189 results about "Stream network" patented technology

Methods for displaying results of a spreading activation algorithm and for defining an activation input vector for the spreading activation algorithm are disclosed. A planar disk tree is used to represent the generalized graph structure being modeled in a spreading activation algorithm. Activation bars on some or all nodes of the planar disk tree in the dimension perpendicular to the disk tree encode the final activation level resulting at the end of N iterations of the spreading activation algorithm. The number of nodes for which activation bars are displayed may be a predetermined number, a predetermine fraction of all nodes, or a determined by a predetermined activation level threshold. The final activation levels resulting from activation spread through more than one flow network corresponding to the same generalized graph are displayed as color encoded segments on the activation bars. Content, usage, topology, or recommendation flow networks may be used for spreading activation. The difference between spreading activation through different flow networks corresponding to the same generalized graph may be displayed by subtracting the resulting activation patterns from each network and displaying the difference. The spreading activation input vector is determined by continually measuring the dwell time that the user's cursor spends on a displayed node. Activation vectors at various intermediate steps of the N-step spreading activation algorithm are color encoded onto nodes of disk trees within time tubes. The activation input vector and the activation vectors resulting from all N steps are displayed in a time tube having N+1 planar disk trees. Alternatively, a periodic subset of all N activation vectors are displayed, or a subset showing planar disk trees representing large changes in activation levels or phase shifts are displayed while planar disk trees representing smaller changes in activation levels are not displayed.

In some embodiments, a method and apparatus for automatically parallelizing a sequential network application through pipeline transformation are described. In one embodiment, the method includes the configuration of a network processor into a D-stage processor pipeline. Once configured, a sequential network application program is transformed into D-pipeline stages. Once transformed, the D-pipeline stages are executed in parallel within the D-stage processor pipeline. In one embodiment, transformation of a sequential application program is performed by modeling the sequential network program as a flow network model and selecting from the flow network model into a plurality of preliminary pipeline stages. Other embodiments are described and claimed.

A distributed incentive mechanism is provided for peer-to-peer (P2P) streaming networks, such as mesh-pull P2P live streaming networks. Video (or audio) may be encoded into multiple sub-streams such as layered coding and multiple description coding. The system is heterogeneous with peers having different uplink bandwidths. Peers that upload more data (to a peer) receive more substreams (from that peer) and consequently better video quality. Unlike previous approaches in which each peer receives the same video quality no matter how much bandwidth it contributes to the system, differentiated video quality, commensurate with a peer's contribution to other peers, is provided, thereby discouraging free-riders.

The invention belongs to the computer vision and mode identification field, particularly relates to a cross-view gait identification device based on the dual-flow generation confrontation network anda training method and aims to solve a problem of non-high cross-view gait identification accuracy. The method specifically comprises steps that a global flow gait energy image of a standard angle is learned through a global flow generation confrontation network model, and local flow gait energy images of standard angles are learned through utilizing three local flow generation confrontation network models. Global gait characteristics can be learned by a global flow model, on the basis of the global flow model, the local flow networks are added, so local gait characteristics can be learned; gait details can be restored through adding pixel-level constraints to a generator of the dual-flow generation confrontation network; through fusing the global gait characteristics and the local gait characteristics, gait identification accuracy can be improved. The method is advantaged in that quite strong robustness for the gait images is realized, and a cross-view gait identification problem can be better solved.

The invention discloses an electromyographic signal gesture recognition method based on a double-flow network. The electromyographic signal gesture recognition method comprises the following steps: 1)collecting electromyographic signals of various gestures of multiple persons, wherein each gesture action of a subject lasts for 12 seconds by wearing a 16-channel acquisition device, steady-state data of 10 seconds are extracted, data preprocessing is carried out, a 300ms time window is selected, and the size of each frame of electromyogram is 300*16, so that a training set is constructed. 2) constructing a double-flow network model, wherein the model is mainly composed of three parts, the first part is a multi-layer CNN and is responsible for extracting spatial features; the second part isa multi-layer LSTM and is responsible for learning time characteristics; and the last part is a feature merging layer which is responsible for feature fusion. 3) training the double-flow network model, and performing gradient descent optimization by adopting an Adam optimizer until convergence, and 4) performing gesture recognition on the sEMG of the arm by using the trained double-flow network model.

The invention provides a behavior recognition method and system based on an attention mechanism double-flow network, and belongs to the technical field of behavior recognition, and the method comprises the steps: dividing an obtained whole video segment into a plurality of video segments with the same length, extracting an RGB image and an optical flow gray-scale image of each frame of each videosegment, and carrying out the preprocessing of the RGB images and the optical flow gray-scale images; carrying out random sampling on the preprocessed image to obtain an RGB image and an optical flowgrayscale image of each video clip; extracting appearance features and time dynamic features of the sampled images by using a double-flow network model introducing an attention mechanism, fusing the appearance features and the time dynamic features according to the types of a time domain network and a space domain network respectively, and performing weighted fusion on a fusion result of the timedomain network and a fusion result of the space domain network to obtain an identification result of the whole video. According to the invention, the video data can be fully utilized, the local key features of the video frame image can be better extracted, the foreground area where the action occurs is highlighted, the influence of irrelevant information in the background environment is inhibited,and the behavior recognition accuracy is improved.

