129 results about "Activity recognition" patented technology

The invention discloses a low-power-consumption warning ground pile awakening method and system based on human activity recognition, and belongs to the technical field of image analysis and intelligent security and protection. According to the method, in a micro-power-consumption mode, the environment is monitored through an image sensor, and when environment changes meet awakening conditions, lightweight human body detection is conducted through an edge AI chip. And if the human activity probability exceeds the confidence coefficient, starting a main camera and a high-computing-power AI chip to carry out deep behavior analysis, fusing a behavior analysis result with a geographic position and a timestamp, generating a risk decision result, and triggering a dynamic response. According to the invention, a hierarchical wake-up and multi-source fusion technology is adopted, on-demand work is realized, power consumption is greatly reduced, early warning accuracy is improved through deep behavior analysis, and the problems of high energy consumption and inaccurate early warning of traditional equipment are effectively solved.

The invention discloses a federal personalized human activity recognition training method based on a super network, which comprises the following steps that: a server randomly selects a plurality of clients to participate in training, and broadcasts embedded network parameters to the clients; and after receiving the embedded network parameters, the client generates an embedded description vector in combination with the local data set and uploads the embedded description vector to the server. And the server generates corresponding personalized model parameters according to the embedded description vector uploaded by each client, and then issues the personalized model parameters to the clients for local fine tuning. And after fine tuning is completed, the client uploads the personalized model parameter update quantity to the server, and the server updates the super network according to the personalized model parameter update quantity, generates an embedded description vector update quantity and returns the embedded description vector update quantity to the client. And the client generates an update gradient for guiding the embedded network according to the update quantity of the embedded description vector by combining the similarity consistency constraint between the embedded space and the personalized model parameter space. And the server collects the update gradients uploaded by all the clients and aggregates the update gradients to complete the update of the embedded network.

The invention relates to the technical field of human activity recognition, and particularly discloses a human activity recognition method and system based on a lightweight hybrid neural network and self-attention migration and a medium thereof, and the method comprises the steps: 1, employing an HHAR data set, and carrying out the preprocessing; step 2, constructing a teacher model based on a CNN-LSTM-Transform hybrid architecture, and constructing a teacher model based on the CNN-LSTM-Transform hybrid architecture; step 3, constructing a student model of a hybrid architecture based on CNN-LSTM-Transform; and step 4, utilizing a self-attention migration mechanism to guide the student model to learn the attention distribution mode of the teacher model, and realizing efficient migration of knowledge. According to the method, the calculation complexity and the storage requirement can be remarkably reduced while the expression ability of the model is maintained.

Systems and methods presented herein include one or more guest activity recognition devices configured to recognize activity of one or more guests within a physical environment of an amusement park. The system also includes one or more ride vehicles of a ride of the amusement park, each ride vehicle including an artificial intelligence entity management system configured to maintain one or more ride vehicle artificial intelligence entities of the ride vehicle based at least in part on the recognized activity of the one or more guests; and one or more features disposed on the ride vehicle and configured to be activated by the artificial intelligence entity management system to simulate the existence of the one or more ride vehicle artificial intelligence entities in accordance with one or more properties of the one or more ride vehicle artificial intelligence entities.

The invention discloses a WiFi signal human body activity identification method based on a time sequence alignment network, and relates to the technical field of human body activity identification. Comprising the following steps: acquiring channel state information; dividing into a training set and a test set; constructing a time sequence alignment network model, introducing a deep large-kernel convolution enhanced backbone network to extract depth feature tensors, respectively inputting the depth feature tensors into a double-branch structure, performing local feature processing on an upper branch, converting the feature tensors into global feature vectors through global average pooling operation on a lower branch, and outputting the global feature vectors; performing joint optimization on the time sequence alignment network model by adopting a cross entropy loss function and a ternary loss function, and obtaining a trained time sequence alignment network model after a plurality of times of training; and inputting the test set into the trained time sequence alignment network model, and evaluating a human body activity identification result. According to the method, local feature dynamic matching is carried out on the depth features extracted by the backbone network, so that actions occurring in different time sequences are aligned, and the recognition accuracy is further improved.

