987 results about "Gesture recognition" patented technology

An autonomous environmental regulation and behavioral monitoring system is disclosed, configured to adapt temperature, lighting, and acoustic conditions based on real-time emotional and physiological data. The system includes a dual-redundant central processor, hierarchical communication networks, multi-angle visual acquisition units, infrared thermometers, and modular environmental subsystems. It detects posture, gestures, facial expressions, and thermal signals to classify user states and apply individualized airflow, light, and sound modulation without relying on external internet connectivity. The system also monitors connected appliances using voltage-based pressure analysis to forecast device degradation. With integrated gesture recognition, privacy-preserving data handling, and predictive adaptation, the invention enables multi-user personalization, long-term learning, and uninterrupted operation within residential, administrative, or healthcare infrastructures.

The invention discloses a real-time dynamic trajectory tracking method and system for millimeter wave radar gesture recognition, and relates to the technical field of gesture recognition tracking, and the method comprises the steps: receiving an echo signal reflected by a gesture, extracting a potential target point cloud, and carrying out the clustering generation of a gesture point cloud sequence; establishing a multi-modal motion model library, dynamically selecting an optimal motion model by adopting graph matching, and generating a prediction state in combination with a gesture point cloud sequence; on the basis of a Poisson multi-Bernoulli hybrid filtering framework, according to the signal-to-noise ratio and spatial distribution of the current gesture point cloud sequence, dynamically adjusting the observation weight, optimizing the observation point cloud, and carrying out optimal association by combining Mahalanobis distance with dynamic time warping; a multi-hypothesis tracking strategy is adopted to maintain trajectory hypothesis, an optimal trajectory is selected through a trajectory scoring mechanism, and Kalman filtering smoothing processing is performed on the optimal trajectory. According to the method, high-precision and low-delay tracking of gesture motion is realized, gesture habits of different users and complex environment interference can be adapted, and meanwhile, relatively high track precision is kept.

The invention discloses a general electromyographic signal processing method and system based on a large self-supervised model. The general electromyographic signal processing method comprises the following steps: step 1, acquiring a multi-source original multi-electrode channel EMG signal X from an electromyographic acquisition device; and finally, performing data unification processing, and finally converting into a space-time activity diagram with a fixed size of 224 * 224. On the basis of the space-time activity diagram and the fatigue state mark, constructing an AEMG for training according to heterogeneous unlabeled EMG data collected by a collection device; performing light-weight Adapter layer fine adjustment on the pre-trained large myoelectricity model to adapt to gesture recognition muscle force regression gait analysis or rehabilitation evaluation downstream tasks; aiming at the problem that the dimension and the structure of myoelectricity data are not matched due to different acquisition devices, acquisition parts and acquisition tasks, original signals are converted into space-time activity diagrams in a unified format through data unification processing, device differences are represented by combining a sensor embedding module, effective alignment of cross-source data is achieved, and the accuracy of the data is improved. And a basis is provided for large-scale data utilization.

The invention relates to a gesture recognition and projection fusion non-contact interaction method and device, equipment and a medium. The method comprises the steps of obtaining multi-user gesture original image data, extracting geometric features and boundary features of a hand contour after image preprocessing and noise removal, and generating recognition result data; secondly, on the basis of three-dimensional coordinate information in the recognition result, a tracking trajectory is analyzed through trajectory continuity, a user identity identifier is matched in combination with a user identity feature library, and user space positioning information is generated; then, a three-dimensional track point sequence is extracted from the user space positioning information, an initial projection area is divided by adopting a dynamic area segmentation algorithm, and an operation area mapping relation is formed; and finally, analyzing the operation area mapping relation to extract gesture action features, generating a control instruction set in combination with a preset rule, and scheduling a conflict instruction to generate an instruction execution sequence. By adopting the method, the scene limitation of traditional contact type interaction can be broken through, and the stability and practicability of non-contact interaction in a multi-user environment are improved.