The invention provides a point cloud scene segmentation method based on knowledge distillation and semantic fusion. The method comprises the steps: constructing a double-flow network frame which comprises dense local branches and sparse global branches; wherein the input of the dense local branch is the local region dense point cloud in the global scene, and the input of the sparse global branch is the sampled global scene point cloud. Secondly, designing a distillation module based on irregular data, performing knowledge distillation by using an Euclidean distance and an adversarial learningloss function, and transmitting local dense detail information to a sparse global branch; and finally, a dynamic graph context semantic information fusion module is designed, and the global features and the local features after detail information enhancement are fused. According to the method, rich detail information of a local scene and rich context semantic information of a global scene are fully and complementarily utilized. Meanwhile, increase of the calculation amount is avoided, and the point cloud segmentation result of a large-scale indoor scene can be effectively improved.

At least three uses of the technology disclosed are immediately recognized. First, a video stream classifier can be trained that has multiple uses. Second, a trained video stream classifier can be applied to monitor a live network. It can be extended by the network provider to customer relations management or to controlling video bandwidth. Third, a trained video stream classifier can be used to infer bit rate switching of codecs used by video sources and content providers. Bit rate switching and resulting video quality scores can be used to balance network loads and to balance quality of experience for users, across video sources. Balancing based on bit rate switching and resulting video quality scores also can be used when resolving network contention.

The invention discloses a video human body behavior recognition method and system based on a multi-mode double-flow 3D network. The method comprises the steps of generating a depth dynamic image sequence DDIS based on a depth video; generating a posture evaluation graph sequence PEMS based on the RGB video; respectively inputting the depth dynamic graph sequence and the posture evaluation graph sequence into a 3D convolutional neural network, constructing a DDIS flow and a PEMS flow, and obtaining respective classification results; and fusing the obtained classification results to obtain a final behavior recognition result. The method has the beneficial effects that the DDIS can well describe the human body motion and the contour of the interactive object in the long-term behavior video by modeling the local space-time structure information of the video. The PEMS can clearly capture the change of the human body posture and eliminate the interference of the messy background. The multi-mode double-flow 3D network architecture can effectively model the global space-time dynamic of the behavior video in different data modes, and has excellent recognition performance.

The invention provides an image fine-grain identification method based on multi-stream multi-scale cross bilinear features. Aiming at the problems of insufficient extraction and insufficient feature utilization of fine-grain features of the image, the image fine-grain identification method utilizes a multi-stream network to extract cross bilinear features, and the features can represent finer local features of the image, thus solving the problem of insufficient feature extraction. The image fine-grain identification method solves the problem of insufficient feature utilization by using a method for enhancing and fusing multi-scale bottom layer bilinear features through image random mixing. Experimental verification shows that the fine-grain identification method based on multi-stream network fused multi-scale cross bilinear features provided by the invention remarkably improves the identification accuracy on a CUB-200-2011 public data set, compared with an existing method, and achievesthe optimal fine-grain identification accuracy.

The invention discloses a behavior recognition method based on a space-time attention enhancement feature fusion network, and belongs to the field of machine vision. According to the method, a networkarchitecture based on an appearance flow and motion flow double-flow network is adopted, and is called as a space-time attention enhancement feature fusion network. Aiming at a traditional double-flow network, simple feature or score fusion is adopted for different branches, an attention-enhanced multi-layer feature fusion flow is constructed to serve as a third branch to supplement a double-flowstructure. Meanwhile, aiming at the problem that the traditional deep network neglects modeling of the channel characteristics and cannot fully utilize the mutual relation between the channels, the channel attention modules of different levels are introduced to establish the mutual relation between the channels to enhance the expression capability of the channel characteristics. In addition, thetime sequence information plays an important role in segmentation fusion, and the representativeness of important time sequence features is enhanced by performing time sequence modeling on the frame sequence. Finally, the classification scores of different branches are subjected to weighted fusion.

The invention discloses an RGB-D multi-modal fusion person detection method based on an asymmetric double-flow network, and belongs to the field of computer vision and image processing. The method comprises the steps of RGBD image acquisition, depth image preprocessing, RGB feature extraction and Depth feature extraction, RGB multi-scale fusion and Depth multi-scale fusion, multi-modal feature channel reweighting and multi-scale personnel prediction. According to the method, an asymmetric RGBD double-flow convolutional neural network model is designed to solve the problem that a traditional symmetric RGBD double-flow network is prone to causing depth feature loss. Multi-scale fusion structures are designed for RGBD double-flow networks respectively, so multi-scale information complementation is achieved. A multi-modal reweighting structure is constructed, the RGB and Depth feature maps are combined, and weighted assignment is performed on each combined feature channel to realize modelautomatic learning contribution proportion. Person classification and frame regression are carried out by using multi-modal features, so the accuracy of personnel detection is improved while the real-time performance is ensured, and the robustness of detection under low illumination at night and personnel shielding is enhanced.