An exemplary virtual IMU extraction system and method are disclosed for human activity recognition (HAR) or classifier system that can estimate inertial measurement units (IMU) of a person in video data extracted from public repositories of video data having weakly labeled video content. The exemplary virtual IMU extraction system and method of the human activity recognition (HAR) or classifier system employ an automated processing pipeline (also referred to herein as “IMUTube”) that integrates computer vision and signal processing operations to convert video data of human activity into virtual streams of IMU data that represents accelerometer, gyroscope, or other inertial measurement unit estimation that can measure acceleration, inertia, motion, orientation, force, velocity, etc. at a different location on the body. In other embodiments, the automated processing pipeline can be used to generate high-quality virtual accelerometer data from a camera sensor.

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.

A system for recognizing human activity from a video stream includes a classifier for classifying an image frame of the video steam in one or more classes and generating a class probability vector for the image frame based on the classification. The system further includes a data filtering and binarization module for filtering and binarizing each probability value of the class probability vector based on a pre-defined probability threshold value. The system furthermore includes a compressed word composition module for determining one or more transitions of one or more classes in consecutive image frames of the video stream and generating a sequence of compressed words based on the deter-mined one or more transitions. The system furthermore includes a sequence dependent classifier for extracting one or more user actions by analyzing the sequence of compressed words to and recognizing human activity therefrom.

Implementations generally use multiple cameras to recognize clinical activities. In some implementations, the method includes acquiring multiple videos of multiple objects in an environment. The method further includes determining one or more key points for each object of the multiple objects. The method further includes recognizing activity information based on the one or more key points. The method further includes calculating workflow information based on the activity information.

This invention discloses a method and system for protecting privacy in wearable devices for human activity recognition, aiming to solve the problems of existing methods struggling to balance privacy and utility, and high resource consumption on the edge. The method first acquires multi-channel time-series signals from the wearable device and constructs fixed-length feature vectors. Based on a sensitive attribute inference model and the number of high-risk features, it divides the device into high-risk and low-risk feature sets. Only the high-risk feature sub-vectors are subjected to perturbation noise to generate perturbation latent representations, which are then uploaded to an edge server along with the low-risk feature set. The edge server generates the final feature vectors and assesses the risk of privacy leakage. An offline evaluation is used to construct a privacy-utility trade-off, dynamically adjusting the perturbation dimension and noise amplitude. This invention achieves targeted protection of high-risk features, maximizing the preservation of key recognition information, reducing edge computing power and communication overhead, adapting to the differentiated needs of different scenarios, and balancing privacy security and activity recognition performance.

The application discloses a fine-grained household power monitoring system and method combined with an intelligent sound box and an electric meter, and belongs to the related field of household electrical appliance energy consumption monitoring.The method comprises the following steps: the system automatically learns the correlation between electrical appliance power and sound, i.e., consistency information and complementary information; power events are divided into high power changes and low power changes, and scene discovery is iteratively performed; sound features and power features with correlation, i.e., key feature pairs, are found, and it is understood through clustering that which key feature pairs belong to the same electrical appliance state; a noise-robust sound-based electrical appliance state recognizer is trained; and the learned consistency information and complementary information, i.e., the recognition result of the electrical appliance state recognizer, is used to realize electrical appliance energy consumption decomposition in a cross-modal correlation fusion manner, and then the power consumption of each type of electrical appliance is inferred.The application helps users understand fine-grained household power consumption in a low device cost and low labeling cost manner, helps users cultivate low-carbon power consumption habits, and can also assist in user activity recognition.

The application discloses a kind of edge cloud cooperation human activity recognition modeling methods based on heterogeneous multi-modal data, and is divided into two stages of centralized pre-training and multi-modal semi-supervised fine-tuning.In the first stage, by dynamic mask contrast learning, pre-train the basic model based on converter using the joint dataset with super-class information, so that it has the ability to extract robust features from any modal combination.In the second stage, the local unlabeled data is used by the end-side client to generate pseudo-labels through weak mask view and calculate losses based on strong mask view for local update;Cloud side aggregates heterogeneous model parameters from different clients, and fine-tunes the global model using a small amount of labeled data.The application effectively alleviates the data heterogeneity problem by separating single-modal feature encoding and cross-modal information fusion, and can efficiently train a high-performance multi-modal human activity recognition model using a small amount of labeled data and unlabeled data while protecting data privacy.

