3785 results about "Euclidean vector" patented technology

Portions from segment boundary regions of a plurality of speech segments are extracted. Each segment boundary region is based on a corresponding initial unit boundary. Feature vectors that represent the portions in a vector space are created. For each of a plurality of potential unit boundaries within each segment boundary region, an average discontinuity based on distances between the feature vectors is determined. For each segment, the potential unit boundary associated with a minimum average discontinuity is selected as a new unit boundary.

Pattern recognition, wherein a sequence of feature vectors is formed from a digitized incoming signal, the feature vectors comprising feature vector components, and at least one feature vector is compared with templates of candidate patterns by computing a distortion measure. A control signal based on at least one time-dependent variable of the recognition process is formulated, and the distortion measure is computed using only a subset of the vector components of the feature vector, the subset being chosen in accordance with said control signal. This reduces the computational complexity of the computation, as the dimensionality of the vectors involved in the computation is effectively reduced. Although such a dimension reduction decreases the computational need, it has been found not to significantly impair the classification performance.

The invention discloses an attention CNNs and CCR-based text sentiment analysis method and belongs to the field of natural language processing. The method comprises the following steps of 1, training a semantic word vector and a sentiment word vector by utilizing original text data and performing dictionary word vector establishment by utilizing a collected sentiment dictionary; 2, capturing context semantics of words by utilizing a long-short-term memory (LSTM) network to eliminate ambiguity; 3, extracting local features of a text in combination with convolution kernels with different filtering lengths by utilizing a convolutional neural network; 4, extracting global features by utilizing three different attention mechanisms; 5, performing artificial feature extraction on the original text data; 6, training a multimodal uniform regression target function by utilizing the local features, the global features and artificial features; and 7, performing sentiment polarity prediction by utilizing a multimodal uniform regression prediction method. Compared with a method adopting a single word vector, a method only extracting the local features of the text, or the like, the text sentiment analysis method can further improve the sentiment classification precision.

An object in a video sequence is tracked by object masks generated for frames in the sequence. Macroblocks are motion compensated to predict the new object mask. Large differences between the next frame and the current frame detect suspect regions that may be obscured in the next frame. The motion vectors in the object are clustered using a K-means algorithm. The cluster centroid motion vectors are compared to an average motion vector of each suspect region. When the motion differences are small, the suspect region is considered part of the object and removed from the object mask as an occlusion. Large differences between the prior frame and the current frame detect suspected newly-uncovered regions. The average motion vector of each suspect region is compared to cluster centroid motion vectors. When the motion differences are small, the suspect region is added to the object mask as a disocclusion.

The invention discloses a short text classification method based on a convolution neutral network. The convolution neutral network comprises a first layer, a second layer, a third layer, a fourth layer and a fifth layer. On the first layer, multi-scale candidate semantic units in a short text are obtained; on the second layer, Euclidean distances between each candidate semantic unit and all word representation vectors in a vector space are calculated, nearest-neighbor word representations are found, and all the nearest-neighbor word representations meeting a preset Euclidean distance threshold value are selected to construct a semantic expanding matrix; on the third layer, multiple kernel matrixes of different widths and different weight values are used for performing two-dimensional convolution calculation on a mapping matrix and the semantic expanding matrix of the short text, extracting local convolution features and generating a multi-layer local convolution feature matrix; on the fourth layer, down-sampling is performed on the multi-layer local convolution feature matrix to obtain a multi-layer global feature matrix, nonlinear tangent conversion is performed on the global feature matrix, and then the converted global feature matrix is converted into a fixed-length semantic feature vector; on the fifth layer, a classifier is endowed with the semantic feature vector to predict the category of the short text.

Embodiments include a motion estimation method performed in a parallel processing system that determines a list of several candidate motion vectors for a macroblock of a video image and retains them through multiple computation passes. All candidate motion vectors are used as potential neighboring predictors, so that the best combination of differential vectors rises to the top of the candidate list. Numerous combinations of differential motion vectors are considered during the process that compares motion vectors among up to eight neighboring macroblocks, instead of simply between pairs of macroblocks. The motion estimation system is configured to use a large number of compute engines, such as on a highly parallel GPU platform. This is achieved by having no dependencies between macroblocks except one per pass. This allows the number of calculations per pass to be very large.