The invention belongs to the technical field of intelligent wireless sensing and radar signal processing, and particularly relates to a millimeter wave radar gesture recognition feature processing method, which is particularly suitable for scenes with environment interference (such as walking of others and static clutter), and specifically comprises the following steps: S1, preliminary filtering by a self-adaptive filter; s2, carrying out improved I CEEMDAN decomposition; s3, I MF component classification and processing; s4, signal reconstruction; the experimental result shows that the method provides a robust and efficient solution for gesture recognition, and can be widely applied to the fields of man-machine interaction, virtual reality, intelligent equipment control and the like.

The invention provides a transient stage electromyographic signal gesture recognition method based on electromyographic activation intensity transfer characteristics and self-adaptive analysis window length. The invention aims to solve the problems of imbalance between spatial features and operation cost, imbalance between recognition accuracy and response speed and low algorithm generalization between individuals and between tasks in the traditional electromyographic signal processing technology when transient stage signals are utilized. The method comprises the following steps: firstly, performing signal acquisition and preprocessing, starting point detection, signal frame formatting, feature extraction and SVM classifier training in an offline mode; in the online mode, the preprocessed signals are collected, a starting point is detected, the signals are formatted, myoelectricity activation intensity transfer features are extracted and input into an offline SVM classifier, the length of an analysis window is dynamically adjusted according to the confidence coefficient, and finally the threshold value of the confidence coefficient is fed back and adjusted in real time. Practice verifies that the method can effectively give consideration to the spatial features and the operation cost, achieves the balance between the recognition accuracy and the response speed in different individuals and tasks, remarkably improves the generalization of the algorithm, and provides more efficient and accurate technical support for the fields of artificial limb control, exercise rehabilitation equipment and the like.

The invention relates to the field of gesture recognition interaction, in particular to a human-computer interaction method and system based on static gestures. Comprising a gesture acquisition module, an image processing and tracking module, a feature recognition and gesture library management module, a memory, a processor, an instruction mapping module, an execution control module and a user feedback module. Brand new gestures and corresponding instructions are created and defined, the interaction flexibility is high, and individual requirements can be fully met.

The invention relates to a non-inductive security monitoring method and system based on a radio frequency identification technology, and relates to the technical field of data processing. The method mainly comprises the following steps: acquiring original radio frequency data and original video data of a target area in a current time period; respectively carrying out feature extraction on the original radio frequency data and the original video data to obtain a radio frequency feature vector and a visual feature vector; performing feature vector fusion on the radio frequency feature vector and the visual feature vector to obtain a fusion feature vector; inputting the fusion feature vector of the current time period and the fusion feature vectors of N time periods before the current time period into a behavior classification model to obtain a behavior prediction result; if the behavior prediction result is limb movement, determining a limb recognition result corresponding to the current time period according to a limb recognition model; and if the behavior prediction result is gesture motion, determining a gesture recognition result corresponding to the current time period according to a gesture recognition model.

The invention discloses a dynamic gesture recognition control system for an intelligent wheelchair, and relates to the technical field of intelligent wheelchairs, the system comprises the following components: an environment sensing module which is composed of a laser radar and an ultrasonic sensor and is used for collecting information of a surrounding environment in real time and constructing an environment map; meanwhile, the position and the shape of the obstacle and the distance information between the obstacle and the wheelchair are accurately recognized; by introducing the dynamic gesture recognition technology, not only can the current static gesture action of the user be recognized, but also the following action intention of the user can be predicted based on parameters such as the speed, the angle and the acceleration of the gesture, so that the intelligent wheelchair can make a response in advance, closer and smoother interaction with the gesture operation of the user is achieved, and in addition, the user experience is improved. In combination with the information of the environment sensing module, the intelligent wheelchair can more accurately understand the intention of the user and perform reasonable obstacle avoidance operation in a complex environment, so that the overall use experience and safety are improved.

The invention discloses an intelligent music playing method based on three-dimensional scene interaction, and relates to the technical field of three-dimensional interaction, and the method comprises the steps: extracting a music feature vector of an input audio, inputting a lightweight neural network to output a music genre label, loading a three-dimensional scene in a three-dimensional scene template library based on the music genre label, and generating an initial three-dimensional music space; materializing a control component in the three-dimensional scene into a three-dimensional interactive object, and detecting the position and posture of the hand of the user in the initial three-dimensional music space to perform music playing control; and driving a dynamic element in the three-dimensional scene based on the music feature vector to control the amplitude of a top light beam, the ripple radius and the particle spacing of the central region, and mapping to the initial three-dimensional music space. According to the method, playing, volume and lyric functions are materialized into interactive objects, gesture recognition is introduced, and natural, visual and low-delay three-dimensional interactive experience is achieved; and an intelligent music playing effect with high immersion, high degree of freedom and high consistency is achieved.

