37 results about "Homograph" patented technology

A device and related methods for word-sense disambiguation during a text-to-speech conversion are provided. The device, for use with a computer-based system capable of converting text data to synthesized speech, includes an identification module for identifying a homograph contained in the text data. The device also includes an assignment module for assigning a pronunciation to the homograph using a statistical test constructed from a recursive partitioning of training samples, each training sample being a word string containing the homograph. The recursive partitioning is based on determining for each training sample an order and a distance of each word indicator relative to the homograph in the training sample. An absence of one of the word indicators in a training sample is treated as equivalent to the absent word indicator being more than a predefined distance from the homograph.

The present invention discloses a vehicle all-round multi-view panorama generation method based on an OPENGL. The method comprises the steps of: collecting images comprising special calibration objects by employing four fish-eye cameras respectively installed at the front and back sides and the left and right sides of a vehicle, employing a DLT (Direct Line Transformation) to calculate coordinatesystems of the four cameras and a homograph matrix H of a world coordinate system, and obtaining an Euler angle and a translation vector between the coordinate systems of the four cameras and the world coordinate system through Rodrigues transformation; on this basis, taking vehicle and ground projections as a center to build a flat-bottom bowl model, employing calibrated inner and outer parameters of the cameras to determine a mapping relation of key points in the model and pixel points in fish-eye camera images, mapping texture of each original image onto the flat-bottom bowl model in real time; and finally, controlling movement of a virtual observation camera according to an assigned path, and performing real-time rendering of the scene to obtain an observation result. The vehicle all-round multi-view panorama generation method based on the OPENGL can effectively solve the problem that a traditional 2D panorama distant object is severely stretched, and has good immersion.

The invention discloses an interactive calibration method and apparatus based on three dimensional reconstruction in a three dimensional monitoring system. The interactive calibration method includes the steps: acquiring a reference background model and monitoring videos; acquiring a plurality of scene images, and connecting a plurality of monitoring cameras corresponding to the monitoring videos, wherein at least one scene image among the plurality of scene images has a part which is overlapped with the site of the monitoring videos; according to the plurality of scene images, performing three dimensional reconstruction so as to generate a three dimensional characteristic point cloud for the scenes, and embedding the three dimensional characteristic point cloud into the reference background model; estimating the current frame gesture of the monitoring videos, and performing automatic calibration on the monitoring cameras; and calculating the homograph from the current frame gesture to the reference background model, and embedding the image projection into the reference background model. The interactive calibration method based on three dimensional reconstruction in a three dimensional monitoring system can realize visual, convenient and unified three dimensional video monitoring effect.

The present invention relates to addresses matching fuzzy index and a corresponding matching method and a chain-type supplement means. The addresses matching fuzzy index herein establishes a set of fuzzy rules based on Chinese phonetic alphabets as well as the characteristics of the Chinese language like lots of phonetically similar words and lots of near homograph words, and a fuzzy processing is conducted on Chinese character strings by changing the same into phonetic character strings, and the fuzzy index is established. The addresses matching fuzzy index adopted in the present invention can strengthen fault tolerance of wrong words occurred in the address and obtain a fuzzy match, whereby a success rate of matching can be markedly increased. The chain-type supplement means herein can replace functions of a standard address database under certain circumstance and provide basis for connections among address factors, therefore increasing matching efficiency. The addresses matching fuzzy index and the corresponding matching method and the chain-type supplement means of the present invention overcome the problems of insufficient application of address characteristics, reduce the problem of low calculation efficiency caused by frequent traversing to the address database, and increase the success rate of matching under the premise of a guaranteed matching rate.

The invention discloses a video synchronization method based on homograph and is used for solving the technical problem that in the prior art, a video synchronization error is relatively high when moving targets carry out non-planar movement. The method comprises the realization steps of carrying out moving target trajectory extraction and background image matching on input videos; obtaining antipodal points of background images; normalizing moving target trajectories and feature points and antipodal points of the background images; through utilization of the background points, the antipodal points and trajectory points, obtaining initial trajectory point matched pairs based on the homograph, and removing wrong trajectory point matched pairs through adoption of epipolar geometry constraints; and extracting time information of the trajectory points, and calculating time model parameters between the input videos through utilization of a random sampling consistent algorithm. According tothe method, when the moving targets carry out the non-planar movement, a relatively high accurate matching ratio can be obtained, and the method is applicable to the field such as multi-video monitoring, target tracking, video splicing, three-dimensional reconstruction and video fusion.

A device and related methods for word-sense disambiguation during a text-to-speech conversion are provided. The device, for use with a computer-based system capable of converting text data to synthesized speech, includes an identification module for identifying a homograph contained in the text data. The device also includes an assignment module for assigning a pronunciation to the homograph using a statistical test constructed from a recursive partitioning of training samples, each training sample being a word string containing the homograph. The recursive partitioning is based on determining for each training sample an order and a distance of each word indicator relative to the homograph in the training sample. An absence of one of the word indicators in a training sample is treated as equivalent to the absent word indicator being more than a predefined distance from the homograph.