The invention discloses a graph-based vision SLAM (simultaneous localization and mapping) method. According to the method, the matching relation between an image and visual feature can be obtained based on the natural feature probability vector representation of the image, and the relative pose between two interframes can be calculated by utilizing the space geometry relation of images. Data association of visual odometry is obtained by utilizing the corresponding relation of continuous images, so that all constraints in an image sequence can be obtained. The camera relative pose is taken as a node in a map, the space constrained relation of image interframes is taken as an edge, so that an estimated track map based on the camera relative pose is constructed. Finally, a maximum likelihood method is employed for optimizing the map, and optimized pose estimation is obtained through a random gradient descent method. Related experiments are performed in the laboratory environment based on the provided method, also the moving track of a robot is displayed, and the validity of the algorithm is confirmed.

An interframe prediction processor designed for high-speed video coding and decoding. The interframe prediction processor has an address selector and a predictor calculator. The address selector receives information about MBAFF type and coding type. When performing motion compensation for a given MBAFF picture with reference to blocks surrounding a current block, the address selector provides memory addresses of reference motion vectors according to the coding type of the current block pair, as well as to the coding type of the reference blocks. The predictor calculator that determines a motion vector predictor for each constituent block of the current block pair based on reference motion vectors read out of the provided memory addresses.

A motion estimating method of H.264 / AVC coder divides integer motion estimation to be rough layer and fine layer being estimated separately in sequence then inputting parallel interpolation of half pixel and fine layer motion estimation into reference frame data and integer pixel data of rough layer motion vector separately and carrying out half pixel motion estimation as well as quarter pixel motion estimation for calculating out final minimum residual error and its corresponding final motion vector by utilizing motion vector corresponding to minimum residual error under optimum prediction mode obtained from rough layer motion estimation.

The invention provides a contour vector feature-based embedded real-time image matching method. The method uses the linear feature based on X and Y direction vectors, and has strong capability of resisting image distortion, noise, shading, illumination changes, polarity inversion and so on. An image pyramid search strategy is used, templates are quickly matched in a high-layer low-resolution image to be tested, and then, a target position is found out accurately by stepwise downward search, so that matching time is reduced greatly. According to the template image specific information, the best pyramid hierarchy number and the best rotation angle step size for the pyramid template matching of each layer are calculated automatically. An image pyramid highest-layer three-level screening matching strategy is provided, treatment is carried out according to the specific content of the image to be tested, and the first level of screening and the second level of screening are carried out; the non-target position is eliminated just by the addition and subtraction and the conditional statements, which is more efficient in the embedded system than using the multiplication and division; and the third level only processes fewer positions meeting the requirements of the above two levels, so that the matching speed is improved greatly. The overall method can realize the work of matching and locating the target at any angle and any coordinate.

This invention extends routing mechanisms that use link metrics for route selection so that: A link metric cross correlation vector is determined for all links, where each element in the vector corresponds to some other link, and reflects the change in the link metric value if a data flow would already use this other link. The invention further describes a specific embodiment where all cross-correlating links are adjacent to each other, i.e., they terminate or originate in a common node. A mechanism is described to create an extended routing graph. This extended graph permits the use of standard polynomial time algorithms that simultaneously construct the optimal route and find the optimal route metric (such as shortest-path algorithms) also for the adjacent link cross-correlating case.

