160 results about "Streaming algorithm" patented technology

A falling detection method and an electronic system using the same are provided. The falling detection method includes: obtaining a video stream, and performing recognition on a person in an image of the video stream to obtain at least one human body feature; calculating based on the human body feature to obtain at least one falling related feature; calculating at least one movement trend in a plurality of directions of the person in the video stream by using an optical flow algorithm; inputting the falling related feature and the movement trend to a falling classifier, such that the falling classifier generates an output result indicating whether a falling event happens to the person; and sending an alarm according to the output result.

The invention discloses a light stream based vehicle motion state estimating method which is applicable to estimating motion of vehicles running of flat bituminous pavement at low speed in the road traffic environment. The light stream based vehicle motion state estimating method includes mounting a high-precision overlook monocular video camera at the center of a rear axle of a vehicle, and acquiring video camera parameters by means of calibration algorithm; preprocessing acquired image sequence by histogram equalization so as to highlight angular point characteristics of the bituminous pavement, and reducing adverse affection caused by pavement conditions and light variation; detecting the angular point characteristics of the pavement in real time by adopting efficient Harris angular point detection algorithm; performing angular point matching tracking of a front frame and a rear frame according to the Lucas-Kanade light stream algorithm, further optimizing matched angular points by RANSAC (random sample consensus) algorithm and acquiring more accurate light stream information; and finally, restructuring real-time motion parameters of the vehicle such as longitudinal velocity, transverse velocity and side slip angle under a vehicle carrier coordinate system, and accordingly, realizing high-precision vehicle ground motion state estimation.

The subject invention provides for a system and method that facilitates concurrent data streaming. In particular, the invention involves initiating a first operation from the optical media at time tx and initiating at least a second operation from the optical media at time ty while the first operation is currently in progress, wherein tx≠ty. The first operation includes reading a real-time data stream to a first buffer. The second operation includes one of reading a real-time data stream and a non-real-time data stream to at least a second buffer. Furthermore, a utility-based analysis can be performed to determine whether to access the first buffer rather than to access the surface of the optical media in order to conduct the second operation. Moreover, the first and at least second operations can be performed in parallel.

A distinct-count estimate is obtained in a guaranteed small footprint using a two level hash, distinct count sketch. A first hash fills the first-level hash buckets with an exponentially decreasing number of data-elements. These are then uniformly hashed to an array of second-level-hash tables, and have an associated total-element counter and bit-location counters. These counters are used to identify singletons and so provide a distinct-sample and a distinct-count. An estimate of the total distinct-count is obtained by dividing by the distinct-count by the probability of mapping a data-element to that bucket. An estimate of the total distinct-source frequencies of destination address can be found in a similar fashion. By further associating the distinct-count sketch with a list of singletons, a total singleton count and a heap containing the destination addresses ordered by their distinct-source frequencies, a tracking distinct-count sketch may be formed that has considerably improved query time.

A browsing interface for browsing Internet information that organizes information downloaded from the Internet in a page-based manner, complete with a functional thickness representation. A number of insertion methods are available for insertion of newly downloaded Internet information. A browsing data file contains information and parameters which control the display of information on the browsing interface. A streaming algorithm pre-downloads data based on predicted user data requests to minimize interruption of the browsing process.

The invention provides a mobile group intelligent perception excitation method with minimized payment as an object. The method solves a problem of minimized payment on the premise that all tasks are completed in a mobile group intelligent perception application, so as to save perception cost for an employer. The method comprises: firstly, providing a group intelligent perception reverse auction model based on time window, providing a polynomial time excitation algorithm based on the model, the algorithm including two steps of primary selection and weight competition, in the primary selection step, using a minimum-cost maximum-flow algorithm to obtain two groups of non-intersected users, in the weight competition step, using a graph theory to find substituted user groups, and according to cost with weight, selecting final selected users, and calculating remuneration of each user. Compared with other excitation mechanisms, the method has lower total amount paid, and is still individually rational and anti-cheating.

