634 results about "Action recognition" patented technology

The invention discloses a motion recognition method and system based on a redox photoelectric memristor, a terminal and a storage medium, and the method comprises the steps: selecting classical motions based on a human body motion data set, extracting time sequence data, and coding the time sequence data into an optical pulse sequence; constructing a reservoir array composed of a plurality of photoelectric memristors, and expanding an optical pulse sequence signal into a high-dimensional state vector; constructing a supervised training model, solving a weight matrix, and constructing a memristive cross array; and outputting an action classification result through simulation domain operation based on the output current multi-path light current signals in combination with the memristor cross array. According to the method, the motion features can be directly fed into a rear-end classification network for action recognition without depending on a complex digital feature extraction algorithm, so that the transmission and processing overhead of redundant data is fundamentally eliminated; a high-efficiency, low-delay and high-robustness hardware solution is provided for real-time and anti-noise motion recognition in scenes such as intelligent monitoring and man-machine interaction.

The invention relates to a method, a system and equipment for intelligently identifying unsafe behaviors of coal mine operating personnel and a medium. The method comprises the following steps: extracting individual trajectory data from an underground video stream through a multi-target tracking algorithm, decomposing continuous actions into an atomic behavior sequence with a space-time mark by using attitude estimation and a space-time diagram convolutional network, and capturing space-time relevance of a behavior chain through an attention mechanism enhanced long and short term memory network to generate a feature vector; risk reasoning is carried out in combination with the coal mine safety knowledge graph to predict the risk event type and probability, and finally early warning information is generated based on a multi-level early warning strategy. According to the scheme, the crossing from isolated action recognition to behavior chain risk prediction is realized, and the early warning capability of potential safety risks in coal mine operation is remarkably improved through fusion of space-time correlation analysis and domain knowledge of the behavior sequence.

The invention relates to the technical field of action recognition, in particular to a competitive sports-oriented skeleton trajectory deep learning compensation method and system, and the system comprises a data preprocessing module, a data integrity evaluation module, a core compensation network, a biomechanical rationality optimization module, a personalized habit encoder and an advanced compensation path. Compared with the prior art that a single trajectory compensation model is generally adopted, challenges of shielding scenes of different degrees are difficult to effectively deal with, and the problem of insufficient short-time shielding compensation precision or trajectory distortion under long-time shielding often occurs, an intelligent routing mechanism based on data integrity evaluation is adopted, and the method has the advantages that the complexity is reduced, and the reliability is improved. Through dynamic switching of a basic compensation path and an advanced compensation path fused with personalized prediction, adaptive processing of different shielding scenes is realized, reconstruction precision under short-time shielding is ensured, track rationality and continuity under a long-time shielding scene are remarkably improved through motion trend prediction, and the reconstruction precision is improved. And the practicability and the reliability of the system in an actual competitive environment are enhanced.

The invention discloses an intelligent wheelchair control method based on multi-source information fusion, and belongs to the field of data recognition and wheelchair intelligent control. The implementation method comprises the following steps: constructing a corresponding relation between gesture actions and wheelchair control directions, and presetting an identification framework; processing the obtained multi-source information to obtain a basic probability distribution result of the evidence corresponding to each single-source information; determining the absolute importance of the evidence corresponding to each piece of single-source information in the fusion process; determining the relative importance of the evidence corresponding to each piece of single-source information in the fusion process; determining the comprehensive importance of the evidence according to the obtained relative importance and absolute importance; according to the comprehensive importance of the evidence corresponding to the single-source information, redistributing a fusion weight for the evidence corresponding to each piece of single-source information to obtain a weighted new evidence; dS evidence fusion is applied, a final fused gesture action recognition result is obtained, the behavior of the wheelchair is controlled according to the gesture recognition result, and intelligent wheelchair control based on multi-source information fusion is achieved.

