.

User authentication, spoofing and replay attack prevention, liveness detection, and user-and-document verification using a live video stream with spatial challenges. A camera of an electronic device captures and transmit a live selfie user-facing video, as part of a user registration process. The user is instructed to spatially move his body or face, such that his face would appear within a first particular on-screen shape; and to also, concurrently or simultaneously, spatially hold in his hand or move a particular an identification document such that it would appear within a second on-screen shape. Optionally, the on-screen shape moves on the screen, and the user is required to spatially move the relevant item to keep it within the boundaries of the moving on-screen shape. The system then analyzes the video via computerized vision, to determine whether the user complied with the spatial manipulation challenges.

The invention provides a facial skin flaw enhancement method based on a Lab color space. The method comprises the following steps: firstly, acquiring an RGB face image and converting the RGB face image into a CIE Lab color space with uniform perception; then, performing differentiation treatment according to the manually selected skin flaw type: for the vascular flaw, extracting statistical characteristics of a component and driving adaptive nonlinear transformation, and generating a grey-scale map which highlights the red flaw; for pigment flaws, nonlinear transformation is carried out on the component L, and then collaborative linear weighting and feature amplification are carried out on the component L, the component a and the component b, so that a grey-scale map with highlighted pigment spots is generated. And finally, coloring the grey-scale map in the Lab color space through adjustable parameters to generate a high-contrast color enhanced image. The method overcomes the dependence on hardware and training data in the prior art, can clearly and adaptively enhance various flaws such as acnes, couperose streaks and color spots, shows robustness under different illumination, and can be widely applied to clinical beauty, later photography and real-time video processing.

Smart dental treatment chair, and a system utilizing Artificial Intelligence (AI) and computerized vision analysis to dynamically monitor real-time progress of a dental treatment, to dynamically report the progress to the patient, and to provide additional benefits to dental patients. A dental treatment chair includes video cameras that capture real time video, and a microphone that captures sound and speech. Computerized vision unit perform analysis of the video, and a Large Language Model (LLM) performs analysis of text extracted from uttered speech, to determine the current step in a multiple-step dental procedure. A display unit is oriented towards the patient, and displays a dynamically-updated progress bar and percentage value, indicating the actual progress of the ongoing dental treatment. Optionally, the smart dental treatment chair also integrally plays music that the patient selects and controls, sprays an aromatic agent, and provides other benefits to the dental patient.

The invention relates to the technical field of medical image processing, in particular to an endoscopic surgery video real-time structure analysis method and system, and the method comprises the steps: carrying out the frame-by-frame semantic segmentation of real-time video data, and generating a segmentation mask of each frame of image; calculating a comprehensive quality score of the target frame based on the segmentation masks of the target frame and the previous frame of the target frame; analyzing a motion amount index between adjacent frames; generating the current length of a sliding time sequence window according to the comprehensive quality score and the exercise amount index, and fusing multiple frames of segmentation masks in the sliding time sequence window to generate a reference mask; extracting a key point from the reference mask, and obtaining a displacement vector of the key point from a previous frame of the target frame to the target frame; generating a pixel-level displacement field according to the displacement vector, deforming the segmentation mask of the previous frame of the target frame to the target frame, and generating a prediction mask of the target frame; and performing superposition display on the prediction mask and the image of the target frame. According to the scheme, the time sequence consistency and stability of the video semantic segmentation result can be enhanced.

A system and method for latent geodesic traversal across multi-axis hyperspaces for real-time video reconstruction and augmentation. Spatiotemporal video data are compressed into navigable latent representations using hierarchical and Lorentzian autoencoders that preserve geometric and temporal structure. A geodesic traversal engine computes paths across spatial, temporal, spectral, and semantic axes, guided by symbolic anchors and spatiotemporal routing protocols. A correlation network restores fine detail, while an augmentation generator synthesizes additional or counterfactual content to enable infinite zoom, continuous multi-scale exploration, and temporally coherent augmentation. A strategy caching system preserves successful traversal patterns for reuse, supporting persistent learning and adaptive real-time performance.

A computer implemented method, system, and non-transitory computer-readable device that may be used in a remote deposit environment. Upon receiving a user request, based on interactions with the UI, the method implements an electronic deposit of a financial instrument by activating a camera on the client device to generate a live video stream of image data of a field of view of at least one camera, wherein the live video stream includes imagery of at least a portion of each side of the financial instrument. The method continues by extracting data fields based on the formation of image objects on one or more sides of the financial instrument from the live video stream of image data. An EFT conversion of extracted data fields may be processed during or subsequent to the extraction process. A message is sent from a payee to a payor requesting the EFT. Upon acceptance, an EFT to the payee occurs. Upon denial, the remote deposit process is completed.

The invention relates to a human traffic monitoring method and system for a dense sitting posture scene, and the method comprises the steps: S1, obtaining the video stream data of the dense sitting posture scene, constructing an original data set, and optimizing the size distribution of an anchor frame; s2, constructing a human traffic monitoring model which is improved and realized on the basis of a YOLOv7 model: in a backbone network, replacing the four starting CBS modules with ODSConv modules, and replacing the two subsequent ELAN modules with E-ELAN-Sim modules; in the head network, a coordinate attention mechanism is embedded in an SPPCSPC module to obtain a CA-SPPCSPC module, the CA module is added in front of a CBS module connected with an ELAN-H module and a Concat module, and meanwhile, a four-scale detection framework is realized in the head network to enhance the small target detection capability; training the constructed people flow supervision model through the training data set; and S3, inputting real-time video stream data into the trained people flow supervision model, and outputting a people flow supervision quantity. According to the method and the system, accurate detection and real-time statistics of the sitting posture target person in the dense sitting posture scene can be realized.