The invention relates to an automatic load distribution method for an intelligent transformer substation. The automatic load distribution method includes performing load flow calculation of the whole network according to network and load data acquired currently by the intelligent transformer substation of a power distribution network; by a reactive power optimization module, on given constraint conditions, solving the optimal solution of an objective function by a search algorithm to obtain schemes on in and out of a reactive power compensation device and transformer tap adjustment; by a power distribution network reconstruction module, on given constraint conditions, solving the optimal solution of the objective function by a virtual stream algorithm to obtain schemes on combined operation of states of network section switches and communication switches; by an interactive computation module, interacting the two sub problems including reactive power optimization and power distribution network reconstruction to approach the optimal solution gradually so as to realize automatic load distribution with the purpose of minimizing power loss. By the automatic load distribution method, power loss of lines is reduced, running of a power grid is optimized effectively, network loss is reduced, analysis speed of the power grid and control reliability are improved, and safety, reliability and economization in running of the power grid is guaranteed. In addition, the automatic load distribution method can be applied to automatic load distribution process of the intelligent transformer substations widely.

The invention relates to streaming algorithms useful for obtaining summaries over unaggregated packet streams and for providing unbiased estimators for characteristics, such as, the amount of traffic that belongs to a specified subpopulation of flows. Packets are sampled from a packet stream and aggregated into flows and counted by implementation of Adaptive Sample-and-Hold (ASH) or Adaptive NetFlow (ANF), adjusting the sampling rate based on a quantity of flows to obtain a sketch having a predetermined size, the sampling rate being adjusted in steps; and transferring the count of aggregated packets from SRAM to DRAM and initializing the count in SRAM following adjustment of the sampling rate.

The invention belongs to the technical field of computer science, and discloses an online public opinion text information emotion polarity classification processing system and method, the online public opinion text emotion polarity is widely applied to a public opinion monitoring system, however, a feature engineering extraction module of a traditional machine learning method is large in text information loss, and the accuracy of a classification model is not high enough. The method comprises the steps of preprocessing data; the method comprises the following steps of: constructing a word vector in a way of pre-training a model fin-tuning through BERT; the BERT model calculates the correlation between the characters in the sentence and each of the other characters; the constructed word vector can better solve the problems of'one-word polysemy 'and'synonym' of Chinese; the loss of word vector representation is greatly reduced; in the classification model, firstly Bi-LSTM is used for effectively learning context information, then Attention is used for capturing main semantic information and effectively filtering valuable public opinion information, finally softmax classification is used, and the performance of an obtained public opinion text emotion polarity classification result is better than that of a current mainstream algorithm.

Displaying requested data from a database includes providing a display containing a scroll bar with a scroll tab therein, determining a total amount of the requested data, sizing the scroll tab in the scroll bar according to the total amount of the requested data and according to an amount of data displayable in the display, and fetching a predetermined amount of data from a first portion of the database. The predetermined amount may be independent of the size of the display. The predetermined amount may be less than the total amount of the requested data. Sizing the scroll tab may include choosing a size that indicates to a user that all of the requested data is being fetched from the database and / or may include choosing a size that is inversely proportional to the total amount of the request data in response to the total amount of the requested data being less than a predetermined value. Choosing a size may include choosing a predetermined size in response to the total amount of the requested data being greater than the predetermined value.

The invention belongs to the technical field of a random fuzzy power flow algorithm for distributed wind power, photovoltaic power generation and other uncertain energy sources connected to a power system, and discloses a power flow algorithm considering the fact that the power of the uncertain energy sources connected to the power system has random fuzzy characteristics. The load of the uncertain energy sources connected to the power system is taken as a random fuzzy variable, and the node load power is randomly simulated in a fuzzy manner; the node load power is embedded to Newton-Raphson power flow calculation to obtain voltage amplitude values and phase angle data of the corresponding nodes of the system; the probability distribution characteristics of the node voltage amplitude values and the phase angles are subjected to extraction and statistics; a probability distribution model suitable for fitting the node voltage amplitude values and the phase angles, and the parameter fuzzy characteristics are analyzed and determined; and the random fuzzy model for the node voltages and phase angles is established. According to the power flow algorithm, the influences on the node voltages of the power distribution network from the uncertainties of the distributed type power supply outputs can be more comprehensively analyzed, so that corresponding guiding evidences can be provided for power generation plan arrangement and dispatching for a large number of distributed wind power, photovoltaic power generation and other uncertain energy sources connected to the power system in the future.