The invention relates to a video action recognition model training method, a video action recognition method and a video action recognition device. The method comprises the following steps: acquiring a sample video frame image and an action category description text corresponding to the sample video frame image, inputting the sample video frame image and the action category description text into a to-be-trained recognition model, and recognizing the action category of the to-be-trained recognition model by an image encoder in the recognition model according to a preset visual prompt vector and the sample video frame image, generating a video embedding corresponding to a sample video frame image, generating a text embedding corresponding to an action category description text by a text encoder in the recognition model based on a preset text prompt vector and the action category description text, and constructing bidirectional comparison loss by taking the video embedding and the text embedding as positive sample pairs, and updating the visual prompt vector and the text prompt vector based on the bidirectional contrast loss to obtain a trained recognition model. By adopting the method, the video action recognition accuracy can be improved.

The invention provides a continuous action recognition method based on a spatial-temporal characteristic dynamic attention fusion network. The method comprises the following steps: constructing an action recognition model; the method comprises the following steps: acquiring an action signal of a to-be-measured target through a millimeter-wave radar, and preprocessing the action signal; constructing a continuous point cloud sequence according to the preprocessed action signal; merging the continuous point cloud sequences; the merged point cloud is input into the trained action recognition model, and a time sequence feature extraction module is adopted to extract a point cloud sequence to carry out adaptive multi-scale time sequence features; performing multi-scale neighborhood aggregation on each frame of point cloud by adopting a spatial feature extraction module; performing complementary fusion on the self-adaptive multi-scale time sequence features and the multi-scale spatial features by adopting a spatio-temporal feature interactive attention fusion module to obtain fusion features; inputting the fusion features into a classifier to obtain a continuous action recognition result; the multi-scale graph convolutional network structure with the adaptive domain weight distribution strategy designed by the invention dynamically adjusts the contribution weight of neighborhood points to feature extraction according to the local density of the point clouds and the spatial geometrical relationship aiming at the characteristics that the millimeter wave radar point clouds are irregularly distributed and the spatiality is difficult to excavate; the problems of coarse granularity and poor robustness of spatial feature extraction of a traditional method are solved.

The invention relates to the field of computer vision, artificial intelligence and intelligent fitness, and particularly discloses a fitness action recognition, counting and quality evaluation method based on machine vision, which comprises the following steps of: extracting video frames and standardizing the video frames, realizing background suppression and human body region enhancement through a semantic segmentation or inter-frame difference method, and outputting a standardized image sequence; detecting key joint points by adopting a pre-training model, and outputting a stable skeleton sequence and candidate action stage data through time sequence consistency filtering; reconstructing a feature tensor, fusing spatio-temporal features through double-branch attention collaboration, and outputting action categories and time sequence compensation parameters in a classified manner; a dynamic threshold method accumulates the number of actions, action qualification is judged in combination with biomechanical constraints, the confidence coefficient is optimized, and a structured result is output; bone rendering, error highlighting, voice generation and personalized training suggestions. According to the method, wearable equipment is not needed, the problems of instable identification, miscounting and the like in a complex scene are solved, and real-time accurate analysis is realized.

The invention provides a method and device for processing sensitive information in a video, electronic equipment and a storage medium, and relates to the technical field of computer vision, and the method comprises the steps: extracting a voice segment in an audio signal, and determining a transliteration text of the voice segment and a phoneme timestamp of the transliteration text; performing semantic analysis on the transliteration text to obtain first sensitive information in the transliteration text, determining a sensitive voice segment corresponding to the first sensitive information based on the phoneme timestamp, and erasing the sensitive voice segment; and finally, sensitive area identification is carried out on each video frame to obtain second sensitive information in each video frame, and the second sensitive information is erased. According to the method, the sensitive information in the video can be automatically and accurately positioned and efficiently purified, the processed video is ensured to meet the requirements of national laws and regulations, the availability of the video in tasks such as action recognition and cross-modal learning can be remarkably improved, and a pure data basis is provided for high-quality visual model training.

