127 results about "Coordinate vector" patented technology

A “Music Mapper” automatically constructs a set coordinate vectors for use in inferring similarity between various pieces of music. In particular, given a music similarity graph expressed as links between various artists, albums, songs, etc., the Music Mapper applies a recursive embedding process to embed each of the graphs music entries into a multi-dimensional space. This recursive embedding process also embeds new music items added to the music similarity graph without reembedding existing entries so long a convergent embedding solution is achieved. Given this embedding, coordinate vectors are then computed for each of the embedded musical items. The similarity between any two musical items is then determined as either a function of the distance between the two corresponding vectors. In various embodiments, this similarity is then used in constructing music playlists given one or more random or user selected seed songs or in a statistical music clustering process.

A “Music Mapper” automatically constructs a set coordinate vectors for use in inferring similarity between various pieces of music. In particular, given a music similarity graph expressed as links between various artists, albums, songs, etc., the Music Mapper applies a recursive embedding process to embed each of the graphs music entries into a multi-dimensional space. This recursive embedding process also embeds new music items added to the music similarity graph without reembedding existing entries so long a convergent embedding solution is achieved. Given this embedding, coordinate vectors are then computed for each of the embedded musical items. The similarity between any two musical items is then determined as either a function of the distance between the two corresponding vectors. In various embodiments, this similarity is then used in constructing music playlists given one or more random or user selected seed songs or in a statistical music clustering process.

The invention is suitable for video monitoring technologies and relates to a method and an apparatus for crowd behavior analysis in video monitoring. The method comprises: obtaining a video stream and dividing each frame of image in the video stream into image sub-blocks with different perspective degrees by taking human height as a size reference by row; extracting deep neural network features in the image sub-blocks; classifying and optimizing the deep neural network features, obtaining a corresponding vector feature set, establishing SVR mathematical models of different kernel functions according to the vector feature set, and making statistics on the number of people in crowds with different density; and adopting a camera calibration method for obtaining coordinate vectors of the deep neural network features, comparing displacements of the deep neural network features of the two continuous frames of image sub-blocks, and calculating a moving direction and a moving speed of the crowd according to the displacement. According to the method and the apparatus, the crowd in the video monitoring is tracked and analyzed to obtain feature information, such as the moving speed, the moving direction, the number of people and the like, of the crowd, the feature information and historical data are analyzed and compared, a state of the crowd is judged, and an alarm is given for an abnormal event.

A 3-D morphology measuring method based on 3-D coding includes preparing N pieces of gratings, dividing said grating by check in chessboard form, projecting said grating and shooting picture then decoding said picture for obtaining actual 3-D coordinate vector of each check and carrying out stereo-matching on said 3-D coordinate vector for obtaining measurement space information.

The invention provides a model rendering method and device. The method comprises the steps that the virtual object model of each virtual object is acquired, wherein the virtual object model is created for a virtual reality scene; the coordinate vector of each virtual object model in a local coordinate system is converted to a coordinate vector in a camera coordinate system; the scene cone of the virtual reality scene is created; according to the coordinate vector of each virtual object model in the camera coordinate system and the scene cone, each virtual object model in the scene cone is acquired; and each virtual object model in the scene cone is successively rendered from being far to a camera to being near to the camera to display the virtual reality scene. According to the invention, the model rendering efficiency and the model rendering display effect are improved.

The invention discloses an edge extraction method of a color image. The method includes the following steps: 1) filtering an original color image to be processed; 2) converting the filtered color image into a color space coordinate system; 3) respectively extracting an image interesting area in each coordinate vector and 4) conducting edge detection on the extracted interesting area under each coordinate vector to obtain an edge image of each coordinate vector and combining and overlaying the edge images of the coordinate vector to obtain a final edge image. Due to the fact that edge detection is only conducted on the interesting areas in the edge extraction method, information processing quantity is less in edge detection, and detection radius of an edge detection operator can be improved. Simultaneously, the color image is converted in the color space coordinate system to be processed in the edge extraction process, so that the method enables edge extraction result to be more complete compared with the prior art that a method for converting the color image into a gray-scale map for processing is adopted.