The invention discloses a lactating sow posture recognition method based on double-flow RGB-D Fast R-CNN. The end-to-end double-flow RGB-D Faster R-CNN algorithm fusing RGB-D image features in the feature extraction stage is provided and used for recognizing five postures of standing, sitting, prone lying, abdominal lying and lateral lying of lactating sows in a free fence sow scene. Based on theFaster R-CNN, firstly, two CNN networks are used for extracting RGB image features and depth image features respectively; Generating an interested area of the RGB image feature map and the depth imagefeature map only by adopting one RPN network by utilizing a mapping relation of the RGB-D image; After pooling the features of the region of interest, realizing splicing fusion of RGB-D features by using an independent network layer; And finally, in the Fast R-CNN stage, introducing an NOC structure to continue to perform convolution extraction on the fused features, and then sending the featuresinto a classifier and a regression device. The invention provides a high-precision and small model fused with RGB-D data information end-to-end and a real-time sow posture recognition method, and a foundation is laid for further analysis of sow behaviors.

The invention discloses a multi-label-based lightweight rapid crowd counting method. A simple and efficient trunk feature extraction network is designed according to the size of a receptive field, anda dense context module is arranged in the trunk feature extraction network, so that information transmission of a network layer is ensured, and the network expression capability is improved; six multi-scale intermediate supervision branches are designed, so that the network can be converged more quickly and stably; an up-sampling module is designed, the resolution is improved step by step, and the quality of a density map is improved, so that accurate counting and accurate positioning are realized; three labels are designed, a crowd counting task based on density is explicitly converted intoa foreground and background segmentation task to assist a regression task of a crowd density map, prediction of the density map and the segmentation map is achieved at the same time, and estimation errors are effectively reduced. Test results of UCF _ CC _ 50, ShanghaiTeck and UCF-QNFR data sets show that the prediction performance of the method is superior to that of a current mainstream algorithm, the prediction speed reaches real time, and the method can be conveniently deployed in terminal equipment.

The invention provides a task allocation method, device and apparatus based on a maximum flow algorithm. The task allocation method comprises the steps of: acquiring work subject information and to-be-allocated task information; building a capacity network according to the work subject information and the to-be-allocated task information, wherein the weight of each edge in the capacity network isdetermined by the processing efficiency of processing of different types of tasks by the work subject; solving all task allocation routes from source nodes to sink nodes in the capacity network by using a preset maximum flow algorithm; allocating the to-be-allocated tasks according to the task allocation routes. According to the invention, the weight of each edge in the capacity network is obtained based on the processing efficiency of the to-be-allocated tasks and / or value generated by the working subject, so that the processing efficiency of the to-be-allocated tasks or the value generated by the working subject can achieve the overall optimization state in final results.

The invention belongs to the field of computer vision. The invention particularly relates to complex backgrounds, shielding and other problems. The invention discloses a target tracking method based on discriminant dictionary learning. Firstly, a target and a background sample are obtained according to the local correlation of the target in time and space; secondly, a dictionary learning model isestablished based on sparse representation, an error item is used for capturing abnormal values generated by shielding and the like, a non-convex MCP function is used for punishing a sparse coding matrix and an error matrix, and inconsistent constraint items are applied to the dictionary so as to improve dictionary robustness and discrimination; mM-IALM optimization method is used for solving theproposed non-convex dictionary learning model so as to obtain better convergence; And calculating a candidate target reconstruction error from the obtained dictionary to construct a target observationmodel, and realizing accurate tracking of the target based on a Bayesian reasoning framework. Simulation results show that compared with an existing mainstream algorithm, the method has higher tracking precision and robustness under the environments of illumination change, scale change, shielding, background clutter and the like.