The invention discloses a geographic and geomorphic characteristic construction method based on laser radar and image data fusion and belongs to the automatic control field. The method specifically comprises 1) obtaining 3D laser point clouds and panoramic pictures of the surrounding environment of a ground unmanned mobile platform at present; 2) matching the 3D laser point clouds and the panoramic pictures and obtaining matched images; 3) dividing the 3D laser point clouds based on different distribution characteristics corresponding to each laser point and carrying out clustering based on a dynamic clustering algorithm of each distribution characteristic to obtain a plurality of region classes; 4) finding passable region classes in the plurality of region classes based on travel ability of the ground unmanned mobile platform; 5)obtaining landform identification vectors of the passable region classes by utilizing a denseness SIFT algorithm; and 6) carrying out landform classification on the passable region classes based on the landform identification vectors and by utilizing a classifier. The method is suitable for passable geographic and geomorphic characteristic construction of the ground unmanned mobile platform.

A system obtains neuro-response data as well as survey based data during discrete choice modeling to evaluate subject decision making processes. A discrete choice model evaluates a decision made by a subject as a function of multiple variables. Neuro-response data vectors and orthogonal survey based data vectors are weighted and combined to generate multi-dimensional vectors. The multi-dimensional vectors are used to estimate the effectiveness of changing particular variables in modifying subject behavior.

The invention discloses an efficient image matching method based on an improved scale invariant feature transform (SIFT) algorithm. The method comprises the following steps of: (1) extracting feature points of an input reference image and an image to be matched by using an SIFT operator; (2) by using a Harris operator, optimizing the feature points which are extracted by the SIFT operator, and screening representative angular points as final feature points; (3) performing dimensionality reduction on an SIFT feature descriptor, and acquiring 64-dimension feature vector descriptors of the reference image and the image to be matched; and (4) initially matching the reference image and the image to be matched by using a nearest neighbor / second choice neighbor (NN / SCN) algorithm, and eliminating error matching by using a random sample consensus (RANSAC) algorithm, so the images can be accurately matched. The method has the advantages that by selecting points which can well represent or reflect image characteristics for image matching, matching accuracy is ensured, and the real-time performance of SIFT matching is improved.

In a motion vector generation apparatus 1 according to the present invention, when motion vectors are sequentially generated for a plurality of images, a target point is determined at a sub-pixel level in any of the plurality of images, a corresponding point corresponding to the target point is searched at the sub-pixel level in another image different from this image, and a motion vector is calculated on the basis of the target point and the corresponding point. Then, the corresponding point is defined as a new target point, and the aforementioned process is repeated to sequentially generate the motion vectors. Thus, the motion vector generation apparatus 1 of the present invention can compute consecutive motion vectors by performing correspondence search at the sub-pixel level in the plurality of images.

A method for tracking objects in three dimensions in a radar system is provided that includes receiving spherical coordinates of an estimated location of each object of a plurality of detected objects, a range rate of each object, and variances for the spherical coordinates and the range rate of each object, determining whether or not each object is currently being tracked, updating a tracking vector for an object based on the object spherical coordinates, range rate, and variances when the object is currently being tracked, and initializing a tracking vector for an object when the object is not currently being tracked, wherein a tracking vector for an object is a process state vector for an extended Kalman filter designed to track an object, elements of the tracking vector including Cartesian coordinates of the object location, the object velocity in three directions, and the object acceleration in three directions.

The invention provides a contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration. The method comprises the first step of obtaining initial three-dimensional point cloud data of the environment where parts of a contact network to be reconstructed are located through motion-sensing peripheral Kinect for Windows, and conducting denoising, simplifying, partitioning clustering, fusing and other preprocessing operations on the initial three-dimensional point cloud data to obtain single-view-angle point cloud data of the parts of the contact network to be reconstructed, the second step of extracting key points through an SIFT algorithm, constructing description vectors of the key points by means of LBP features of uniform patterns and determining the corresponding relations between the key points in different point clouds according to the distances between the vectors, the third step of completing point cloud registration through a rough registration method and an ICP fine registration method and obtaining the complete three-dimensional point cloud data of the parts of the contact network to be reconstructed, and the fourth step of completing three-dimensional reconstruction through the Poisson surface reconstruction method and obtaining a three-dimensional model. According to the method, the key factor is point cloud registration which is the key step influencing the three-dimensional reconstruction speed; the description vectors of the key points are constructed by means of the LBP features of the uniform patterns, so that vector dimensions are reduced, the matching speed of the corresponding relations is increased, registration is accelerated, and the three-dimensional reconstruction speed is increased.

