311 results about "Activity detection" patented technology

The invention provides an intelligent sound box voice processing method and system based on artificial intelligence, and the method comprises the steps: obtaining audio data and mouth shape video data, carrying out the processing of the audio data and the mouth shape video data, and carrying out the multi-modal feature fusion, and obtaining a fusion feature; performing bimodal voice activity detection on the fusion features to obtain effective voice data; performing context sensing recognition of audio and video fusion on the effective voice data to obtain a first text; constructing a user feature model, and performing semantic understanding on the text based on the model to obtain an understanding result; performing intention recognition and slot filling based on the understanding result to obtain user intention and key information; generating a response strategy in combination with the user intention, the key information and the environment perception data; generating response voice according to the response strategy; and monitoring feedback information of the user to the response voice in real time, and updating the user feature model and the response strategy evaluation model based on feedback. According to the scheme, the voice can be recognized more accurately, and the safety and robustness of the system are enhanced.

The invention relates to the technical field of voice processing, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a voice intention recognition method, device and equipment and a medium, and the method comprises the steps: obtaining a to-be-processed voice signal, carrying out the voice activity detection processing of the voice signal, dividing the voice signal into a plurality of voice segments, analyzing semantic contents of the plurality of voice segments, determining semantic correlation information of each voice segment, analyzing sound source attributes of the plurality of voice segments, determining sound field type information of each voice segment, screening out a target voice segment from the plurality of voice segments according to the semantic correlation information and the sound field type information, and executing intention recognition processing based on the target voice segment to generate an intention recognition result. According to the invention, through a dual analysis mechanism of semantic correlation information and sound field type information, effective screening of voice segments is realized before voice recognition, non-target voice or interference segments are effectively prevented from being sent to an intention recognition model, and the accuracy of a recognition result is improved.

The invention relates to the technical field of computer security, and discloses a computer remote login identification device based on artificial intelligence, which comprises a user side device module used for collecting biological identification data and input behavior data of a user; the authentication and verification module is used for executing multi-mode identity authentication and dynamically adjusting an authentication strategy; the behavior analysis and risk assessment module is used for analyzing a behavior mode and a login environment of the user; the database and storage module is used for storing user identity information, behavior characteristic data and authentication logs; and the security monitoring and alarming module is used for monitoring the login activity of the user in real time and detecting potential security threats. According to the invention, the design based on multi-modal authentication greatly improves the security of remote login. By combining a plurality of authentication means such as face recognition, fingerprint recognition and voice recognition, the system can effectively prevent security vulnerabilities such as password stealing, fingerprint counterfeiting or facial image attack.

A video gateway system at a worksite is coupled to multiple cameras on a network, and backs-up video streams generated by the cameras to a backend cloud backup video storage system and a frontend (cache) video storage system. The video gateway system generates an aggregated video asset from a plurality of streams of video from the multiple cameras, and generates metadata and a backup report associated with the video asset. The video asset, metadata, and backup report are stored on the backend cloud backup video storage system in an file system, and are also stored on the frontend (cache) video storage system such that for at least a period of time, the video asset and the associated the metadata and backup report are stored on both the cloud backup video storage system, facilitating quick access and retrieval to the stored video for retrieval for streaming, activity detection, and other uses.

A method includes obtaining a first attack sequence comprising at least a first attack phase, a second attack phase, and a third attack phase. The method further includes determining that a first number of alerts generated for the first attack phase based on a set of detection operations and a second number of alerts generated for the third attack phase based on the set of detection operations satisfy a threshold criterion. The method further includes determining that a third number of alerts generated for the second attack phase based on the set of detection operations fails to satisfy the threshold criterion. The method further includes determining that the set of detection operations is to be modified to detect future malicious activities corresponding to the second attack phase.

A training data set which includes event sequences representing actions of respective users of an application, properties of the users, and dynamic attributes associated with events such as the elapsed time between successive events, is prepared. A machine learning model which provides probabilistic predictions of next events of input event sequences is trained using the training data set. A trained version of the model is stored.