The invention provides a binocular camera-based living body identification method. The method comprises the steps of collecting human face images corresponding to an object under visible light and infrared light by utilizing a binocular camera; extracting human face key points corresponding to the human face images respectively in the human face images under the visible light and the infrared light; obtaining human face key point coordinate vectors, reflected in the binocular camera, of the target object; performing distance measurement according to the human face key point coordinate vectors, and calculating distances from the human face images to the camera; and under a light source with the same frequency and the same distance, calculating whether the light intensities of the human face key point coordinate vectors are in a preset range or not, judging whether the object providing a human face is a living body or not according to a calculation result. The invention furthermore provides a binocular camera-based living body identification apparatus. Under the condition of obtaining the frequency of the light source, whether the object is the living body or not is judged according to the reflected light intensities of different materials under the same distance, so that other devices do not need to be additionally purchased and the manufacturing cost is reduced.

Disclosed is a method for determining coordinates of a mobile rover. The method includes determining a vector of one-shot code coordinates of the mobile rover. The method also includes determining a vector of phase increments by determining full phase differences for each navigation satellite in a plurality of navigation satellites in view at a discrete time interval (called a time epoch) and at a previous time epoch in a plurality of time epochs. A vector of radial range increments is determined from the full phase differences. A vector of rover phase coordinate increments is also determined using the vector of radial range increments. The vector of one-shot code coordinates and the vector of rover phase coordinate increments are then filtered to determine, at each time epoch, smoothed code coordinates of the mobile rover. Measured phase increments are cleared up from abnormal measurements.

The invention provides a face detection and restoration method based on an RGBD camera. The method comprises the following steps: 1) constructing a binocular camera system based on a RGB camera and aninfrared camera, and obtaining image data from the RGB camera and the infrared camera respectively; 2) carrying out face detection on the two types of image data; 3) determining face feature vectorsin the image data; and 4) calculating an approximate coordinate position on an infrared image of a face key point coordinate vector on an RGB image by utilizing corresponding matrixes of two camera images, and selecting a face key feature vector with the minimum distance from the face key feature vectors detected on the infrared image as an optimal matching result. Since depth images are calculated based on the infrared images, the depth images generally keep a one-to-one correspondence relation with infrared image pixels, and the depth image face and the infrared image face have the same position; and through combined processing of RGB camera images and the depth images, face detection and matching of the two types of data can be finished while collecting the RGBD data, which makes it possible for high-level applications in the RGBD-based face image processing field.

The invention discloses a target locating method of a miniature drone full-strapdown down looking camera. The target locating method comprises the step that the camera focal length is calculated through the camera field angle and the width of a camera square pixel array, then the distance length from the pixel position to the origin of coordinates is obtained through the camera focal length and the position of a target on the pixel array, the unit coordinate vector of the target in the camera coordinate system is obtained through the distance length from the pixel position to the origin of coordinates and the position of the target on the pixel array, the relative distance from the target to a drone is obtained through estimation in combination with a Kalman filter algorithm, the relationbetween geographic position information of the target and the relative distance from the target to the drone is obtained through coordinate conversion, and geographic position information of the target is calculated through the distance from the target to the drone, the drone position information amount obtained through measurement of a satellite navigation system and the unit coordinate vector ofthe target in the camera coordinate system. By means of the target locating method, the high locating precision can be obtained, and part of measurement errors can be removed.

The invention discloses a human body behavior recognition method based on a convolutional neural network and a recurrent neural network, and the method comprises the steps: tracking a human body behavior through a sensor, and collecting a three-dimensional coordinate vector group and an RGB video of a human body joint in a time period; training the three-dimensional coordinates of the human jointby using a recurrent neural network RNN to obtain a time feature vector; and training the RGB video by using a convolutional neural network (CNN) to obtain a space-time feature vector, finally combining the time feature vector and the space-time feature vector and normalizing, feeding the normalized space-time feature vector and the normalized space-time feature vector to a classifier of the linear SVM, finding a parameter C of the linear support vector machine (SVM) by using the verification data set, and finally obtaining a comprehensive identification model. According to the method, the problem of overfitting of action classification of the model in the model training process can be solved, and meanwhile, the human body behavior recognition efficiency and accuracy can be effectively improved.