The invention belongs to the technical field of exercise training and analysis, and particularly relates to an intelligent badminton action analysis system and method based on a motion sensor, and the system comprises a sensor data collection module, a real-time data processing module, an action recognition analysis engine, an intelligent data storage module and a visualization module. The method comprises the following steps: receiving and buffering sensor data in real time; periodically extracting data segments by using a sliding window to carry out multi-feature (an acceleration peak value, an angular velocity peak value, a peak value ratio and the like) calculation; based on a predefined multi-feature threshold rule set, identifying action types (such as kicking, flat ball drawing and the like) and evaluating the quality; an intelligent strategy is adopted to selectively store data so as to optimize the system load; and finally, a result is visually fed back in real time. The method realizes objective, quantitative and real-time analysis of badminton actions, has the advantages of high recognition accuracy, high system real-time performance, efficient data processing and the like, and is suitable for autonomous training of athletes and auxiliary guidance of coaches.

The invention provides an action recognition method based on adaptive skeleton grouping and direction sensitive space-time modeling, and relates to the technical field of skeleton action recognition. Comprising the steps of skeleton action sequence input and feature embedding, adaptive skeleton grouping and direction weighting space-time modeling, trunk modeling and feature refinement, adaptive time down-sampling, multi-stream feature fusion and classification output and action recognition loss evaluation. The method comprises the following steps: acquiring an introduction result through skeleton action sequence input and feature embedding, acquiring an adjustment result through adaptive skeleton grouping and direction weighted space-time modeling, acquiring a processing result through trunk modeling and feature refinement, acquiring a reconstruction result through adaptive time downsampling, performing multi-stream feature fusion and classified output, and finally executing action recognition loss evaluation. According to the method, the skeleton action recognition precision is improved, and meanwhile, the complexity of long sequence modeling calculation is reduced, so that the skeleton action recognition real-time performance and deployment efficiency are improved, and the problem that the skeleton action recognition precision is not high in the prior art is solved.

The invention discloses a table tennis motion detection system based on pure visual perception. The table tennis motion detection system comprises a table tennis detection and event monitoring module, a human body-racket posture estimation module and a time sequence action recognition and quality evaluation module. The table tennis ball detection and event monitoring module is based on an improved DTTNet model and realizes table tennis ball track detection and event identification such as table touching, net touching and net passing through through multi-frame video stack and deformable convolution; the human body-racket posture estimation module adopts a lightweight detection network and high-resolution posture estimation network combined reasoning structure, outputs human body key point and racket three key point coordinates, and constructs a human body-racket combined kinematics model; the time sequence action recognition and quality evaluation module fuses multi-view video and posture features, classification and quality evaluation of table tennis technical actions are achieved through time sequence boundary matching and multi-scale feature analysis, the table tennis technical action recognition and quality evaluation system has the advantages of being high in detection precision and high in real-time performance, and automatic recognition and quantitative evaluation of table tennis movement can be achieved.

The invention discloses a three-dimensional reconstruction analysis method based on an artistic gymnastics motion trajectory, and relates to the technical field of motion recognition, and the three-dimensional reconstruction analysis method based on the artistic gymnastics motion trajectory comprises the following steps: S1, obtaining a target and parameters, and collecting a real-time image; s2, setting a personalized body display target, and constructing a virtual person three-dimensional motion trajectory model; s3, performing time sequence segmentation to obtain a virtual stage action track set, and extracting virtual action feature parameters; s4, time sequence segmentation is carried out, action track parameters are extracted, and a real-time action track parameter set is obtained; s5, generating a real-time three-dimensional model of the artistic gymnastics S6, performing time sequence segmentation to obtain a real-time stage action track set, and extracting real-time action characteristic parameters; and S7, performing comparative analysis, and outputting an evaluation report. According to the invention, through combination of a multi-view synchronous camera technology and a three-dimensional grid modeling algorithm, the accuracy of action analysis and evaluation is improved.