Techniques and apparatuses are described that perform audioplethysmography and motion-sensing data fusion. Fusing motion-sensing data with audioplethysmography expands the situations in which audioplethysmography can operate. In one aspect, audioplethysmography and motion-sensing data fusion can be used for motion-artifact filtering. With motion-artifact filtering, noise caused by motion of a user can be attenuated to improve sensitivity and accuracy for audioplethysmography. This performance improvement expands the ability of audioplethysmography to support use cases during situations in which the user engages in an activity. In another aspect, audioplethysmography and motion-sensing data fusion can expand a functionality of a hearable to include activity detection and / or activity classification. While activity detection and / or activity classification can provide additional contextual information for other use cases associated with audioplethysmography, either can also be used to control an operation of the hearable and / or a computing device.

An apparatus for detecting fraudulent activity includes a device property data processing circuit, a security analysis circuit, an alert circuit, and an alert provisioning circuit. The device property data processing circuit is structured to interpret device property data corresponding to a device registered with an Internet of Things (IoT) device registry. The security analysis circuit is structured to determine, based at least in part on the device property data, that the device is subject to a fraud event. The alert circuit is structured to generate, responsive to the determined fraud event, a message that identifies the device. The alert provisioning circuit is structured to transmit the message.

Systems, methods, and other embodiments associated with quasi-supervised clustering for activity pattern characterization and anomalous activity detection are described. In one embodiment, a method generates a first sparse similarity matrix for nearest neighbors of a plurality of data points. The data points each characterize a pattern of activity associated with an account. The method generates a second sparse similarity matrix for random neighbors of the plurality of data points. The method recursively clusters the plurality of data points based on the first sparse similarity matrix. The method quasi-supervises the recursive clustering based on the second sparse similarity matrix to stop the iterative clustering when the data points are split into N clusters. The value of N is not pre-determined. The method detects that the individual data point has changed clusters, indicating anomalous activity. And, the method generates an electronic alert that the anomalous activity is associated with the account.

Embodiments of the technology described herein identify and mitigate phishing attempts by analyzing user input using a client-side proxy component and a proxy server. Embodiments disclosed herein provide systems, methods, and computer-storage media for employing proxy server capabilities in conjunction with a snapshot capturing an image or video recording of a target action input by a user into a software application. Certain embodiments disclosed herein employ proxy server capabilities to capture a snapshot and / or screen recording based on a user authorization or approval. For example, the proxy server proactively captures the snapshot or screen recording prior to, during, and after the user performing a target action. From the snapshot, certain embodiments extract snapshot features or determine enriched-contextual event data that is used to perform a mitigation action, generate a security mitigation score, or update an administrator portal activity log for an authorized administrator.

The invention relates to the technical field of radar remote sensing, discloses a high-order loose body historical activity detection method based on multi-source DEM data, and aims to solve the problem that the historical activity of a high-order loose body in a glacier coverage area is difficult to accurately and quantitatively evaluate in the prior art. The scheme mainly comprises the steps that drainage basin units are automatically divided based on DEM data, and glacier influence areas are screened; multi-source DEM data are fused, and system errors are eliminated through geographic positioning deviation correction, elevation distortion deviation correction and track mode deviation correction; establishing a penetration depth physical model, and correcting the radar penetration depth in the glacier area by using the C / X wave band elevation difference optimization parameter; calculating an elevation change rate by adopting robust regression; and calculating an activity comprehensive index by integrating multiple factors, and identifying a hot spot region and a time period through probability estimation and spatio-temporal clustering. According to the method, automatic and high-precision quantitative evaluation on the historical activity of the high-position loose body is realized, and the method is particularly suitable for geological disaster early warning and risk evaluation in high mountain areas.

Techniques are described herein that are capable of performing AI-based malicious activity detection using a threat actor profile. An alert is received. The alert indicates that a potentially anomalous event has occurred with regard to an entity. A profile of a threat actor is generated using information that describes behavior of the threat actor. An artificial intelligence (AI) model is triggered to determine whether the threat actor performs a malicious activity with regard to the entity by providing an AI prompt as an input to the AI model. The AI prompt includes the profile of the threat actor and a description of the potentially anomalous event. The AI prompt requests a determination whether the threat actor performs the malicious activity with regard to the entity.

A method for processing cross-modal question answering based on large model, an apparatus, and a storage medium are suggested, which relates to the field of artificial intelligence technologies such as speech interaction processing, large models, machine learning and natural language processing. The specific implementation includes: performing an activity detection on a target speech input by a user; in response to detecting a pause in the inputting of the target speech, obtaining a first text corresponding to a first input speech before the moment of the pause in the target speech; performing a text response processing using a pre-trained speech question answering processing system based on the first text and the first input speech.