A method for multidimensional scaling (MDS) of a data set comprising a plurality of data elements is provided, wherein each data element is identified by its coordinates, the method comprising the steps of: (i) applying an iterative optimization technique, such as SMACOF, a predetermined amount of times on a coordinates vector, said coordinates vector representing the coordinates of a plurality of said data elements, and obtaining a modified coordinates vector; (ii) applying a vector extrapolation technique, such as Minimal Polynomial Extrapolation (MPE) or reduced Rank Extrapolation (RRE) on said modified coordinates vector obtaining a further modified coordinates vector; and (iii) repeating steps (i) and (ii) until one or more predefined conditions are met.

The invention relates to a self-adaptive finite time convergence sliding-mode control method of a robot and belongs to the technical field of control. The method comprises steps as follows: establishing a robot kinetic equation considering uncertainty, determining a sliding mode surface, then introducing n-dimensional self-adaptive updating rate, calculating value of active power/moment required for driving the robot finally, and driving a robot system based on the value so as to enable generalized coordinate vectors of the robot to converge to a steady-state or trace command signals in finite time. The method has the characteristics and benefits as follows: firstly, the problem of chattering of the sliding-mode control is solved greatly while high-precision control is realized; secondly, possible influence of different dynamic characteristics in all directions of freedom degrees of the robot system can be eliminated; thirdly, the problem of moment saturation and chattering of the robot system at the starting stage can be solved; fourthly, kinetic compensation guaranteeing real-time performance of the control can be realized. The method is applicable to the robot systems with structure types of parallel connection, series connection, series-parallel connection and the like.

The invention discloses an load prediction method and system based on LLSVM optimization. The load prediction method based on LSSVM optimization comprises steps of discriminating and correcting abnormal data after obtaining original history load data, constructing a characteristic vector, performing k average value clustering on characteristic vectors, choosing a input variable according to a clustering effect, 2) using a punishment factor C of an LSSVM model and a kernel function width parameter sigma as position coordinates of a particle in a searching space for a particle swarm algorithm, using a particle coordinate vector value [C,sigma] having a smallest fitness value as an output of the particle swam algorithm, (3) using optimized and processed input variable data as input and output of the LSSVM model, and using the [C,sigma] obtained from the step 2 to solve a load prediction regression equation and using the regression equation to perform load prediction. The load prediction method and system based on LSSVM optimization can correct abnormal data, find the most suitable punishment factor and the kernel function parameter and improve accuracy of the load prediction based on the LSSVM.

The application provides a robot hand and eye calibration method and device used in a robot system. The robot system comprises a mechanical arm, a calibration plate and a panoramic camera keeping fixed relative to a mechanical arm base, wherein the calibration plate is fixed at the mechanical arm tail end, and the relative positions of the mechanical arm tail end and the calibration plate remain unchanged. The method comprises the following steps of obtaining a relative position coordinate B of the mechanical arm tail end in a calibration plate coordinate system according to a plurality of calibration plate images shot by the panoramic camera in a first state, recording a plurality of position coordinate vectors D of the mechanical arm tail end in a mechanical arm base coordinate system ina second state, and solving to obtain a conversion matrix E between a panoramic camera coordinate system and the mechanical arm base coordinate system according to a first conversion matrix A1 between the calibration plate coordinate system and the panoramic camera coordinate system and the relative position coordinate B of the mechanical arm tail end in the calibration plate coordinate system, so that the accuracy is ensured and the process is simple and convenient.

The embodiment of the invention provides an expression recognition method and device, electronic equipment and a storage medium. The expression recognition model comprises a convolutional neural network model, a full connection network model and a bilinear network model. The method comprises steps of in the expression recognition process, preprocessing the to-be-identified image to obtain a face image and a key point coordinate vector; carrying out operation on the face image through a convolutional neural network model to output a first feature vector, performing operation on the key point coordinate vector through the full connection network model to output a second feature vector, performing operation on the first feature vector and the second feature vector through the bilinear networkmodel to obtain second-order information, and further obtaining an expression recognition result according to the second-order information. In the process, the prior expression information containedin the face key points is considered, the robustness to postures and illumination is good, and the expression recognition accuracy is improved. Furthermore, when the expression intensity is low, the expression can be correctly identified.

