190results about How to "Solve the large amount of calculation" patented technology

The invention discloses a man-machine cooperation device and method based on a Kinect video camera. The man-machine cooperation device and method based on the Kinect video camera is mainly based on a Kinect video camera which is launched by the Microsoft Corporation recently so as to achieve detecting and tracking of a human body target. The man-machine cooperation device and method based on the Kinect video camera can accurately judge the position of a target object and the intention of a tester by utilizing Kinect human skeleton detecting technology, and therefore the risks which occur in the process of handover are avoided. Besides, a rail rapid transit (RRT) algorithm based on working space is provided to achieve route planning. Through repeated experimental verification, the system can conduct interpersonal handover safely.

The invention discloses a robust lens distortion correction method. The invention provides a method for separating lens distortion from camera parameters to be solved separately. The method comprisesthe following steps of: performing a feature extraction from a checkerboard calibration object by firstly performing angular point detection utilizing a Shi-Tomasi operator, then performing screeningaccording to the specific characteristics of symmetry and large variance of the checkerboard angular point, and finally obtaining an sub-pixel coordinate with accurate checkerboard angular point by utilizing a sub-pixel optimization algorithm; establishing a camera imaging model and a lens distortion model to obtain an optimized objective function with respect to a distortion parameter; and solving the distortion parameter by utilizing a nonlinear optimization algorithm for distortion correction. According to the robust lens distortion correction method, a calculation of all main distortion parameters can be accomplished with only one calibration object image containing the checkerboard, the model is simple, the calculation efficiency and stability are high, the application adaptability isgood in an industrial field, and the calculation accuracy is comparable to a traditional method.

The embodiment of the invention discloses a structured entity recording method and device, a server and a storage medium. The structured entity recording method comprises the steps that candidate entities related to structured entities to be recorded are selected from a knowledge graph, the structured entities to be recorded are determined as associated entities according to priori attribute information of the type to which the candidate entities belong and a preset model, the associated entities and the candidate entities are merged, and the associated entities are recorded into the knowledgegraph. According to the embodiment of the invention, the problems are solved that when an existing knowledge map records entities, since entity analysis is conducted on existing entities in the knowledge graph once in the process of adding one entity, the calculation amount is large, the consumed time is long, and the accuracy of entity analysis is not high; the candidate entities are selected, and then prior knowledge is utilized to fuse the preset model so that the efficiency and accuracy of entity association can be effectively improved; the calculation amount is reduced, and therefore theknowledge graph can easily and efficiently record the structured entities.

The invention discloses a method for reconstructing a three-dimensional facial expression model based on a monocular video. Extra multi-angle shooting is not needed, A universal 3D face model is directly driven from a neutral expression image frame in an input monocular video to deform so as to generate a personalized three-dimensional face template; and expressing the deformation of the three-dimensional face expressions corresponding to different frames as the change of the personalized three-dimensional face template in the 3D vertex flow of the three-dimensional space, and solving the coarse-scale geometric model of the face expressions through the consistency with the change of the 2D optical flow. While the shape accuracy of a coarse-scale reconstruction model is improved by utilizing dense optical flow, the shooting requirement of an input video is broadened, details are added on a recovered coarse-scale face model through a light and shade recovery shape technology so as to recover fine-scale face geometric data, and a three-dimensional face geometric model with high fidelity is reconstructed.

This patent discloses a convolution neural network accelerator based on PSoC device, including an off-chip memory, a CPU, a feature map input memory, a feature Map Output Memory, a bias memory, a weight memory, a direct memory access to the same number of cells as neurons. The calculation unit comprises a first-in first-out queue, a state machine, a data selector, an average pooling module, a maximum pooling module, a multiplication and addition calculation module and an activation function module, wherein the calculations in the multiplication and addition calculation module are executed in parallel, and can be used for a convolution neural network system of various architectures. The invention fully utilizes the programmable part in the PSoC (Programmable System on Chip) device to realize the convolution neural network calculation part with large calculation amount and high parallelism, and utilizes the CPU to realize the serial algorithm and the state control.

This invention provides a quick satellites choosing method for together fitted satellites system. It chooses satellites indirectly in this way: first, confirm the number of satellites in need. Then calculate the quadrant elevation and azimuth angle of visible satellites, set them to different parts and choose a group of satellites which has a medium quadrant elevation. Disposal the azimuth angle of the choosing satellites, combine the quadrant elevation information to remove some improper satellites. This satellite choosing method is quick, convenience and needs less calculation than the traditional one. Even it is not as precise as before, but it makes the navigational computer working more easily and guarantee the precision of together fitted navigation.