The invention discloses a 3D human body action recognition method based on geometry-contour multi-frequency, relates to the technical field of computer vision and behavior recognition, and provides an end-to-end learning framework from point cloud sequence input to action recognition output. The framework innovatively fuses two core processes of geometry-contour space structure modeling and multi-frequency time decomposition modeling: the geometry-contour space structure modeling accurately depicts macroscopic contour postures and microscopic geometry details of a human body to comprehensively capture motion information; the latter decomposes the time attitude sequence into a low-frequency trend flow and a high-frequency detail flow, and digs action mode rules under different frequencies; and through cooperative work of the two, the accuracy of point cloud action recognition and the robustness in a complex scene are remarkably improved.

The invention discloses an automatic classroom director method and electronic equipment. Audio data and video data of a current classroom scene are acquired; based on the audio data, sound source direction positioning is carried out through a microphone array, and the current sound source direction is determined; face key points are extracted according to the video data, lip movement recognition is carried out according to the face key points, and a lip movement recognition result is obtained; extracting human body key points according to the video data, and performing action recognition according to the human body key points to obtain a target action recognition result; based on the sound source direction, the lip movement recognition result and the target action recognition result, a current shooting picture of the camera device is controlled, and the current shooting picture of the camera device is a director picture; and outputting and displaying the director picture. According to the application, audio and video multi-mode information is fused, high-precision identification and natural and accurate automatic picture switching of the speaker and the interaction object in the classroom scene are realized, and the automation level of director and the overall classroom recording and broadcasting effect are improved.

The invention relates to the technical field of railway maintenance education, in particular to a high-speed railway four-electricity system intelligent teaching system and method based on AI, and the method comprises the steps: a teaching data obtaining module collects teaching process data, student learning tracks and industry fault data; the virtual simulation modeling module is fused with a BIM model and a technical specification text to construct a four-electric system twinborn body; the AI intelligent analysis engine constructs student ability portraits through a machine learning algorithm, generates a personalized learning path and intelligently answers questions; the virtual-real teaching interaction module builds a digital-intelligent practical training scene, carries out three-dimensional visual error correction and remote guidance through real-time motion recognition, and outputs practical training data. The cross-professional fault simulation module carries out cross-professional fault linkage simulation deduction through an association rule mining algorithm and outputs a deduction result; and the teaching effect evaluation module establishes a binary evaluation system combining process and practical operation skill assessment, and outputs teaching effect data. Therefore, the problems of fixed teaching strategy, low precision and the like in the prior art are solved.

The invention discloses a ViT model lightweight method for video action recognition, which comprises the following steps of: performing motion estimation in a data processing stage, calculating to obtain motion intensity and time redundancy scores of tokens, and performing window division on video frames according to the scores. Afterwards, a window-global token merging strategy is adopted to effectively merge redundant tokens, the first three layers of the model process local information through adaptive window merging, and from the fourth layer, the model repeatedly executes global merging according to the depth of the layers so as to realize gradually enhanced feature representation; therefore, the number of tokens is reduced, the complexity of the ViT model in space-time self-attention calculation is reduced, and the problem of high calculation resource consumption caused by space-time information redundancy of the current ViT model in a video task is solved. Calculation overhead and energy consumption can be remarkably reduced in tasks such as video action recognition, action detection and time sequence event understanding, and meanwhile the characterization capacity of a key dynamic area is kept.

The invention provides a gesture action recognition method, electronic equipment and a storage medium, and the method comprises the steps: collecting a continuous gesture image sequence, carrying out the track space-time diagram structure construction of image frames in the continuous gesture image sequence, obtaining a track space-time diagram structure containing track node features and edge correlation features, and carrying out the track evolution feature analysis, identifying a change mode of track node features along with time and a dynamic adjustment rule of edge correlation features, and generating a track evolution feature sequence; inputting the track evolution feature sequence into a preset gesture semantic analysis model, and generating a gesture semantic feature vector representing the gesture action intention through time sequence correlation modeling and semantic mapping processing; and performing similarity comparison on the basis of the gesture semantic feature vector and a pre-constructed gesture category feature library, determining a gesture action category corresponding to the continuous gesture image sequence, and outputting a gesture action recognition result. According to the invention, the gesture action type corresponding to the continuous gesture image sequence can be reliably identified.