A computer-implemented system and method for pupil-directed retinal projection in near-eye displays are disclosed. The computer-implemented system provides smart glasses with directed physical pixels that project light beams / signals directly onto user's retina based on pupil position and size. The glasses comprise frame, lenses with directed pixel layers, sensors for tracking pupil movement. Each directed pixel generates multiple virtual pixels by rapidly changing its emission angle. Each directed pixel with pupil's real-time tracking, provides automatic adjustments to virtual contents, ensuring accurate alignment of images with field of view of users while identifying vergence-accommodation conflict. The glasses function as prescription lenses, virtual reality displays, augmented reality devices without traditional optical systems. Additional features include depth sensors, cameras, connectivity to peripheral devices. The glasses enable seamless blend of virtual and real-world experiences, creating immersive “Mixverse” environment. Various input methods, including gesture recognition and brain-computer interfaces, allow for intuitive control and interaction.

The invention discloses a gesture recognition sensing method and system based on a Wi-Fi signal. The method comprises the following steps: acquiring a Wi-Fi signal in an environment, and outputting a two-dimensional data matrix containing amplitude information and phase information; uniformly converting the data matrix into a standard input format adaptive to a sequence encoder, and pre-training the codec through a mask self-supervision reconstruction mechanism; keeping the structure and parameters of the pre-trained encoder, replacing the decoder with a gesture recognition network formed by a lightweight classification network, and carrying out model end-to-end training by using a supervised optimization mode; and inputting actually acquired original signal data containing amplitude information and phase information by using the model subjected to end-to-end training, and outputting a corresponding gesture category. According to the method, the original Wi-Fi CSI signal is directly utilized, non-contact gesture recognition with low cost, high precision and strong generalization capability is realized, and the method is suitable for deployment of the mobile terminal and the edge equipment in an actual environment.

A method for augmented reality assisted automotive inspection and automatic ordering of automotive parts and / or services begins with the detection of an automobile inspection session's initiation using an AR headset worn by a technician. The system then generates relevant AR elements for the inspection session, which may include technician workflows, part identifications, part descriptions, and ranked ordering options for parts and services. These AR elements are displayed as an overlay to the technician's field of view through the AR headset. The method includes detection of a technician's selection of a part or service via voice command, gesture recognition, or gaze tracking. Finally, the selected part or service is automatically ordered without requiring any physical manual input from the technician, streamlining the entire inspection and ordering process.

A system and a method for automatically assisting physicians during patient encounters includes transcribing and diarizing physician-patient interactions, extracting clinical concepts, and combining data with patient history data to provide suggestions to the physicians. A speech and gesture recognition module captures audio from patient-doctor interactions and transcribes the audio into text in real-time. A natural language processing module diarizes the transcribed text to attribute speech. A clinical concept extraction module identifies and extracts clinical concepts from transcription. A clinical recommendation module integrates real-time data with historical patient records and analyzes the combined data set to generate evidence-based suggestions to the physician. A physician's own historical data is validated to generate a unique profile for each physician. Interactions with the physician are analyzed to understand decision-making patterns. The physician's profile is updated and future physician decisions are forecast based on past behavior.

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 discloses a question answering system and a construction method for a motor vehicle driving teaching knowledge graph, and aims to overcome the defects of knowledge coverage integrity, scene adaptability, reasoning depth and interaction naturalness in the prior art. A traffic regulation document, coach voice and short videos and vehicle-mounted real-time data are integrated through a data acquisition module to cover rules and experience knowledge, a knowledge construction module constructs a space-time dynamic knowledge graph containing static and dynamic nodes based on Dgraph, space-time association is achieved by means of GeoHash coding, real-time updating is completed through an MQTT protocol, and the time-space dynamic knowledge graph is stored through a storage module. The intelligent reasoning module calls a DoWhy model to construct an'operation, principle and result 'causal chain in combination with a physical formula so as to output deep interpretation, and the interaction module achieves multi-mode immersion teaching through speech synthesis, gesture recognition and 3D animation, dynamically optimizes knowledge pushing according to student operation data, and achieves multi-mode teaching. The comprehensiveness, the real-time performance, the scientificity and the interaction naturalness of driving teaching are improved, and the method is suitable for intelligent driving teaching scenes.