In some embodiments, a fiber optic cable may be monitored while reducing false positives from disturbances in trusted areas. In some embodiments, a system includes a light source, a first remote termination unit in a trusted area, and a second remote termination unit in another trusted area. The light source emits light through the fiber optic cable, which passes through the first remote termination unit, an untrusted area, and then the second remote termination unit. The remote termination units reflect portions of the light back towards the light source. The system uses a prediction model to detect disturbances in the untrusted area based on the reflected light data.

Systems, methods, and techniques are directed to detecting potential anomalous activity based on changes in a security graph. In an example, a security system receives a first snapshot of a graph representative of a tenant account of a network-based system corresponding to a first timestamp. The security system receives a second snapshot of the graph corresponding to a second timestamp. The security system determines a first change in the graph based on the first and second snapshots and a second change related to the first change. The security system detects a potential anomaly based on the first and second changes. Responsive to detecting a potential anomaly, the security system causes a mitigation step to be performed with respect to the tenant account. In a further example, the security system determines relationships between a sequence of changes satisfies a cumulative anomaly criterion.

Techniques and apparatuses are described that perform audioplethysmography and motion-sensing data fusion. Fusing motion-sensing data with audioplethysmography expands the situations in which audioplethysmography can operate. In one aspect, audioplethysmography and motion-sensing data fusion (202) can be used for motion-artifact filtering (206). With motion-artifact filtering (206), noise caused by motion of a user (106) can be attenuated to improve sensitivity and accuracy for audioplethysmography. This performance improvement expands the ability of audioplethysmography to support use cases during situations in which the user (106) engages in an activity. In another aspect, audioplethysmography and motion-sensing data fusion (202) can expand a functionality of a hearable (102) to include activity detection (208) and / or activity classification (210). While activity detection (208) and / or activity classification (210) can provide additional contextual information for other use cases associated with audioplethysmography, either can also be used to control an operation of the hearable (102) and / or a computing device.

The invention relates to the field of electric digital data processing, and discloses a data processing method for dynamically updating a knowledge graph, which comprises the following steps of: presetting a basic weight for a knowledge unit, calculating an activity index of the knowledge unit attenuating along with time, and broadcasting an updating pulse signal to a neighbor unit when the activity index exceeds a threshold value, and the neighbor unit autonomously judges whether to trigger updating or continue to propagate the signal according to the signal attribute. According to the method, system resources are naturally focused on a high-value area through a life-driven self-organizing updating mechanism in the knowledge unit, global scanning redundancy is avoided while the knowledge timeliness is guaranteed, efficient transmission of key updating and reliable awakening of a dormant area are achieved by means of signal priority coupling and a neighborhood activity detection mechanism, and the reliability of the system is improved. And the resource efficiency contradiction in dynamic updating of the knowledge graph is solved.

The invention discloses a lightweight self-calibration time synchronization method for a LoRa sensor network, and belongs to the technical field of wireless communication in the Internet of Things (IoT). According to the method, a time synchronization beacon is broadcasted through a gateway TDMA time slice, and low communication overhead, low hardware complexity and high-precision synchronization are realized in combination with a channel activity detection (CAD) technology of a terminal and a dynamic calibration strategy. The method comprises the following core steps: synchronizing TDMA periodic broadcast time of a gateway, and configuring the number of lead codes; a terminal detects the active state of a channel through CAD, a progressive calibration strategy is adopted, a hardware RTC clock compensation value is dynamically adjusted, and steady-state oscillation caused by fixed calibration is avoided. The method is suitable for industrial monitoring, smart cities and other low-power-consumption scenes, and solves the problems of dependence on high-precision hardware and high communication overhead in the prior art.

Systems and methods of detecting activity at point of sale are provided. In one exemplary embodiment, a method is performed by a POS system having an optical sensor device, an optical scanner device and a load sensor device. The optical sensor device has a field of view that includes a first region about the POS system and the load sensor device is operable to measure a load of an object while positioned in a bagging area of the POS system. Further, the first region includes a second region associated with a certain object type that can be positioned in the bagging area without having to be scanned or weighed. The method includes obtaining a successive image captured by the optical sensor device to enable a determination that the target object corresponds to the certain object type that is exempt from being scanned or weighed based on the successive image.