The invention discloses a video copy and paste blind detection method based on a dense scale-invariant feature transform (SIFT) stream. The video copy and paste blind detection method includes the following steps: step A, extracting key frames of video to be detected and non-key frames corresponding to each key frame; step B, confirming a copy area and a paste area of each key frame extracted in the step A so as to obtain a copy and paste detection result picture of each key frame; and step C, using a dense SIFT stream algorithm to obtain a copy and paste detection result of the video to be detected according to a copy and paste detection result of each key frame. Compared with the prior art, the video copy and paste blind detection method based on the dense SIFT stream is low in algorithm complexity, can accurately detect a copy and paste forgery area in the video, and has good robustness for rigid body target detection.

The invention discloses a no-reference image quality evaluation method based on a self-attention mechanism GAN network, and particularly relates to the field of image processing, which is composed ofthree parts: a generation network, a discrimination network and a quality evaluation network. The generation network performs feature extraction and recovery reconstruction on an input distorted image, and the discrimination network distinguishes the reconstructed image from an undistorted image as much as possible. According to the method, the adversarial learning intensity is enhanced by addingthe self-attention module and improving the model structure, and a more reliable simulation'reference graph' is output; experiments are carried out on LIVE and TID2013 data sets, results show that theSARAN algorithm provided by the invention is superior to a current mainstream algorithm in overall objective evaluation of quality evaluation of non-reference distorted images, has good performance for different distortion types, and shows that the SARAN algorithm has high correlation with a subjective evaluation result, and better meets the perception consistency of a visual perception system (HVS) on the image quality.

The invention relates to a method, a device and equipment for optimizing the analysis performance of an intelligent video, and the method comprises the steps: (1) carrying out a reference piperine test on a video file for the acceleration of an offline video file, and setting an optimal file slice number; slicing the video file, and issuing a slicing task to the GPU; calling a GPU to decode the slice file, and calling back a decoding result to an algorithm directly through a video memory address, and reducing the performance loss without the video memory-main memory copy, wherein the video analysis algorithm takes the decoded video memory address, calls a GPU for algorithm acceleration and outputs an analysis result; (2) optimizing and expanding the number of paths for real-time video stream algorithm analysis; and calling the GPU to decode each path of real-time video, calling back a decoding result to the algorithm directly through a video memory address, setting double caches by analgorithm end, storing decoded data in multiple paths, transmitting the decoded data to the algorithm for GPU batch processing, and switching the two cache functions after batch processing is completed to achieve the purpose of minimum system delay.

The invention discloses a rapid inter-frame mode selection method for an AVS (Advanced Audio Video Coding Standard) coder. The method comprises the following steps of: preselecting an optimal mode from a 16*16 mode, a 16*8 mode, a 8*16 mode and a 8*8 mode by virtue of a vision perception judging model and pixel point edge information; breaking through the dependence on data of interframe mode selection by virtue of a rapid interframe mode selection algorithm facilitating the realization of hardware to ensure that the interframe mode selection does not need to wait the relevant reconstruction data and becomes the premise for realizing an efficient interruption-free mode selection stream line; designing 5-level stream algorithm which can be realized oriented at the hardware and is capable of efficiently calculating rate distortion cost values to allow the application and popularization of the practical hardware for the interframe mode selection on the basis of rate distortion cost values to be possible; and finally, determining an optimal mode according to cost values of three candidate modes including a preset optimal mode, a direct mode and an intra mode. The invention also discloses an AVS interframe mode selection device. According to the invention, the coding property of the hardware coder is remarkably improved.

The invention discloses a crowd safety degree evaluation method based on crowd density and pedestrian speed, which comprises the following contents: adopting a neural network model regression to predict the crowd density by using a feature vector composed of a foreground area and a perimeter of the obtained video frame image; based on an LK optical flow improved pyramid Lucas-Kanade optical flow algorithm, processing continuous image frames to obtain the pedestrian velocity, i.e. the average velocity of individual pedestrians; constructing a fuzzy inference system to evaluate the crowd safetygrade: selecting evaluation factors and determining its membership function; establishing crowd attributes and a safety risk rule base, and carrying on fuzzy reasoning on the input according to the rule base; defuzzifying the reasoning results, and then evaluating the risk level of crowd safety. The invention establishes a mathematical model for crowd density, pedestrian speed and output crowd safety risk grade, and adopts a fuzzy inference method to determine the influence of each element of crowd attributes on the early warning grade, which is reasonable in design and simple in operation, and improves practical applicability.