The invention provides a sparse planar formation optimization method based on spatial gain. The method comprises the following steps that an array is initialized, a coordinate matrix formed by coordinate vectors of array elements is obtained, and spatial gain corresponding to the coordinate matrix is calculated; iteration parameters are set; iterative optimization is carried out on the initialized array, 5-N array elements in the array are sequentially disturbed in each iterative optimization, new coordinates of the array elements generated after each disturbance is finished are judged, the new coordinates meeting the judge conditions replace old coordinates of the array elements, the coordinate matrix is renewed, the interactive process is repeated until a set interactive total number is reached, an array manifold is received, and a coordinate matrix (x0,y0) which is renewed after iterative optimization is used as an array position of a sparse planar matrix. When an angle-directed face array is needed, a cost function can comprehensively evaluate side-lobe level and main-lobe beam width, and by means of the method, the sparse planar array with narrow main-lobe beam width and low side-lobe level can be generated.

The invention belongs to the technical field of radioactive particle implantation, and discloses a method for realizing radioactive particle implantation by using laser guidance, comprising the following steps: S1, marking a target area on the body surface of a patient; S2, obtaining the image examination data of the patient target area and introducing the image examination data into the processing system to generate a disease body model; 3, introducing the disease body model into a computer three-dimensional treatment plan system for dimension development, determining and adding a source lineof a puncture needle path, arranging radioactive particles on the source line, and derive the radioactive particles in the form of a three-dimensional data packet; S4, introducing the three-dimensional data packet into the orbit equipment and emitting laser rays according to the three-dimensional coordinate vector of the distribution source line; and 5, inserting a particle needle along that laser ray and implant radioactive particles. The invention adopts laser technology to guide, combines the advantages of freehand puncture and template puncture, shortens the preparation period before operating, lowers the cost, and can change the implantation scheme in time according to the change of the tumor in the patient body, and has strong flexibility.

The invention provides a method for predicting the stability of vehicles, and belongs to the field of automotive technologies, and the method can be used for solving the problem that in the prior art, methods for predicting the stability of vehicles are not high in accuracy and low in reliability. The method comprises the following steps: S10, detecting an acting force Fi exerted on each wheel of a target vehicle by the ground; S20, analyzing and calculating a force-center coordinate vector as shown in the specification of the target vehicle; and S30, substituting the force-center coordinate vector as shown in the specification into a vehicle force-center/stability transfer function which is as shown in the specification and is pre-stored in a vehicle-mounted controller of the target vehicle so as to calculate a stability signal Scar of the target vehicle, and according to the value of the stability signal Scar, judging the current stability state of the target vehicle; a test vehicle and the target vehicle are same in type or similar in technical parameter, the test vehicle is defined as a vehicle for calibrating test data under experimental conditions, and the target vehicle is defined as a vehicle normally running on the road. The judging process in the method is simple and convenient, and the method is accurate in judgment and high in reliability.

The invention discloses a self-focusing method for high-resolution moving ship SAR imaging. The method comprises the following steps: carrying out coarse compensation and range direction matched filtering processing on a baseband echo signal by adopting inertial navigation data; coarsely focusing the imaged original image; screening out a plurality of strong scattering regions from the original image, and respectively estimating a local phase error of each strong scattering region by using a weighted phase gradient self-focusing method; calculating a coefficient matrix of a second-order errorpolynomial; constructing coordinate vectors X and Y of the whole original image, and fitting a phase error of each pixel point at each slow moment by utilizing a coefficient matrix; and compensating the original image by adopting the phase error estimation matrix. According to the invention, the problem that a traditional self-focusing method cannot perform unified estimation compensation on space-variant phase errors caused by ship motion during moving ship imaging is solved, and accurate focusing of moving ship high-resolution imaging is achieved.