The invention provides a text sentiment classification method based on deep learning and feature fusion, which learns text representation by integrating hierarchical features and comprises the following steps: designing vector representation of different environments for sentiment words according to text polarity and negative words; performing part-of-speech sampling to obtain vector distinguishing homomorphic words; optimizing word vectors in combination with emoticons, and learning word vector-based simple sentence features by using a neural network model; learning word vector-based simple sentence features through a word sequence; splicing the two parts to obtain sentence-level features; for a document at least containing two simple sentences, inputting a sentence vector sequence\ intoan upper-layer neural network to learn document features based on the neural network, segment heads, averaging segment tails and sentences containing summary words to obtain document features based onrules, and the two parts are spliced to obtain document-level features;nd for a specific task, inputting the single sentence or document features into the random forest classifier to predict the emotion category. Compared with a basic model, the accuracy of text sentiment classification can be effectively improved.

The invention discloses a fundamental matrix and homograph matrix estimation method and system based on GPU (Graphics Processing Unit) parallel speedup. The method is realized in the following steps that: for a plurality of images, on the basis of feature points, utilizing a matching algorithm to obtain the matching list of each image pair, wherein the matching list comprises the matching information of the feature points of the image pairs; due to the restriction of the capacity of a video memory, only selecting M image pairs each time to carry out parallel computation, and carrying out parallel random sampling on feature point matching pairs in the M image pairs; according to a sampling result, carrying out the parallel computation to obtain corresponding candidate fundamental matrixes or candidate homograph matrixes and a corresponding interior point number; then, in the plurality of candidate fundamental matrixes or candidate homograph matrixes which belong to the same image pair, obtaining and optimizing the candidate matrix with the largest interior point number to obtain the final fundamental matrix or homograph matrix. By use of the method, the calculation time of the fundamental matrix and the homograph matrix can be greatly shortened.

The present invention discloses a robust semi-calibrating down-looking image epipolar rectification method and system. The robust semi-calibrating bottom view image epipolar rectification method comprise the following steps: S1, obtaining an image shot by a down-looking camera of an unmanned aerial vehicle at each imaging moment, an intrinsic parameter matrix of the camera, and attitude parameters provided by an inertial navigation system of the unmanned aerial vehicle, and constructing a virtual imaging environment in parallel with a ground level; S2, respectively extracting feature points from the images at different imaging moments, constructing an initial feature point corresponding set, obtaining a feature point corresponding set after transformation according to the initial feature point corresponding set; S3, constructing a probability expression model, calculating an included angle between the course of the unmanned aerial vehicle and the true north direction; and S4, calculating a homograph matrix, and performing epipolar rectification for the image at each imaging moment according to the homograph matrix. The robust semi-calibrating down-looking image epipolar rectification method can quickly achieve rectification, and can obtain high-precision rectified image, thus the robust semi-calibrating down-looking image epipolar rectification method has important guiding significance in the application fields such as unmanned aerial vehicle auxiliary navigation.

The invention relates to an affine reconstructing method based on parallel planes and infinite points, which is applied to the affine reconstructing under the condition that a group of parallel planes and a pair of infinite points are in a scene, the method comprises the steps of shooting two images comprising three visual planes (wherein one group is the parallel planes); extracting the image coordinate of each point required by the image, determining the homographic matrix corresponding to the parallel planes in the scene, establishing a linear constraint equation to the infinite homograph by using the constraint of the infinite point to the infinite homographic matrix, and linearly solving the matrix; finally, establishing a camera matrix to solve the world coordinate of a spatial point so as to reconstruct the affine of an object in the scene. The operations involved in the solving process are all linear operations, and therefore, the affine reconstructing method can avoid certain error and has certain precision.

The invention relates to a method and to a device for identifying objects from images from a camera and can be used in particular in camera-based driver assistance systems. The method for detecting objects from a series of images from a vehicle camera comprises the following steps: a) capturing a series of images by means of the vehicle camera, b) determining corresponding features in two consecutive images, d) determining a plurality of planes in space by associating adjacent corresponding features with a plane in space, f) detecting objects while taking into account the planes determined instep d). According to a preferred embodiment, the method comprises the following step: c) computing homographs for adjacent corresponding features for determining the plurality of planes; and / or d2) associating adjacent corresponding features with a bottom plane, a back wall plane, or a side wall plane. Because the method according to the invention can be designed to be very fast, generic object detection or scene interpretation is possible for nearly the whole image in real time.

A word semantic embedding apparatus includes: a data storing unit to store a lexical semantic network including word dictionary data and word semantic data; a word list generating unit to extract vocabularies having a predetermined part of speech from the stored word dictionary data and generate a list of words to be learned; a processing data generating unit to bring the generated list of words to be learned and the word semantic data of a word to be learned included in the list of words to be learned from the data storing unit and process the data suitable for word embedding learning to generate processing data; and a word embedding learning unit to learn the word to be learned through the word embedding learning using a learning model formed of an input / output layer and a projection layer with the generated processing data to generate a word vector.