The invention discloses a children autism risk assessment method and system based on multi-modal collaborative reasoning and a vertical domain large model, and the method comprises the steps: collecting the audio data of a to-be-assessed child and a parent in a parent-child interaction link, and collecting the video data through collection equipment; analyzing and marking the collected audio data, and generating a sounding record and a transcription text with a timestamp; performing real-time analysis and labeling on the collected video data, and generating sight line, expression and action recognition records with timestamps; performing independent analysis on the preprocessed multi-modal data, performing feature fusion and collaborative analysis on the multi-modal data such as audio, text and video by using a multi-modal collaborative reasoning method, and generating a scientific data index for autism risk assessment; training a vertical domain large model special for autism risk assessment; and inputting the generated data indexes into a vertical domain large model, and outputting targeted assessment suggestions by the large model in combination with daily information of the children to be assessed obtained by interviews and conversations with parents.

The invention discloses a personnel state analysis method and system based on limb movement recognition, relates to the technical field of limb movement recognition, and determines the movement state of a corresponding limb part based on recognition of each movement-expression combination. And the driving response state of the personnel is determined based on the action state of each limb part, the corresponding part priority and the driving state of the truck, so that the accuracy of the driving response state of the personnel is improved. Therefore, the change event of the facial expressions is determined according to the plurality of facial expressions at different time, and the driving fatigue state of the personnel is determined according to the change event of the facial expressions and the lane changing frequency of the truck; according to the method, the voice interaction event of the person in the driving process is collected, the multiple state key contents are determined according to recognition of the voice interaction event, the state analysis system of the person is determined according to the multiple state key contents, the driving response state of the person and the driving fatigue state, and the accuracy of the state analysis system of the person is improved.

The embodiment of the invention discloses an action processing method, device, system and equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring point cloud data corresponding to each target object in a target space; recognizing the action of the target object based on the point cloud data to obtain the action corresponding to the target object; if the action is the target action, indicating the movable equipment to move to an area corresponding to the target object, and obtaining feedback information obtained by interaction between the movable equipment and the target object; and processing an event corresponding to the target action based on the target action and the feedback information. According to the method, action recognition is carried out by acquiring the point cloud data of the target object, a user does not need to wear any equipment, user privacy leakage is avoided while the action recognition accuracy is improved, meanwhile, the mobile equipment is linked to interact with the user, the event is processed according to the feedback information obtained through interaction, and the user experience is improved. And the action processing efficiency and the event processing capability are improved.

The invention relates to the technical field of action recognition, in particular to an AI-based motion posture recognition system, which comprises a key point extraction module, a graph structure construction module, a node trajectory modeling module, an action segmentation analysis module and a posture recognition output module. According to the method, the spatial position of each node of the human body in the continuous image under the time sequence structure is extracted, and the graph structure sequence based on the topological relation between the nodes and the motion trend is established, so that the dynamic modeling and the structure expression of the skeleton motion can be realized, and the trajectory characteristics of node position change, speed fluctuation, direction change and the like are analyzed; according to the method, key paragraphs with inversion and abnormity in actions are effectively identified, action segmentation classification is carried out in combination with frame sequence spacing and key node activity states, accurate analysis of complex action processes is realized under the condition of not depending on external hardware, and the accuracy and timeliness of posture identification are improved in a mode of matching with a standard template.