The application discloses an activity recognition method based on a variational autoencoder and a time sequence network, and belongs to the technical field of behavior feature recognition, and comprises the following steps: S1, collecting an original data set of activity signals and performing pretreatment; S2, extracting latent features based on S1 by using a VAE, and outputting a reconstruction error; S3, obtaining time dynamic features based on S1 by using a CNN-GRU; S4, performing feature fusion based on the results of S2 and S3 by using a self-attention mechanism, performing soft assignment by using a deep embedding clustering layer, and performing classification by using a classifier, and completing model training; S5, testing, judging whether input is OOD data, and outputting a category. The application effectively recognizes and processes OOD data by judging the reconstruction error and the output of the deep embedding clustering layer, and enhances the robustness and generalization ability of the model; by combining the VAE, the CNN, the GRU and the self-attention mechanism, space-time features in IMU signals are comprehensively extracted, the accuracy of activity recognition is improved, time sequence information is effectively reserved, and the model training efficiency is improved.

Embodiments of the present disclosure relate to activity recognition methods utilizing automatic training based on inertial sensors. A technical advance is disclosed that utilizes inertial sensor data associated with a device to determine a new feature array and determine whether the new feature array is within an existing class within a state space associated with the inertial sensor data. In response to the new feature array being included in the existing class, the new feature array is added to the existing class and a representation of the existing class in the state space is updated based on the new feature array and an existing representation of the existing class. In response to the new feature array not being included in the existing class, a new class is created based on the new feature array.

In one embodiment, a method includes accessing a current sensor signal from an inertial measurement unit (IMU) of an earbud in a current earbud orientation worn by a user and accessing a baseline-orientation IMU sensor signal for the earbud in a baseline earbud orientation. The method further includes estimating, for the earbud and based on the current sensor signal and the baseline-orientation IMU signal, an orientation transformation matrix that transforms the current sensor signal from the current earbud orientation to the baseline earbud orientation; and transforming the current sensor signal from the current earbud orientation to the baseline earbud orientation using the orientation transformation matrix.

Some embodiments relate to a system for automated gaming recognition, the system comprising: at least one image sensor configured to capture image frames of a field of view including a table game; at least one depth sensor configured to capture depth of field images of the field of view; and a computing device configured to receive the image frames and the depth of field images, and configured to process the received image frames and depth of field images in order to produce an automated recognition of at least one gaming state appearing in the field of view. Embodiments also relate to methods and computer-readable media for automated gaming recognition. Further embodiments relate to methods and systems for monitoring game play and / or gaming events on a gaming table.

The invention relates to the technical field of voice signal processing. The invention discloses a human body activity identification method and device based on a sound event, a terminal and a storage medium, which can reduce the dependence on manual marking data, reduce the cost of manual marking and improve the accuracy of human body activity identification of a human body activity model. The human body activity recognition method based on the sound event comprises the steps of collecting audio data under the condition that the sound event is detected; filtering the audio data to obtain audio features; the audio features are input into a human body activity model for human body activity recognition processing, a human body activity recognition result corresponding to the audio data is obtained, and the human body activity model is formed by training an unlabeled data set and a labeled data set.

An object of a system according to an exemplary embodiment is to detect a security threat of a server in real time and quickly deal with the security threat.SOLUTION: A system according to an embodiment includes a real-time analysis unit, a suspicious activity detection unit, a threat handling unit, a report generation unit, and an automatic learning unit. The real-time analyzer analyzes the behavior of the server in real time using the generated AI. The suspicious activity detection unit detects a suspicious activity from the data analyzed by the real-time analysis unit. The threat handling unit identifies and handles a threat based on the suspicious activity detected by the suspicious activity detection unit. The report generation unit generates a security report based on the information on the threat addressed by the threat handling unit. The automatic learning unit automatically learns based on the report generated by the report generation unit to improve the security measure.SELECTED DRAWING: Figure 1