The invention provides a gesture recognition method based on an intelligent glove and the intelligent glove. The method is applied to artificial intelligence. The method comprises the following steps: acquiring tactile data, inertial data and stretching data; generating tactile window data, inertial window data and stretching window data corresponding to the tactile data, the inertial data and the stretching data respectively based on a preset window; performing feature extraction on the tactile window data, the inertial window data and the stretching window data by using a pre-trained gesture recognition network model to obtain a tactile feature vector, an inertial feature vector and a stretching feature vector; performing feature fusion on the tactile feature vector, the inertial feature vector and the stretching feature vector to obtain a fused feature vector; and performing gesture recognition based on the fused feature vector to obtain a gesture recognition result. According to the invention, efficient and accurate complex gesture recognition is realized.

The invention discloses a gesture recognition method and system based on an active acoustic sensing field, and the method comprises the steps: transmitting a frequency modulation continuous wave through a loudspeaker on head-mounted equipment, and receiving an echo signal reflected by the hand of a wearer through a microphone on the head-mounted equipment; performing frequency mixing, low-pass filtering and analog-to-digital conversion on the echo signal to obtain a digital baseband signal; performing range correction and noise reduction processing on the digital baseband signal to obtain a hand echo signal; inputting a hand echo signal into the multi-scale time sequence encoder, and capturing long time sequence dependence to obtain multi-scale spatial-temporal characteristics; performing trajectory tracking on the multi-scale spatial-temporal features by the hand key point regression branches to obtain hand key point features; inputting the hand key point features and the multi-scale spatial-temporal features into the gesture recognition branch to obtain a gesture recognition result; according to the method, multipath interference and spatial fuzziness in the acoustic sensing process can be overcome, and high-precision and fine-grained gesture recognition is realized.

The invention provides a posture recognition method and system based on a posture recognition neural network, and relates to the technical field of posture recognition, and the method comprises the steps: constructing a standard library containing safety and dangerous posture types, and marking a plurality of key points to generate a soft label sample data set; defining the key points as graph nodes, and constructing a feature matrix and an adjacent matrix; a double-branch fusion network model is designed, key point topological structure features are extracted through image convolution branch, image global context features are extracted through global visual branch, and attitude similarity spectral vectors are output after fusion; and finally, the attitude state is judged by calculating the geometric difference degree between the image to be analyzed and the most similar attitude template on the key point distance and the joint angle and comparing the overall difference between the image to be analyzed and the safety / danger camp. According to the method, the dual advantages of data driving and rule verification are combined, the recognition robustness and decision interpretability of dangerous postures are remarkably improved, and the safety risk is effectively reduced.

The invention provides a crane commander gesture recognition method based on a multi-branch G-LSA millimeter wave radar, and the method comprises the steps: transmitting a frequency modulation signal through an FMCW millimeter wave radar, and receiving an echo signal reflected by a gesture; performing frequency mixing and processing on the emission signal and the echo signal to obtain a beat frequency signal, and acquiring gesture signal data of a commander in a real application environment through ADC sampling; performing signal preprocessing on the sample to obtain micro-Doppler features of the sample, establishing a corresponding point cloud data set D, and performing unified processing on the micro-Doppler features of the sample to obtain a three-dimensional mixed feature tensor; constructing a gesture recognition model based on multi-branch G-LSA, training the model, and optimizing the weight of the model to obtain a gesture recognition model; aDC data of gesture actions of a crane commander are collected, and a three-dimensional mixed feature tensor of the ADC data is obtained and input into a multi-branch G-LSA gesture recognition model; an identification result is sent to a crane cab; according to the method, a safer and more reliable gesture interaction scheme is provided for high-precision and high-risk scene optimization such as crane command.