The invention belongs to the technical field of image processing and computer vision, provides a lightweight grouping shuffling convolution personnel activity detection algorithm based on improved YOWOv3, and is applied to the field of real-time video behavior detection. According to the scheme, an original CSP residual module is reconstructed, and standard convolution is replaced by a dynamic grouping convolution and channel shuffling mechanism, so that the calculation complexity is remarkably reduced while the precision is ensured. The core innovation comprises the following steps: dividing an input channel into a plurality of subgroups for parallel processing, and reducing a parameter quantity to 1 / 4 of a standard convolution; feature interaction is realized through cross-group channel rearrangement, and the model expression ability is improved; convolution kernel parameters are multiplexed in the same network stage, memory occupation is further reduced, and efficient spatial-temporal feature fusion is realized in combination with a 3D convolution spatial-temporal detection head. According to the method, an edge calculation scene is adapted through lightweight design, real-time performance and detection precision are effectively balanced in a video analysis task, and an efficient solution is provided for dynamic behavior recognition in a low-resource environment.

The invention discloses an air-ground cooperative termite detection method and system based on multi-modal data fusion. The method comprises the following steps: S1, collecting a first termite detection data set through an air detection platform; s2, performing local detection on the soil and wall hidden area through a ground detection platform to construct a second termite detection data set; s3, performing preliminary preprocessing on the first termite detection data set and the second termite detection data set in a ground data processing system; s4, generating fusion termite detection feature data; s5, outputting the optimized fusion termite detection feature data; s6, generating a preliminary termite detection result; s7, sending the third termite detection data set to a ground data processing system; s8, generating updated fusion termite detection feature data; and S9, inputting the updated fusion termite detection feature data into the termite detection model, and generating a final termite detection result in combination with the preliminary termite detection result. According to the invention, the termite activity detection requirements in a large-range and complex environment can be better met.

The invention relates to a voice activity detection method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring audio sampling data, and extracting acoustic features from the audio sampling data; inputting the acoustic features into a convolutional neural network to obtain a first intermediate feature vector; inputting the first intermediate feature vector into a graph convolutional neural network for spatial relationship construction to obtain a second intermediate feature vector containing spatial relationship information among a plurality of feature channels; inputting the second intermediate feature vector into a multi-head attention mechanism network, and performing multi-head attention calculation based on rotation position coding on the second intermediate feature vector through the multi-head attention mechanism network to obtain a third intermediate feature vector; and performing voice activity detection according to the third intermediate feature vector. By adopting the method, the accuracy of voice activity detection can be improved.

Systems, methods, and other embodiments associated with quasi-supervised clustering for activity pattern characterization and anomalous activity detection are described. In one embodiment, a method generates a first sparse similarity matrix for nearest neighbors of a plurality of data points. The data points each characterize a pattern of activity associated with an account. The method generates a second sparse similarity matrix for random neighbors of the plurality of data points. The method recursively clusters the plurality of data points based on the first sparse similarity matrix. The method quasi-supervises the recursive clustering based on the second sparse similarity matrix to stop the iterative clustering when the data points are split into N clusters. The value of N is not pre-determined. The method detects that the individual data point has changed clusters, indicating anomalous activity. And, the method generates an electronic alert that the anomalous activity is associated with the account.

There is described a computer-implemented method to monitor operational performance of a system based on sensor data associated with the operation of the system. Received sensor data is processed to provide a plurality of input signals for an Artificial Neural Network, ANN. This processing involves transforming the sensor data into a plurality of components that are characteristic of the operation of the system and normalised with respect to activity values within the ANN, each component having a corresponding component value, and for each component, determining a difference value between the normalised component value and a respective feedback value from the ANN and generating an input signal representative of the difference value. By monitoring activity associated with the ANN, anomalous activity indicative of a departure from normal operation of the system is detected.

Systems and methods presented herein include guest activity detection devices configured to detect activity of guests of an amusement park, and to send data indicative of the detected activity; an artificial intelligence entity management system configured to analyze the data indicative of the activity of the guests, and to modify properties of guest-specific artificial intelligence entities based at least in part on the analyzed data indicative of the activity of the guests; and park features disposed within a physical environment of the amusement park and configured to be instructed by the guest-specific artificial intelligence entities to modify a guest experience for the guests of the amusement park in accordance with the properties of the guest-specific artificial intelligence entities.