The invention firstly provides a laser radar SLAM algorithm-CPFG (Closet Probability and Feature Grid) algorithm based on a feature probability grid map. The algorithm utilizes three-dimensional laser radar data to create and update a grid map of line, surface and Gaussian distribution characteristics and occupancy probability in real time, and combines a robust mahalanobis distance as an optimization function to carry out real-time pose estimation. The algorithm mainly comprises three steps of point cloud preprocessing, matching of the point cloud and the characteristic probability grid map and pose estimation, and updating of the feature probability grid map. Compared with several current mainstream algorithms, the laser radar SLAM algorithm has better performance in the aspects of real-time performance and positioning precision. Afterwards, attitude information of inertial navigation is fused, the high displacement precision of the laser radar SLAM and the low attitude drift characteristic of inertial navigation are combined, the relative positioning precision of the method can reach about one thousandth, and the method has wide application prospects in the field of unmanned driving positioning.

The invention discloses a station area grid theoretical line loss calculation method giving consideration to line impedance and load uncertainty based on a fuzzy flow algorithm, and the method comprises the steps: firstly inputting a network structure, impedance membership degree information and load membership degree information, and carrying out the parameter initialization; secondly solving the value range of each fuzzy variable at a membership degree of zero through an alpha-cut set method; thirdly solving the maximum value of network loss in the value range through a genetic algorithm; fourthly solving the minimum value of network loss in the value range; fifthly substituting the value of the variable at the membership degree of one, and calculating the corresponding network power loss through a three-phase flow algorithm; finally obtaining the possibility distribution of network loss through the solved maximum and minimum values of network loss at all membership degrees.

The embodiment of the invention discloses a data processing method and device, a server and a storage medium, and the method comprises the steps of obtaining initial medical data from at least one hospital information system, the initial medical data comprising medical sub-data of at least two dimensions; on the basis of different dimensions corresponding to the medical sub-data, adopting a distributed message publishing and subscribing system to carry out shunting processing on the initial medical data, and obtaining multiple pieces of branch data; analyzing and processing each piece of branch data through a streaming computing platform to obtain a plurality of pieces of analyzed data; associating and storing each of the analytic data in an object- In the relational database, associatingall the stored analysis data to form target medical data corresponding to the initial medical data; when a query request sent by a client is received, adopting a streaming algorithm to carry out object-object query on the client; and searching the target medical data in the relational database, and sending the target medical data to the client, so that the processing efficiency of the medical datacan be improved.

A distinct-count estimate is obtained in a guaranteed small footprint using a two level hash, distinct count sketch. A first hash fills the first-level hash buckets with an exponentially decreasing number of data-elements. These are then uniformly hashed to an array of second-level-hash tables, and have an associated total-element counter and bit-location counters. These counters are used to identify singletons and so provide a distinct-sample and a distinct-count. An estimate of the total distinct-count is obtained by dividing by the distinct-count by the probability of mapping a data-element to that bucket. An estimate of the total distinct-source frequencies of destination address can be found in a similar fashion. By further associating the distinct-count sketch with a list of singletons, a total singleton count and a heap containing the destination addresses ordered by their distinct-source frequencies, a tracking distinct-count sketch may be formed that has considerably improved query time.