The invention provides a playing control method and device, electronic equipment and a storage medium, and the method comprises the steps: inputting collected gesture collection information into a set recognition model; wherein after the set recognition model outputs a gesture prediction result corresponding to the gesture acquisition information, gesture recognition is carried out based on the gesture acquisition information; under the condition that the set recognition model outputs a gesture prediction result, controlling a preloading player to load a monitoring video stream based on the gesture prediction result; and under the condition that the set recognition model outputs a gesture recognition result, controlling the playing of the monitoring video stream based on the gesture recognition result and the preloading player. According to the method and the device, the user can realize switching of the monitoring video stream through the gesture, a more convenient playing control mode is provided for the user, and the required video is loaded in advance through the gesture prediction, so that the time required for switching the monitoring video stream is shortened, the switching of the monitoring video stream is smoother, and the use experience is further improved.

The invention discloses an intelligent cabin interaction priority dynamic allocation system and method and electronic equipment, and relates to the field of cabin interaction, and the system comprises a multi-mode signal input layer which is used for continuously receiving original interaction instruction signals of a voice recognition module, a gesture recognition module, an eye movement tracking module and a touch screen module; the context sensing layer is used for collecting vehicle state data, driver state data and environment state data in parallel; the dynamic priority arrangement engine comprises a multi-factor fusion and weight calculation module and an arbitration and scheduling module; the multi-factor fusion and weight calculation module is used for normalizing the context data into a weight influence factor and calculating a real-time dynamic priority score of each interaction mode; the arbitration and scheduling module is used for comparing priority scores and selecting an optimal execution mode, and sending suppression signals to other modes at the same time; and the instruction execution layer is used for receiving the final instruction of the arbitration and scheduling module and issuing the final instruction to a vehicle-mounted application for execution.

The invention discloses a skeleton gesture recognition method based on an adaptive space-time decoupling network, and the method is characterized in that a skeleton gesture sequence is obtained and preprocessed, and multi-modal fusion features are obtained; constructing an adaptive space-time decoupling network, wherein the adaptive space-time decoupling network comprises N adaptive space-time decoupling units connected in sequence, a self-supervised time-channel adaptation module fusing the space-time features of each mode, a global average pooling layer, a random inactivation layer and a full connection layer, and the N adaptive space-time decoupling units, the self-supervised time-channel adaptation module, the global average pooling layer, the random inactivation layer and the full connection layer are connected in sequence; inputting the multi-modal fusion features into an adaptive space-time decoupling network for training; performing skeleton gesture recognition by adopting the self-adaptive space-time decoupling network obtained by training, and finally outputting a category prediction score of the skeleton gesture; the method has the advantages that spatial representation is enhanced, interference is decoupled, the modeling efficiency of multi-granularity long and short time sequence dependence is improved, and therefore the recognition precision of skeleton gestures is improved.

A distributed ultra-wide band (UWB) radar system and methods for operating the same are disclosed. The distribute radar system uses a plurality of separate UWB radios or nodes that do not use a centralized clock or source of time. However, the distributed UWB system operates to provide radar-like functionality and fine-grain sensing capabilities through synchronization based on line-of-sight (LOS) signal processing and noise estimations. Synchronized channel impulse responses (CIRs) can be processed for general object detection and tracking, gesture recognition, or even micro-motions such as monitoring vitals

The invention relates to the field of obstacle avoidance and gesture control, and particularly provides an intelligent obstacle avoidance and gesture control method and system fusing machine vision and ultrasonic waves. Step 2, preprocessing and time-space synchronization; step 3, carrying out multi-modal fusion; and step 4, decision output. According to the invention, collaborative obstacle avoidance of fusion of the machine vision sensor and the ultrasonic sensor is realized, the detection range is expanded, and the environmental adaptability is improved; and the gesture recognition accuracy and the response speed are improved through multi-modal sensor fusion. Through the fusion effect of the machine vision and the ultrasonic sensor, a new-generation sensing and interaction system which is high in reliability, high in accuracy, low in delay and intelligent and adaptive is finally formed.