The invention discloses an intelligent security and protection management system and method based on abnormal target tracking, and relates to the field of security and protection management, which analyzes the space-time accessibility of a target in an environment topology after the target is lost, models the dynamic evolution process of target characteristics, and compares the judgment reference of identity re-identification from the static global appearance, so that the identification accuracy of the target is improved. And transferring to a dynamic probabilistic spatio-temporal search model with perspectiveness. And furthermore, based on the model, actively predicting a time-space window where a target may appear and an attenuated appearance portrait, performing targeted search and matching in a specified real-time video stream, and realizing reliable re-association of the identity according to the confidence of a matching result. Therefore, two key variables, namely environment topology and time decay, are introduced into a tracking mechanism, so that active prediction of the space-time position and the appearance state of the missing target can be realized, and the continuity and the accuracy of identity tracking can be guaranteed under the complex condition that the target is separated from the visual field for a long time.

The invention relates to the technical field of unmanned aerial vehicle detection, and discloses a highway bridge hidden danger target identification and position determination method based on an unmanned aerial vehicle. The method comprises the following steps: acquiring a real-time video of a bridge by using an unmanned aerial vehicle carrying a multispectral sensor, and performing multi-scale segmentation on a current frame image to obtain a bridge surface and a suspected hidden danger target area; extracting a suspected hidden danger target area texture feature through a gray level co-occurrence matrix, morphological closed operation and spatial domain filtering, and inputting a hidden danger classification model to obtain type information; based on the bridge surface geometric structure model and the center coordinate of the suspected hidden danger target area, calculating the offset of the suspected hidden danger target area relative to the bridge datum line; and generating three-dimensional space coordinates in combination with the offset and the type information, and marking the three-dimensional space coordinates in a bridge structure chart. According to the method, multi-spectral sensing, multi-scale image processing and three-dimensional modeling technologies are fused, accurate recognition and three-dimensional positioning of bridge hidden dangers are achieved, hidden danger information can be dynamically updated, and the detection efficiency and precision are improved.

The invention provides a subway platform edge behavior abnormity real-time identification and early warning method based on deep learning, and relates to the technical field of deep learning, which comprises the steps of obtaining real-time video data, performing pedestrian detection and track extraction, calculating behavior characteristic parameters and a dynamic distance threshold value, performing danger grade division on pedestrians, and performing early warning and early warning. According to the method, time sequence feature extraction is carried out on potential dangerous targets, a behavior abnormity scoring mechanism is established, timely early warning of abnormal behaviors is realized, the accuracy and real-time performance of safety monitoring of the subway platform can be effectively improved, and the occurrence rate of safety accidents can be reduced.

The invention discloses a real-time video abnormal behavior detection method based on multi-mode collaborative learning. The method comprises the following steps: S1, collecting video, audio and environment data through a camera, a microphone and an environment sensor; s2, the collected multi-modal data is preprocessed; s3, an improved multi-modal variational auto-encoder is adopted to generate a multi-modal feature vector; s4, inputting the multi-modal feature vector into a multi-modal diffusion model to generate a multi-modal fusion feature representation; s5, dynamically adjusting feature weight distribution based on a zebra optimization algorithm, and optimizing de-noising parameters and step size setting; s6, outputting an abnormal behavior detection result and a confidence score, and triggering an alarm module to give an alarm in real time; and S7, dynamically adjusting model parameters, and uploading abnormal data to a cloud platform. Through the multi-mode collaborative learning method, high-precision detection and quick response of real-time video abnormal behaviors in a complex dynamic environment are realized, and the safety and stability of an intelligent monitoring system are effectively improved.

The invention discloses a power transmission line defect detection method based on an improved Yolovlln model. The method comprises the following steps: firstly, inputting a defect image training data set into an improved Yolovlln network model for training to obtain a target detection model; the camera collects real-time video image information and sends the real-time video image information to the defect detection unit; and finally, obtaining a defect category and position identification result by a target detection module in the defect detection unit. The improvement method of the Yolovlln is as follows: a loss function CIoU is replaced by ShapeIoU; an optimizer SGD is replaced by AdamW; the method comprises the following steps: replacing a C3k2 module with a C3k2-H module for a structure in a backbone network; a PSA attention mechanism in the C2PSA is replaced with a BRA attention mechanism; an original Yolo detection head is replaced by a task dynamic alignment detection head, so that the recognition capability of the model on a multi-scale target is improved. The defect detection method adopted by the invention can accurately and efficiently detect various abnormal defects in the power transmission line, and has relatively high practicability and application prospect.

Disclosed is a computer-implemented method and system for training a subject-specific machine learning model to infer inherent subject features from recorded or live video data. The system preprocesses the visual and audio channels, converting audio to text, and employs multiple pre-trained extraction models to generate feature embeddings. Ground truth data is obtained to guide training, where weights are assigned to produce and combine predicted feature values. Model performance is optimized by minimizing error. The trained feature extraction models are deployed on an edge device, while the subject-specific model resides in the cloud. A lightweight edge model, derived via knowledge distillation and model compression, supports local inferencing with reduced reliance on cloud resources. Synchronization ensures iterative updates for sustained accuracy.