The invention provides a multi-target traffic behavior rapid extraction method based on video recognition. The method comprises the following steps: foreground target extraction: training a part of images of a to-be-processed video by using a Gaussian mixture model to obtain an initial background, training the initial background by using a visual background extraction method, further extracting acontour of a target in the foreground by using a background difference method according to the trained initial background, and performing filtering and morphological processing on the contour; contourtarget tracking: extracting feature points of each target in the to-be-processed video by using an angular point detector, and tracking each target by using a KLT optical flow algorithm to generate acorresponding image track; contour track post-processing: performing separation correction and smoothing processing on the image track of each target, and converting the image coordinates of each target in each frame of the to-be-processed video into real coordinates according to a perspective transformation principle; and result output, obtaining the corresponding traffic parameters including the speed, and judging the category to which the target belongs.

The invention proposes an improvement to increase the speed at which a Trickle algorithm can propagate update information in a multi-hop network, while still controlling the number of redundant transmissions and ensuring scalability of the algorithm. It does this by intelligently adapting the size of a listen-only period (t1, t2), depending on the fact if a node has recently received an update or not.

The invention relates to the technical field of computer vision, in particular to an embedded visual computing system for face recognition, counting and temperature measurement. After edge computing equipment is started, the system is started to run; a FaceSDK / BodySDK algorithm module is initialized; a camera stream acquisition module acquires a video stream from predefined camera information; a video stream is pushed to a video stream algorithm processing module in a queue form by a single frame and is analyzed according to a loading algorithm; and the analyzed structured result is processed according to the business requirement and then presented to the user in a visual result form. According to the invention, the defects in the prior art are overcome, and the service scenes of employee sign-in, guest greeting, frequent hall approaching and the like are solved through face comparison by setting applications of face recognition, guest greeting, frequent hall approaching and the like based on face detection and face comparison algorithm derivation; and an AIOT framework based on the intelligent edge calculation is adopted, so the purpose of deploying the central server in an internet end or a far-end machine room can be achieved.

The invention relates to a real-time video stream analysis acceleration method, a device and apparatus. The method aims at path number optimization and expansion of real-time video stream algorithm analysis, and the GPU is called to decode each path of real-time video, a decoding result is called back to the algorithm directly through a video memory address, double caches are set by an algorithm end, storing decoded data in multiple paths, transmitting the decoded data to the algorithm for GPU batch processing, and switching the two cache functions after batch processing is completed to achieve the purpose of minimum system delay.

The invention discloses an A3C-SRU-based intelligent vehicle traffic flow converging method and system. The implementation method comprises the following steps of 1, adopting environmental parametersand vehicle parameters by devices such as a digital camera, a multi-line laser radar, a millimeter-wave radar and a gps positioning system; 2, establishing a simulation environment platform by utilizing simulation software according to the environment parameters and the vehicle parameters extracted in the step ; 3, setting parameters and constraint conditions of a reinforcement learning algorithmaccording to the simulation environment in the step 2; 4, training by using an A3C-SRU algorithm according to the simulation environment constructed in the step 2 to obtain a decision of an imported traffic flow scene; and 5, obtaining the optimal action sequence obtained in the step 2 according to the model obtained in the step 4, storing the trained model, and inputting the model into the intelligent vehicle to realize a traffic flow importing task. According to the A3C-SRU-based intelligent vehicle afflux traffic flow algorithm of the invention, real-time afflux traffic flow tasks can be effectively realized according to the settings of the steps 1-5.

The invention discloses an IO disk falling method and device for a storage system and the storage system. The IO disk falling method comprises the steps: obtaining the maximum IO issuing rate according to the used cache proportion of a cache; obtaining the number of fragments according to the IO maximum issuing rate and a preset fragment size, and generating fragments with the number of fragmentsin a cache; storing each piece of received IO data to a respective corresponding fragment; and enabling the fragments meeting the preset disc falling condition to be subjected to disc falling. According to the IO disk falling method, IO data on each fragment form a large IO for disk falling; the IO merging process is simplified; and the delay of the drop time delay of the batch drop IO data is reduced, and the consistency of the upper limit of the actual rate and the upper limit of the rate of the leaky bucket current limiting algorithm is ensured on the basis of meeting the requirement of full flash storage of the batch drop IO data, so that the IO drop can be accurately controlled by subsequently utilizing the drop time delay and the dynamic regulation rate of the fragments.