The invention relates to a method and a system for detecting and screening end defects in circular parts. The system comprises an optical amplifier, a control computer, a controller, a pneumatic nozzle device, a step motor and a conveyor belt. The method includes the steps of 1, photographing a circular part to be detected by the optical amplifier to obtain an optically amplifier image of the end of the circular part to be detected; 2, subjecting the optically amplified image to image processing and defect recognition by the control computer, and judging whether the circular part is qualified or not according to defect conditions; 3, allowing the controller to receive a command of controlling the computer, control the pneumatic nozzle device to operate, remove unqualified parts, and control the conveyor belt to convey next part to a detection position. The method and the system have the advantages that end defects in circular parts can be reflected truly, whether an object is qualified or not can be judged accurately, automation level is high, and detection efficiency is improved.

The invention discloses a method and a device for optimizing a multidimensional cube model. The method for optimizing a multidimensional cube model comprises the following steps: building an all-dimension multidimensional cube model based on a dimension set, wherein the dimension set includes all data dimensions in a service model; updating the dimension set based on acquired user behavior data and according to preset updating rules; and regenerating a multidimensional cube model according to the updated dimension set. According to the technical scheme of the invention, the dimension set is updated based on the acquired user behavior data, and then, a multidimensional cube is regenerated. The problem that the updating of a multidimensional cube at present requires a large amount of calculation and depends on the experience of the designer is solved. A multidimensional cube can be updated automatically. Updating of the dimension set based on the operation data in a database is closer to the actual need of users. Thus, the updated multidimensional cube model is smarter and more practical.

The invention discloses an image identification method and apparatus. The image identification method includes the steps: inputting an image to be identified into a trained first convolution neural network which includes P groups of convolution layers, wherein each group of convolution layers includes two convolution layers; the sizes of the convolution cores of the two convolution layers are respectively a 1*N first convolution core and an N*1 second convolution core; by means of each group of convolution layers in the first convolution neural network, extracting the characteristics of the image to be identified and obtaining the characteristics, of each group of convolution layers, to be extracted; and according to the characteristics, of each group of convolution layers, to be extracted, determining the identifying result of the image to be identified. The technical scheme of the image identification method and apparatus solves the problem that in the related technologies, a square convolution core causes high computational complexity and high computing cost.

The invention discloses a screen content display method and a screen content display device. The screen content display method includes a first step of receiving an updating instruction sent by a user and confirming an updated area according to the updating instruction, a second step of reading pixel points in the updated area one by one, wherein the upper layer of the pixel points is a image layer and the lower layer of the pixel points is a video layer, a third step of configuring the transparency of the image layer of the pixel points according to the pixel value of the image layer of the pixel points for the pixel points which are read, mixing the pixel value of the video layer of the pixel points with the pixel value of the image layer when the transparency value is bigger than 0 and smaller than 255, and storing a mixed result to a corresponding position of a cache region, and a fourth step of displaying the pixel values stored in the cache region on a screen. Through adoption of the screen content display method, the problem in the prior art that calculated amount of screen contents is large in the displaying process is solved. Consequently, calculation procedures are simplified, power consumption of the device is reduced and user experience is improved.

The invention is applicable to the technical field of computer vision and image processing, and provides a lane line detection method and device, terminal equipment and a readable storage medium. Inputting the road image into a trained neural network model for processing, and outputting a detection result of a lane line in the road image of the current scene; wherein the trained neural network model is obtained by training according to sample images in a training set and a semantic segmentation model, and the sample images in the training set comprise collected road images of a plurality of scenes and marked images corresponding to the road images of the plurality of scenes. According to the method and the device, the problems that most of deep learning models used for lane recognition atpresent are relatively large in calculation amount, relatively complex in model and unfavorable for meeting the requirement on real-time performance in an actual application scene of an automatic driving task can be solved.

The invention relates to an image processing method and device. The image processing method comprises the steps of acquiring background statistical information of the current scene image, wherein the background statistical information comprises a background mean value and a standard deviation value of a first color area of the current scene in a color space to which the first color area belongs; carrying out matting processing on the current scene image according to the background mean value and the standard deviation value, and acquiring foreground image information of the current scene image; and carrying out fusion processing on the foreground image information and a virtual background image so as to acquire a fused scene image.