The embodiment of the invention provides a virtual IMU self-supervised human body activity identification method, system and device and a medium. The method comprises the steps of obtaining a to-be-identified action category; processing according to the to-be-recognized action category and a preset language model to obtain a natural language action description; generating a three-dimensional action sequence according to the natural language action description and a preset action generation mechanism; inputting the three-dimensional action sequence into a target virtual IMU (Inertial Measurement Unit) for calculation to obtain virtual IMU data; obtaining a training sample set according to the virtual IMU data to extract a preset encoder to obtain encoder parameters; and determining the type of the action to be identified according to the encoder parameters and a preset task classifier. According to the embodiment of the invention, a large number of diversified training samples can be generated, and the generated virtual data better fits the actual action characteristics, so that the quality of the training samples is improved, the to-be-recognized action category can be accurately recognized, and the accuracy and robustness of action recognition are improved.

The invention relates to the technical field of data analysis, in particular to an Internet of Things time sequence analysis method based on a self-adaptive hypergraph neural network, which comprises the following steps: acquiring a plurality of human activity identification data, determining a selected data volume according to data fluctuation, determining activity label similarity according to information association degree, and obtaining a time sequence analysis result; according to the activity label similarity, human activity identification data is selected as training data according to the overall similarity and the selection similarity or according to the amplitude change rate; according to the motion complexity of the training data, determining whether to increase and adjust the scale feature level to determine a target scale feature level; and the features of the points are connected through Tensor, and classification is carried out through Linear to obtain a prediction category result corresponding to the training data. The method can improve the recognition precision of the human activity recognition task.

Disclosed are artificial intelligence (AI)-based security risk prediction system and method for targets to be protected in a cloud environment. The method includes: collecting cloud logs and system logs for the targets to be protected in real time; learning all activity logs included in the cloud logs and the system logs for the targets to be protected of a corresponding member company through an AI algorithm; identifying a new activity among activities for the targets to be protected based on a learning process through the AI algorithm, and in response to the identified new activity being a new activity related to security, identifying a first activity pattern comprising the corresponding new activity; identifying an order of an preparatory activity for the new activity in the first activity pattern; identifying a risk score corresponding to the order of the preparatory activity for the new activity; and calculating a risk score of each target to be protected by summing identified risk scores of all new activities.

System and methods are provided for mitigating click farm fraud by receiving network data and sensor data from a plurality of computing devices, extracting one or more features from the sensor data and the network data for each of the devices. The features represent one or more of a local physical environment and communication channel environment associated with a device of the plurality of computing devices. The method includes determining one or more subsets of the plurality of computing devices based on environmental and network characteristics of the one or more features, identifying, based on the one or more subsets and detected influencer activities, co-located computing devices, and responsive to determining that a count of the co-located computing devices is greater than a predetermined count, sending a session terminating command to one or more servers in communication with the co-located computing devices to mitigate click farm fraudulent activities.

The application discloses a human action recognition method, system and device based on WiFi channel state information imaging and a readable storage medium, and the method comprises the following steps: collecting WiFi channel state information, and extracting amplitude information changing with time in the channel state information; constructing a Gram angle difference field matrix based on the amplitude information changing with time, and converting the Gram angle difference field matrix into a Gram angle difference field image; inputting the image data into a convolutional neural network model to extract image features, and outputting a recognition result, wherein the convolutional neural network comprises four convolutional layers, each convolutional layer is followed by a batch normalization layer and a rectified linear unit layer, after the rectified linear unit layer, a maximum pooling layer is used to extract main features of adjacent regions, an adaptive average pooling layer is used after the fourth convolution to specify the output size of a feature vector, and a classifier with a Dropout layer and a linear layer is arranged after the adaptive average pooling layer. The application realizes higher activity recognition accuracy with lower model complexity.

Various embodiments provide methods performed by a user equipment (UE) including receiving sensor data associated with activity of a user from a plurality of inputs at a first-time instance, determining a state of the user at the first-time instance based on the sensor data, and identifying, by the processor, at least one attribute of the user or the activity of the user based on the sensor data and the state of the user collected at multiple time instances including the first-time instance. Identifying the at least one attribute may use an inference model to receive selected sensor data as an input and to provide an output classifying the state of the user at the multiple time instances as an attribute of the user.