The invention discloses a multi-mode control method for voice interaction and gesture recognition of an atmosphere lamp, and belongs to the field of intelligent control. The method comprises the following steps: S1, constructing a multi-mode sensing module, and arranging a bidirectional linear array microphone, a dual-lens infrared depth camera and a space calibration laser transmitter; s2, extracting voice features by adopting an extended spectrogram domain attention network, and generating a first semantic vector; s3, dynamic attitude features are extracted through the time-airspace sparse convolutional network; s4, performing modal alignment and cooperative coding on the two semantic vectors based on a bidirectional gating fusion mechanism, generating a joint interaction instruction vector, and solving modal conflicts by a time sequence synchronization and confidence regulation strategy; s5, inputting the vector into a multi-task intention classifier, analyzing a user intention and outputting a unique atmosphere lamp control code; and S6, transmitting the control code to a driving unit to realize light control. The beneficial effects are that voice and gesture fusion control is realized, and atmosphere lamp interaction intelligence and response precision are improved.

The embodiment of the invention relates to a gesture recognition method and system based on cross-modal feature alignment. The method comprises the following steps: acquiring a radar echo signal and a visual image of a gesture action; performing feature extraction on the radar echo signal based on a fusion attention mechanism to obtain a radar feature vector; performing multi-layer convolution processing on the visual image to extract a high-level semantic feature, performing dynamic feature extraction on the visual image based on a fusion space-time attention mechanism, and fusing the extracted high-level semantic feature and the dynamic feature to obtain a visual feature vector; performing time sequence alignment, hidden space alignment and semantic alignment on the radar feature vector and the visual feature vector under the constraint of total alignment loss; performing dynamic weighted fusion processing on the aligned radar feature vector and visual feature vector based on a cross-modal attention mechanism to obtain a fusion feature; and recognizing a gesture category according to the fused feature to obtain a gesture recognition result. According to the embodiment of the invention, the data of each sensor are effectively fused, and the recognition precision is improved.

The present disclosure provides an apparatus configured to perform gesture recognition and communicate with a smartphone, comprising one or more sensors configured to detect movement of at least one wearable component associated with a user, one or more memories configured to store gesture data, and one or more processors, coupled to the one or more memories and the one or more sensors, configured to capture, via the one or more sensors, motion data corresponding to movement of the at least one wearable component, input the motion data into a machine learning model trained to predict gestures, output, by the machine learning model, a gesture identifier based on the motion data, and transmit, via a wireless communication interface, the gesture identifier to a smartphone device.

The invention discloses an industrial operation gesture recognition model based on deep learning, and the model comprises the following steps: S1, collecting image data of an operation gesture of a worker, and carrying out the preprocessing of the image data; s2, constructing a gesture recognition model by using a convolutional neural network as an infrastructure; s3, training the model by using the preprocessed gesture image data set; s4, integrating the trained gesture recognition model into an industrial operation auxiliary system; the model based on deep learning can automatically learn the complex features of the gesture image and accurately recognize various gestures, the accuracy of industrial operation gesture recognition is effectively improved, misjudgment and missed judgment are reduced, product quality and production efficiency can be improved, for example, in a complex gesture recognition test, the accuracy of industrial operation gesture recognition is improved, and the accuracy of industrial operation gesture recognition is improved. Compared with a traditional method, the accuracy is improved by more than 20%.

The invention relates to the technical field of three-dimensional gesture recognition, and discloses a three-dimensional hand key point recovery method and system. The method comprises the following steps: acquiring a hand image and three-dimensional hand key point coordinates; inputting the three-dimensional hand key point coordinates into a hand posture encoder, and extracting potential structure feature vectors; searching a nearest neighbor code word of the potential structure feature vector in the structured codebook, and constructing a pseudo-quantization feature through rotation alignment; performing image coding on the hand image, and extracting image guide features; performing conditional diffusion modeling on the noise-added attitude code word index, and gradually predicting and restoring discrete representation of an original attitude under the assistance of image guide features to obtain a restored attitude code word index; and performing table look-up mapping and decoding on the recovered attitude code word index to obtain a three-dimensional hand key point coordinate prediction result. According to the method, the precision and robustness of three-dimensional hand key point estimation in a complex shielding scene are remarkably improved, and the method is particularly suitable for the application fields of man-machine interaction, virtual reality and the like.