The invention discloses a bi-directional prediction method for video coding / decoding. When bi-directional prediction coding at the coding end, firstly the given forward candidate motion vector of the current image block is obtained for every image block of the current B-frame; the backward candidate motion vector is obtained through calculation, and the candidate bi-directional prediction reference block is obtained through bi-directional prediction method; the match is computed within the given searching scope and / or the given matching threshold; finally the optimal matching block is selected to determine the final forward motion vector, and the backward motion vector and the block residual. The present invention achieves the object of bi-directional prediction by coding a single motion vector, furthermore, it will not enhance the complexity of searching for a matching block at the coding end, and may save amount of coding the motion vector and represent the motion of the objects in video more actually. The present invention realizes a new prediction coding type by combining the forward prediction coding with the backward.

The invention relates to an entity relationship joint extraction method and system based on an attention mechanism. The method comprises the following steps of: converting an entity marked in trainingdata and a triple of a relationship into a form that each word corresponds to a predefined type of tag; Mapping each word in the sentences of the training data into a corresponding word vector, inputting the word vectors into a neural network model based on an attention mechanism, and performing training through a back propagation algorithm to obtain a label prediction model; And inputting the sentences needing to be subjected to entity relationship extraction into the trained label prediction model, predicting a label corresponding to each word, and obtaining entity relationship triples existing in the sentences according to the corresponding relationship between the labels and the words in the triples. The system comprises a preprocessing module, a model training module and a result processing module. According to the method, by more effectively utilizing the key information in the sentences, the joint extraction performance of the relational entities is improved, and the method hasgood practicability.

A method and a system for determining a product vector for computation of a Euclidean distance for performing Dynamic Time Warping of a test signal and a template signal are provided. Low-rank factorized vectors are determined for the template signal. The low-rank factorized vectors are processed along with the test signal for determining the product vector. The product vector is thereafter usable for the determination of a Euclidean distance between the test signal and the template signal, and for performing dynamic time warping of the test signal and the template signal.

Techniques for deriving eigenvectors based on steered reference and used for spatial processing. A steered reference is a pilot transmission on one eigenmode of a MIMO channel per symbol period using a steering vector for that eigenmode. The steered reference is used to estimate both a matrix Σ of singular values and a matrix U of left eigenvectors of a channel response matrix H. A matrix Ũ with orthogonalized columns may be derived based on the estimates of Σ and U, e.g., using QR factorization, minimum square error computation, or polar decomposition. The estimates of Σ and U (or the estimate of Σ and the matrix Ũ) may be used for matched filtering of data transmission received via a first link. The estimate of U or the matrix Ũ may also be used for spatial processing of data transmission on a second link (for reciprocal first and second links).

A method of encoding a motion vector includes: selecting one of a plurality of predicted candidate motion vector sets by using motion information of neighboring blocks of a current block; selecting one of predicted candidate motion vectors within a selected predicted candidate motion vector set, as a predicted motion vector; encoding a differential motion vector representing a difference between a current motion vector or motion vector of the current block and a selected predicted motion vector; and encoding a predicted motion vector index indicating the selected predicted motion vector. As a motion vector is encoded after selecting an efficient predicted candidate motion vector set, the size of a differential vector to be encoded can be reduced without necessarily encoding additional information to indicate which set of predicted candidate motion vectors has been selected, resulting in improved compression efficiency of motion vectors and in turn the improved video compression efficiency.