The invention relates to a rubbing sound emission signal denoising method based on discrete fraction cosine transform. The invention comprises the technical proposal that: a sound emission signal is extracted from an emission sensor of a friction point and superimposed with noise; the signal comprising noise is sampled, quantized and subframed and added with windows to obtain a discrete time sequence; the sequence is carried out discrete fraction Fourier transform with the fraction a to obtain F(k), then the signal after being filtered is carried out the discrete fraction Fourier transform with the fraction 2-a and the combination of frames; the signal after being denoised is output. The rubbing sound emission signal denoising method solves the disadvantages that a sound emission source has diversity, transient property and a noise source has complexity and can not gain satisfactory effect. The rubbing sound emission signal denoising method can obtain good denoising effect to the rubbing sound emission signal under the environment of a plurality of nonstationary noise, strengthen the sound emission signal, has robustness to the diversity and transient property of the sound emission source and the complexity of the noise source, effectively solves the problem of the large calculation amount in the prior method and has wide applicability.

The embodiment of the invention discloses a CT three-dimensional reconstructing method and system. The method comprises the steps of sending two-dimensional projection images obtained by CT at at least two rotation angles to at least two lower computers; controlling projection matrixes of all pixel points in the two-dimensional projection images received by all the lower computers through calculation, and storing the calculation result into memories of all the lower computers; according to the projection matrixes and the current iteration state of three-dimensional body data to be reconstructed, controlling all the lower computers to calculate an iterative attenuation weight and returning; according to the iterative attenuation weight, updating the current iteration state; returning to execute control over the projection matrixes and the current iteration state which are stored in the memories of all the lower computer, and calculating an iterative attenuation weight and returning till a preset ending condition is met; adopting a current iteration state after iteration ends as a three-dimensional representing result of a CT. By utilizing multiple lower computers to conduct distributed parallel calculation, the problems of low calculating efficiency, low reuse performance of calculating result and the like are overcome, and the three-dimensional reconstructing image quality is guaranteed.

Disclosed are a system and a method for integrating mobile phone test data. The system comprises a mobile phone test platform and a file server. The mobile phone test platform is connected with the file server through a network. Data acquisition software used for acquiring the mobile phone test data is arranged in the mobile phone test platform. The file server comprises a test data receiving module, a test data analyzing module and a test database. The test data receiving module receives the mobile phone test data acquired by the mobile phone test platform, and the test data analyzing module analyzes mobile phone test detail data according to the mobile phone test data and saves the mobile phone test detail data into the test database. The system and the method for integrating the mobile phone test data can automatically acquire the mobile phone test data of a mobile phone in the process of testing in real time, and automatically extract the mobile phone test detail data from the test data to classify and gather the mobile phone test detail data. Thus, the problems that, in the prior art, manual acquisition and analysis of the mobile phone test data is large in computing amount and low in efficiency and accuracy are solved, and timeliness of mobile phone test data acquisition is ensured.

The embodiment of the invention discloses a pre-coding matrix selection method, a device and a system. The method comprises the steps that pilot frequency information sent by a sending end is received through the current channel; according to the pilot frequency information, an equivalent channel matrix of the current channel is estimated; inner product and modulo operation are respectively carried out on the equivalent channel matrix and each pre-coding matrix in a preset pre-coding codebook collection, so as to acquire a number of module values; the pre-coding matrix corresponding to the maximum module value in a number of module values is selected as a target matrix; and the target matrix is fed back to the sending end, so that the sending end selects the target matrix to carry out pre-coding on a signal sent through the current channel. The embodiment of the invention further discloses a pre-coding matrix selection device and system. According to the invention, the pre-coding matrix selection method, device and system are provided; a pre-coding selection process can be simplified; the calculation amount in the pre-coding selection process is reduced; and the problems of large calculation amount and complicated operation of the pre-coding matrix selection method in the prior art, are solved.

The invention discloses a method, a device and a system for tracking a rigid body. The method comprises steps of acquiring an image, acquired by a camera, of a rigid body and determining a three-dimensional coordinate of feature points in the image relative to the camera; acquiring an initial posture, acquired by an IMU, of the rigid body and a rigid body posture corresponding to a shooting period of the camera, and determining identity information of feature points of the image; according to the identity information of the feature points, obtaining matched feature point pairs; according to the matched feature point pairs, calculating rotation and translation information of the feature points in the image; and according to the rotation and translation information, calculating position information of the rigid body in an optical system. According to the invention, problems of big calculated amount of the current feature matching algorithm and severe limitation of application scenes are solved; a problem of low space positioning precision caused by low resolution of images acquired by the camera in the computer graphics-based tracking algorithm is solved; an objective of complementing advantages of the computer graphics-based tracking algorithm and the IMU-based tracking algorithm is achieved; and tracking precision of the rigid body is improved.