The invention provides an in-band network telemetry bearer flow selection method and system, and the method comprises the steps: obtaining a network state telemetry demand; acquiring network topologyinformation of a to-be-tested network and distribution information of user service flows, wherein the distribution information of the user service flows comprises average length of data packets of each user service flow and paths flowing through a switch; determining a target switch and telemetry metadata needing to be collected by the target switch according to a network state telemetry requirement; and according to the network topology information, the target switch, the telemetry metadata needing to be collected by each target switch, the average length of the data packets of each user service flow, and the path of each user service flow passing through the switch, selecting at least one user service flow as a telemetry bearer flow, the telemetry bearer flow covering all the target switches. According to the user service flow selected by the in-band network telemetry bearer flow selection method provided by the invention, a network administrator can acquire the required network state in time.

The invention discloses an action detection method based on asymmetric multi-stream, which comprises the following steps: extracting an RGB image and an optical stream from a prior video, and training to obtain a trained RGB image single-stream network and an optical stream single-stream network; extracting image flow characteristic information and optical flow characteristic information of each frame in the priori video, and training an asymmetric double-flow network by combining an action label; respectively extracting image flow characteristic information and optical flow characteristic information of each frame in the target video to be detected through the trained RGB image single-flow network and optical flow single-flow network, obtaining segment characteristics of the target video, inputting the segment characteristics into the trained asymmetric double-flow network, and calculating to obtain a video classification vector; selecting potential actions from the video classification vectors to obtain an action recognition sequence of the potential actions; and completing the action detection is completed through the action recognition sequence. According to the action detection method, the asymmetry between the image flow and the optical flow is considered, and the accuracy of action recognition and action detection can be improved.

The invention discloses a video behavior identification method based on a spatio-temporal feature fusion deep learning network, which adopts two independent networks to extract time and space information of a video respectively, adds LSTM learning video time information to each network on the basis of a CNN, and fuses the time and space information with a certain strategy. The accuracy of the FSTFN on a data set is improved by 7.5% compared with a network model without introducing a space-time network proposed by Tran, the accuracy on the data set is improved by 4.7% compared with a common double-flow network model, a segmentation mode is adopted for a video, each video sample samples a network composed of a plurality of segments and inputs the segments into a CNN and an LSTM, and by covering the time range of the whole video. The problem of long-term dependence in video behavior recognition is solved, a visual attention mechanism is introduced to the tail end of the CNN, the weight ofa non-visual subject in a network model is reduced, the influence of the visual subject in a video image frame is improved, and the spatial features of the video are better utilized.

The invention discloses a double-flow network behavior recognition method based on multi-level spatial-temporal feature fusion enhancement. According to the method, a network architecture based on a space-time double-flow network is adopted, and the network architecture is called as a multi-level space-time feature fusion enhancement network. The method aims at solving the problems that the effectof shallow features is ignored and the complementary features of a double-flow network cannot be fully utilized due to the fact that category probability distribution of two flows is only fused at the last layer in a traditional double-flow network. According to the method, a multi-level spatial-temporal feature fusion module is provided, and multi-depth-level mixed features are captured throughthe spatial-temporal feature fusion module at different depth levels of double flows so as to make full use of a double-flow network. In addition, in the network, all features weakens the effect of those features that contribute greatly to classification are equally treated. According to the method, a grouping attention enhancing module is provided in the network, and the saliency of effective areas and channels on features is automatically enhanced. Finally, the robustness of the behavior recognition model is further improved by collecting the classification results of the double-flow networkand feature fusion.

The invention discloses a three-branch network behavior identification method based on multipath space-time feature enhanced fusion. The method adopts a network framework based on a space-time double-flow network, and the network framework is called as a multipath space-time feature enhanced fusion network. The method aims at solving the problems that double-flow information is not fully utilizeddue to the fact that a double-flow network only fuses top-layer space-time features, and feature fusion interaction is insufficient due to the fact that a feature fusion stage is located behind a global sampling layer. According to the method, a compression bilinear algorithm is utilized to perform dimension reduction on multi-layer corresponding spatial-temporal features from a double-flow network, and then fusion is performed, so that the interaction between fusion features is increased and the fusion effect is enhanced while the memory required by the fusion features is reduced. Besides, amulti-scale channel-space attention module is provided in the fusion flow, effective features in the fusion features are enhanced, and invalid features are suppressed. Finally, long-term time information in the video is captured in combination with the thought of a time segment network TSN, and the robustness of the behavior recognition model is further improved.