A space vector pulse-width modulator (SVPWM) and a method implemented by the modulator. A precalculation module accepts Ua and Ub modulation indexes and in response thereto, outputs modified Ua and Ub information; a sector finder has a U module which receives the modified Ua information and outputs a U sector; and a Z module which receives the U sector and the modified Ub information and outputs a Z sector. The U sector and the Z sector are 2-phase control signals for implementing 2-phase modulation. For 3-phase modulation, the SVPWM and method further possess an active vectors calculation module and an assign vectors module which receive the modified Ua and Ub information and the U sector, and which calculate active vectors for 3-phase modulation; a zero vector selector which receives the Z sector and calculates zero vectors for 3-phase modulation; and a PWM counter block which receives the active vectors and zero vectors and outputs 3-phase control signals for implementing 3-phase modulation. The SVPWM and method may have a symmetrical PWM mode, an asymmetrical PWM mode, or both. Advantageously there may also be a rescale and overmodulation module which receives duration information corresponding to the vectors and in response thereto, detects the occurrence of overmodulation. Overmodulation may be detected in response to a negative zero vector time. The module may respond to overmodulation by clamping the zero vector time to zero and rescaling the active vector times to fit within the PWM cycle. The rescaling may restrict a voltage vector to stay within hexagonal boundaries on the space vector plane, while preserving voltage phase.

Neural Network Loadflow (NNL) computation method is invented involving input vector composed of net nodal injection of real and reactive powers and diagonal elements of conductance and susceptance matrices multiplied by the squared voltage magnitude components of the flat-start normally used as initial solution estimate guess for loadflow solution by conventional NRL or SSDL methods. Training, and testing / validating input-output data sets are generated by applying uniform and non-uniform scaling factors applied to base case loads at PQ-nodes, resistance and reactance of transmission line branches. These scale factors are increased until loadflow solution by conventional methods such as Newton-Raphson Loadflow and Super Super Decoupled Loadflow methods diverge. Divergence of loadflow methods are due to node voltage, node angle, and numerical instabilities. Voltage magnitude and phase angle values in the solution before divergence are respective stability limits and voltage magnitude and phase angle values in loadflow solution provide direct measure to the respective stability margins. Also Suresh's diakoptiks based feature selection technique is presented for ANNs calculating one node variable with one neuron each in their output layers.

A method re-identifies objects in a pair of images by applying a convolutional neural network (CNN). Each layer in the network operates on an output of a previous layer. The layers include a first convolutional layer and a first max pooling layer to determine a feature map, a cross-input neighborhood differences layer to produce neighborhood difference maps, a patch summary layer to produce patch summary feature maps, a first fully connected layer to produce a feature vector representing higher order relationships in the patch summary feature maps, a second fully connected layer to produce two scores representing positive pair and negative pair classes, and a softmax layer to produce positive pair and negative pair probabilities. Then, the positive pair probability is output to signal whether the two images represent the same object or not.

A method for sparse channel estimation in MIMO OFDM systems with a plurality of subchannels having the same sparsity structure is presented. The inventive method comprises initializing a plurality of residual vectors and observation generating matrices modeling the channel, sending a pilot signal for each subcarrier, converting the pilot signals to tap positions, detecting an optimal tap position, updating the residual vectors by removing the one residual vector having the optimal tap position, updating the generating matrices in accordance with the optimal residual vector, calculating weighted residuals based on the updated residual vectors, and repeating the steps, except initializing, until a stopping condition is met, wherein the updated observation matrices estimate the sparse channel. In one embodiment, the observation generating matrices are omitted. In one embodiment, multiple vectors are removed during one iteration. Pilot placement and pilot allocation techniques are presented to optimize the method.

An integrated rainfall calculation method using X-band dual-polarimetric radar measurement data includes a precipitation classification step of classifying hydrometeors into four types of snow, rain / snow, rain and non-meteorological target through a fuzzy logic technique using a correlation coefficient (cross correlation coefficient, ρhv), features of a measured differential propagation phase (Ψdp(r)) or differential propagation phase (φdp) and a signal-to-noise ratio (SNR) as input variables (input feature vector); a specific differential phase calculation step of separately calculating a specific differential phase by applying a specific differential phase using a total difference of differential phase and signal-attenuation corrected reflectivity for the rain among the classified hydrometeors and applying a specific differential phase calculated using a filtering method for the other hydrometeors; and a rainfall calculation step of calculating rainfall by using a relation between the specific differential phase and the rainfall and using the separately calculated specific differential phase.