A method for reducing voice noise, including: acquiring an algorithm activity detection result corresponding to a current audio frame to be processed by detecting the current audio frame using a predetermined voice activity detection algorithm; acquiring a target activity detection result corresponding to the current audio frame by merging a model activity detection result corresponding to a previous audio frame and the algorithm activity detection result corresponding to the current audio frame, wherein the model activity detection result is outputted by a predetermined voice noise reduction network model; acquiring an initial noise reduction audio frame by performing, based on the target activity detection result, noise estimation and noise elimination on the current audio frame; and outputting a target noise reduction audio frame and a model activity detection result corresponding to the current audio frame by inputting the initial noise reduction audio frame into the predetermined voice noise reduction network model.

The embodiment of the invention provides a sealed seed source automatic sorting system and device, and belongs to the technical field of seed source sorting. According to the method, a double-channel feeding module is used for synchronous vibration sorting and directional conveying of double-specification seed sources; the single seed source separating and feeding module is used for controlling alternate capture of double-specification seed sources and airflow jet type output of the seed sources through dynamic station switching of a switching plate; the activity detection module is used for measuring the activity of the seed source through the flow direction control of the Y-shaped inlet connecting pipe and the retention mechanism of the detection pipe; and the discharging automatic sub-packaging module is used for positioning the seed source collecting bottle and the material injection guide pipe in the target activity interval through the rotary storage disc to insert and release the seed source collecting bottle and the material injection guide pipe so as to carry out graded collection of the activity of the seed source. Through a double-channel parallel feeding mechanism, synchronous treatment and dynamic adaptation of seed sources of different specifications are achieved, and the overall productivity is remarkably improved. And the box body design and the automatic split charging process are combined, so that the safety of the working environment is effectively guaranteed.

The invention relates to the field of audio processing, and discloses a human voice activity detection method and device, computer equipment and a storage medium. The method comprises the following steps: collecting audio and video data of a user, and dividing the audio and video data into a video stream and an audio stream; detecting the video stream by using a FaceMesh model to obtain a lip opening degree and a head attitude angle, and determining a lip motion state according to the lip opening degree and the head attitude angle; blocking the audio stream to obtain a plurality of audio blocks, performing noise reduction on each audio block to obtain a plurality of noise-reduced audio blocks, and detecting each noise-reduced audio block by using a silhoVAD model to obtain a human voice activity probability and an audio cache queue; and inputting the lip motion state and the human voice activity probability into a state machine, introducing an audio cache queue by the state machine, and outputting a speaking identifier and a corresponding audio clip. Speaking recognition is cooperatively completed in combination with machine vision and hearing signals, and the accuracy of whole human voice activity detection is improved.

The invention discloses an agricultural environment perception and robot-assisted picking method based on vision, and the method employs an end-to-end detection, tracking and classification frame, comprises five independent modules: object detection and tracking, feature extraction, activity recognition, target position estimation, and a robot scheduler, obtains information through employing an RGB camera and a two-dimensional laser radar, and carries out the detection and classification of the information. By customizing long and short-term memory network processing data, compared with other methods, an end-to-end architecture adopted by the method can enable the system to comprehensively process the whole process from activity detection of a picker to auxiliary operation of a robot, and adopted multi-camera information fusion can accurately estimate the positions of the picker in different distance ranges, so that the accuracy of picking is improved. And the behavior activity of the picker can be accurately predicted by adopting the customized long-short-term memory network.

The invention relates to a Java background service deployment method and system based on double-node dynamic routing, and the method comprises the steps: obtaining the health state of each node based on TCP port activity detection and / or UDP heartbeat packet detection; judging the nodes of which the detection failure times exceed the preset times as fault nodes, and setting the fault nodes to be in an offline state; determining a traffic allocation weight according to the real-time working state of the node; performing routing distribution on the client request based on the traffic distribution weight; route distribution is dynamically adjusted based on the traffic allocation weight, the resource utilization rate of the nodes is improved, and the situation that the load of a single node is too heavy is avoided; when a fault node occurs, the other node is automatically switched to respond to the request, so that the fault switching time is shortened; and the data consistency between the double nodes is adjusted based on the Redis cluster and the MySql cluster. Based on cooperation of the steps, the problems of failure switching delay, low resource utilization and low data consistency in a traditional scheme are solved.