The invention discloses disambiguation of homographs in speech synthesis. Systems and processes for disambiguating homographs in speech synthesis are provided. In one exemplary process, voice input containing homographs may be received from a user. The speech input may be processed using an automatic speech recognition system to determine phoneme strings corresponding to homographs pronounced by the user in the speech input. The correct pronunciation of the homograph can be determined based on the phoneme string or at least one of using an n-gram language model of the automatic speech recognition system. A dialogue response to the voice input may be generated, wherein the dialogue response may include homographs. The dialogue response may be output as voice output. The homograph in the dialog response may be pronounced in the voice output according to the correct pronunciation.

The invention discloses a video synchronization method based on homograph and is used for solving the technical problem that in the prior art, a video synchronization error is relatively high when moving targets carry out non-planar movement. The method comprises the realization steps of carrying out moving target trajectory extraction and background image matching on input videos; obtaining antipodal points of background images; normalizing moving target trajectories and feature points and antipodal points of the background images; through utilization of the background points, the antipodal points and trajectory points, obtaining initial trajectory point matched pairs based on the homograph, and removing wrong trajectory point matched pairs through adoption of epipolar geometry constraints; and extracting time information of the trajectory points, and calculating time model parameters between the input videos through utilization of a random sampling consistent algorithm. According tothe method, when the moving targets carry out the non-planar movement, a relatively high accurate matching ratio can be obtained, and the method is applicable to the field such as multi-video monitoring, target tracking, video splicing, three-dimensional reconstruction and video fusion.

The invention provides a binocular video splicing device based on an SOPC. The binocular video splicing device comprises an SOPC system making a NiosII soft-core processor as a core, wherein a pair of CMOS imaging sensors are connected with a pair of binocular video acquisition modules respectively through FPGA ports; the binocular video acquisition modules are connected with a binocular video storage module; one path of the output end of the binocular video storage module is connected with a binocular video output module through a feature extraction co-processor, and the other path of the output end of the binocular video storage module is connected with a binocular video display module; the binocular video output module is further respectively provided with the NiosII processor and an upper computer, and the binocular video display module is further connected with a VGA displayer. According to a binocular video splicing method, feature points are roughly matched through a Euclidean distance method, then partial mismatching points are extracted by means of a KNN method, a homograph matrix is calculated by the adoption of a RANSAC method at last, and real-time splicing for single-frame binocular video data is achieved through cylinder space conversion and linear weighing fusion.

The invention relates to the technical field of image processing, which discloses a method for realizing rapid fusion of human face images, and solves the problems of complicated operation and low processing efficiency of professional image processing software in the traditional technology. In the present invention, at first carry out human face detection respectively to given target image and background image and obtain corresponding human face characteristic coordinate; Then, carry out normalized representation to two groups of human face characteristic coordinates obtained; Estimate the face orientation of the background image, and then perform homography transformation on the target image; then, based on the facial feature points in the background image, deform and stretch the facial feature points in the target image; finally, transform The final target image and background image are automatically segmented, and an exponential attenuation gradient coefficient template with the distance between the pixel point and the feature boundary as a factor is applied to the image segmentation, so as to finally obtain the background image fused with the face image of the target image.

The present invention discloses a robust semi-calibrating down-looking image epipolar rectification method and system. The robust semi-calibrating bottom view image epipolar rectification method comprise the following steps: S1, obtaining an image shot by a down-looking camera of an unmanned aerial vehicle at each imaging moment, an intrinsic parameter matrix of the camera, and attitude parameters provided by an inertial navigation system of the unmanned aerial vehicle, and constructing a virtual imaging environment in parallel with a ground level; S2, respectively extracting feature points from the images at different imaging moments, constructing an initial feature point corresponding set, obtaining a feature point corresponding set after transformation according to the initial feature point corresponding set; S3, constructing a probability expression model, calculating an included angle between the course of the unmanned aerial vehicle and the true north direction; and S4, calculating a homograph matrix, and performing epipolar rectification for the image at each imaging moment according to the homograph matrix. The robust semi-calibrating down-looking image epipolar rectification method can quickly achieve rectification, and can obtain high-precision rectified image, thus the robust semi-calibrating down-looking image epipolar rectification method has important guiding significance in the application fields such as unmanned aerial vehicle auxiliary navigation.

The invention relates to a Chinese character input method based on a specific Latin alphabet tone for inputting Chinese characters through a computer keyboard. The main feature is that on the basis of the current Chinese phonetic alphabet, a specific Latin letter is added after a final letter, and the final letter together forms a dialect final. Method represents pitch. These specific Latin letters are x, v, q, j, etc., which are only combined with i and u initial consonant letters to form pinyin. The same letter means Yinping and light tone. The corresponding keys on the keyboard are: x key means Yangping, v key means upward adjustment, q key means de-adjustment. The j key and the h key distinguish homographs. This program has the advantages of easy memorization, learning and computer operation, and is conducive to the realization of the standardization, standardization and popularization and application of the common language of the country-common and standardized Chinese characters.