The invention relates to the technical field of artificial intelligence and behavior recognition, in particular to a learning activity recognition method and device based on a memory module, and the method comprises the steps: generating a plurality of video clips according to a classroom video, and inputting the plurality of video clips into a pre-constructed audio recognition model, so as to output at least one subtitle file; generating a key teacher instruction meeting a preset association condition with the learning activity according to the subtitle file; and inputting the key teacher instruction, the classroom video frame and the instruction text into a pre-constructed learning activity identification artificial intelligence model to obtain a learning activity identification result of each video clip. Therefore, the problems of low teaching behavior recognition accuracy, incomplete semantic understanding and difficulty in effectively supporting automatic teaching analysis and precise teaching application requirements caused by difficulty in adapting to complex scenes such as multi-subject interaction, dense language communication and long video in a classroom due to the fact that related technologies are mainly based on a universal activity recognition model are solved.

An apparatus using a lightweight transformer model for human activity recognition in a portable device, includes: a data collector configured to collect data for human activity recognition based on an mmWave radar; a data processor configured to perform a data processing process to process mmWave data and convert the mmWave data into an input form for a model; a lightweight GST model part configured to generate a feature vector by combining a grouped attention mechanism, which splits an input sequence into several small groups and independently calculates attention within each group, and a sparse attention mechanism, which calculates attention only for selected location pairs rather than calculating attention for all location pairs, to classify an output class through a fully connected layer; and a human activity recognition result outputter configured to output human activity recognition results based on human activity types classified in the lightweight GST model part.

The invention discloses a radar human body activity identification spectrum diffusion generation method based on a dual-attribute condition, and belongs to the technical field of crossing of radar perception and multi-modal artificial intelligence. According to the method, natural language semantic prompt and a condition diffusion generation framework are combined, and two semantic conditions of'character identity 'and'action category' are simultaneously introduced in a generation process by utilizing a cross-modal semantic alignment and hierarchical semantic modulation technology; and a micro-Doppler radar spectrogram with a real time-frequency structure and semantic consistency can be accurately synthesized according to semantics. The method provided by the invention can be used for automatic generation and data enhancement of high-quality pseudo actual measurement data, and supports training, verification and robustness evaluation of a downstream human body activity recognition model, so that the actual measurement data acquisition and labeling cost is remarkably reduced, the recognition precision and generalization ability are improved, and the method is suitable for popularization and application. And actual deployment and popularization of application scenes such as radar-based intelligent security and protection, smart home, remote health monitoring and the like are promoted.

The present subject matter relates to a system (100) and a method (300) for providing automated resolution to one or more anomalous events in an enterprise information technology (IT) environment. The system (100) integrates a processor (201) and a memory (202) that stores instructions to execute various tasks. The system (100) monitors activities within the enterprise IT environment, identifies one or more anomalous events, and correlates the identified anomalous events with one or more predefined resolution workflows. Each workflow includes specific operating instructions tailored to address the identified anomalies. Upon detecting an anomaly, the system (100) extracts the relevant operating instructions from the corresponding workflow and executes them to resolve the issue. Thus, the system (100) significantly reduces the mean time to resolve by automating the detection and resolution of anomalies, thereby improving operational efficiency and minimizing downtime in the enterprise IT environment.

An audio-visual pilot activity recognition system including one or more image collectors, one or more audio collectors, and a processor configured to collect at least one image signal from the one or more image collectors, collect at least one audio signal from the one of more audio collectors, and determine a pilot activity based on the collected at least one image signal and the collected at least one audio signal.

The application discloses a cross-user activity recognition method based on source domain data screening and training feature constraint. The data distribution of different users is calculated through optimal transmission distance, and source domain data screening is performed according to the distribution distance of the users, thereby reducing the problem of poor domain adaptation migration. Data enhancement is performed in the screened source domain data, so that the source domain data distribution is more continuous, and the influence of the reduction of the source domain data is alleviated. Then, a joint loss of a regularization random drop loss, a classification loss and a hierarchical feature maximum mean error is constructed to train a network model, and through the training feature constraint and the hierarchical feature maximum mean error alignment method, better domain adaptation effect is realized. The application improves the generalization capability of the model in cross-user activity recognition.