The invention relates to the technical field of new energy equipment operation and maintenance, in particular to a new energy operation and maintenance field operation behavior intelligent identification and compliance monitoring system and method, and the system comprises a multi-source sensing layer, an edge intelligent layer, a compliance reasoning layer, a command linkage layer and an evidence and auditing layer. The multi-source sensing layer adopts a multi-source acquisition device to monitor operation and maintenance field personnel, tools and environment states, and acquires and fuses multi-source data; the edge intelligent layer performs space-time alignment on the fused data, performs action recognition and interactive behavior modeling analysis, and outputs a result; the compliance reasoning layer performs compliance judgment based on the result to form a risk quantification result; the command linkage layer performs measures according to the risk quantification result and the risk level grading response; and the evidence and auditing layer forms an evidence chain and a compliance auditing log which cannot be tampered according to the evidence and auditing layer so as to realize evidence storage traceability.

The invention discloses a gesture recognition control method and system for a range hood and the range hood. According to the gesture recognition control method for the range hood, two sets of infrared geminate transistor sensors are controlled to emit infrared emission signals with preset coding values in a synchronous period; acquiring the signal intensity, the signal coding matching degree and the receiving time interval of two groups of corresponding infrared receiving signals after the two groups of infrared sending signals are reflected, judging whether an effective gesture judgment condition is met or not based on the signal intensity, the signal coding matching degree and the receiving time interval, and when the effective gesture judgment condition is met, executing the step 2; if yes, the gestures currently detected by the two sets of infrared geminate transistor sensors are determined to be effective gestures, and the range hood is controlled to execute control actions corresponding to the control instructions based on the control instructions corresponding to the effective gestures; coding verification, intensity verification and sequential logic-based dual-channel correlation judgment are carried out on the two groups of infrared receiving signals, so that efficient, reliable and high-interference-resistance non-contact waving action recognition can be ensured.

The invention belongs to a biological signal identification and data processing technology, and relates to a multi-sensor fusion action identification system for shoulder rehabilitation evaluation. Comprising a flexible strain sensor, an inertial measurement unit, a multi-channel data acquisition module, a data preprocessing module and an action recognition module, the flexible strain sensor comprises a plurality of sensing areas and connecting areas which are distributed around the rotator cuff area; the multichannel data acquisition module acquires multichannel sensing signals, and the multichannel sensing signals are preprocessed by the data preprocessing module. The action recognition module adopts a multi-modal feature fusion algorithm based on a bidirectional cross-modal interactive attention and time period specificity adaptive weight mechanism to realize dynamic correlation modeling between a strain sensing signal and inertial measurement data in a unified time sequence space, and cross-modal time sequence features are fused; capturing the global and local dependency relationship of the shoulder movement, and recognizing various rehabilitation actions. The method effectively overcomes the static limitation of the prior art, and remarkably improves the precision and robustness of shoulder rehabilitation evaluation.

The invention relates to the technical field of equipment operation tests, particularly discloses a simple scene equipment operation action rapid processing method based on a test video, and is used for solving the problems of low action recognition precision, insufficient spatial-temporal feature fusion, low detection speed and more redundant data in the prior art. Comprising the following steps: accurately intercepting motion start-stop frames, extracting the frames according to a preset frame rate, implementing geometric transformation and color augmentation on the frame-extracted images, respectively extracting spatio-temporal dynamic features and static spatial features by adopting a 3D convolutional network and a 2D convolutional network in parallel, integrating the spatio-temporal features by a channel fusion attention module based on a Gram matrix, and extracting the spatial and temporal dynamic features and the static spatial features by adopting the channel fusion attention module based on the Gram matrix. A thermodynamic diagram is generated based on Grad-CAM to assist in key area positioning, a RefineDet network is input to generate a detection result through ARM coarse regression and negative anchor filtering, ODM fine regression and non-maximum suppression in sequence, and efficient and accurate positioning and classification of operation actions are carried out with assistance of a dynamic frame abandoning and time sequence correction strategy; according to the invention, through the multi-scale residual error convolution and the anchor frame network, accurate positioning of slow action of equipment is realized.

The invention provides an intelligent video monitoring early warning method and system based on a multi-mode visual model, and the method comprises the steps: introducing a low-rank decomposition attention mechanism and an improved attention mechanism into an improved YOLOv8 algorithm, so as to obtain a detection model; detecting a specific target in the video based on the detection model; a ByteTrack algorithm is adopted to track a specific target, and an image of the specific target is cut out; performing short-time-sequence action recognition on the image by adopting a SlowFast algorithm; and a Qwen-VL multi-mode reasoning model is adopted to analyze the long-time-sequence content of the video, and the analyzed long-time-sequence content is combined with the RAG knowledge base to realize dynamic anomaly judgment. According to the method, the purpose of effective and continuous tracking can be achieved, short-time-sequence action recognition and long-time-sequence content analysis can be carried out, and dynamic anomaly judgment can be effectively achieved.

The invention provides a piece counting method applied to a sewing machine, and the method comprises the steps: determining the movement distance of a sewn object when a position sensor detects the movement of the sewn object; if the moving distance of the sewing object meets a moving distance threshold value, acquiring a user hand action image recorded by an image sensor within a set time after the sewing object moves, and analyzing the user hand action image to obtain a hand action sequence; if it is confirmed that the hand action sequence contains the piece counting symbolic actions or the number of times of the hand actions contained in the hand action sequence is lower than the change threshold value, piece counting accumulation operation of the sewn object is executed. According to the method, the distance sensor is combined with the hand action recognition for automatic piece counting, manual piece counting is not needed, and the piece counting accuracy can be effectively improved. The invention further provides a piece counting system applied to the sewing machine, a computer readable storage medium and electronic equipment which have the above beneficial effects.

This application relates to the field of computer vision technology, and in particular to a computer vision-based human action recognition method and system. The method includes real-time acquisition of user action videos, decomposing them into continuous video frames; deploying sensors on key human feature points to acquire angle data of all key human feature points in each video frame, including roll angle, yaw angle, pitch angle, angular velocity, and angular acceleration; performing preliminary correction on the image of each video frame based on the roll angle, yaw angle, and pitch angle from the angle data of all key human feature points in each video frame to generate a first corrected image; performing secondary correction on the first corrected image based on the angular velocity and angular acceleration from the angle data of all key human feature points in each video frame to generate a second corrected image; performing action recognition on the second corrected image, and sequentially combining the actions of single frames to output the user's complete action. This application effectively solves the problem of decreased recognition accuracy caused by user posture deviations and dynamic interference during movement, improving the stability and robustness of action recognition.

The application discloses a CSI-based position-independent human activity continuous learning recognition method, which comprises the following steps: 1, collecting CSI action sample data; 2, pre-processing the CSI action sample data; 3, constructing positive samples by randomly scaling the pre-processed samples in the time dimension; 4, constructing a multivariate time graph neural network and extracting CSI action sample features; 5, calculating the similarity between the sample feature values and the positive samples and the feature values of the remaining samples, obtaining a comparison loss, and optimizing the feature extraction network; 6, freezing the feature extraction network, sending the features obtained from the input samples into a classifier for training to obtain a classification model. When the application continuously learns new action categories, the user does not need to retrain the feature extraction network, and the new and old action recognition in any position in the room can be realized by providing limited position new category samples to train the classifier, and the practicability is relatively high.

The invention relates to an action recognition system for online education, and the system comprises a real-time snapshot device which is disposed at a client of remote education and is used for carrying out the real-time snapshot processing of an environment where a remote education user is located, so as to obtain and output a corresponding snapshot processing image; and the contact notification mechanism is used for executing on-site broadcasting of notification information corresponding to small action recognition when the contact between the head and the hand is intelligently recognized based on various image processing data by adopting the multiple trained Hough neural network. According to the invention, the Hough neural network after multiple times of training can be introduced to intelligently identify whether the head and the hand are in contact based on various image processing data, and when the head and the hand are intelligently identified to be in contact, on-site broadcasting of notification information corresponding to small action identification is executed. Therefore, intelligent detection and warning of illegal small actions of remote education customers are realized.