The invention belongs to the field of parking lot path planning in intelligent transportation, and discloses an intelligent parking control method and system, a storage medium, a computer program anda terminal. The method comprises the following steps: constructing a superposed data energy field through the obtained physical information of a parking lot, and completing the recommendation of parking spaces; performing regular statistics on the parking duration of all parking spaces of the parking lot, fitting a statistical result, and superposing the fitting result and the original data energyfield to complete parameter updating; calculating real-time weights of all roads through real-time physical information of the parking lot, constructing a real-time undirected weighted graph of the parking lot, and planning a real-time optimal path by using an improved Dijkstra algorithm; and displaying the result and continuously optimizing the result by utilizing the obtained real-time physicalinformation of the parking lot until a user finishes parking. According to the invention, the problems caused by many parking spaces and irregular distribution of parking spaces in large parking lotsare effectively solved, and the method has better dynamics and real-time performance.

The invention relates to a pulse width modulation method and device based on an equivalent zero vector, belonging to the technical field of power electronic converters. The invention determines the action time of the modulated wave space vector according to a space vector algorithm, adopts the adjacent symmetrical vector with equal time to replace the zero vector in the modulated wave space vector, and ensures the voltage output of the three-phase converter to achieve the effect of reducing the common-mode voltage of the converter port on the basis of the voltage output of the three-phase converter. The invention completely retains the existing space vector calculation part, obtains the carrier signal by calculating the voltage instruction through the space vector algorithm, and greatly saves the calculation amount and improves the efficiency compared with an existing vector modulation method for reducing the common-mode voltage.

The invention relates to a group target identification method based on a high-resolution range-profile and a single-pulse measurement angle. The method is used for radar group target identification. For the disadvantages that the traditional group target identification method requires each sub-target to move relative to a radar and the computation burden is comparatively large, the invention provides the group target identification method based on the high-resolution range-profile and the single-pulse measurement angle. Based on the high-resolution range-profile and the single-pulse measurement angle information, the clustering is accomplished by adopting a mean value drifting clustering algorithm on a two-dimensional plane, and HRRPs of various sub-targets are extracted without utilizingmovement information of the target; the method has no constraint on a condition whether each sub-target is in movement, and a requirement of requiring each sub-target to move relative to the radar inthe traditional method is solved. By utilizing scattering point amplitude information, the method improves the traditional mean value drifting clustering algorithm from two aspects: initial value selection and sample weight, the iteration efficiency of the mean value drifting algorithm is improved, and the problem that the traditional method is large in computation burden is solved.

The invention discloses a method for measuring the inner wall water film thickness of a microwave humidity sensor on the basis of linear fitting. The measuring method comprises the following steps that the resonant frequency of a microwave measurement resonant cavity in a TE111 mode on the condition that no water film exists is calculated; the water film thickness and the corresponding frequency deviation caused by the water film thickness are calculated, wherein 0<hi<=200 micrometers, and 0<i<=N; linear fitting is conducted on the water film thickness hi and the corresponding frequency deviation deltaf<w-1,i> caused by the water film thickness; the resonant frequency of the microwave measurement resonant cavity in the TE111 mode is measured; the resonant frequency deviation caused by the water film is calculated; the water film thickness is calculated through a linear relation between the water film thickness and the frequency deviation caused by the water film thickness. According to the method, the water film thickness hi and the corresponding frequency deviation deltaf<w-1,i> caused by the water film thickness present the linear relation on the basis of the work of the cylindrical waveguide resonant cavity in the TE111 mode, the inner wall water film thickness of a wet steam humidity sensor is quickly and accurately measured, and convenient conditions are provided for eliminating measuring errors brought by the water film thickness on a humidity measuring result and improving the humidity measuring precision.

The invention relates to a pulse Doppler radar ranging improvement method based on a one-dimensional set method. The method includes the steps of establishing a finding table, and making each numerical value in the finding table be 0; obtaining the apparent distance threshold s[i] and the maximum fuzzy degree f[i] corresponding to the ith pulse repetition frequency P[i] of a radar transmitting signal; obtaining the apparent distance x[i] corresponding to the ith pulse repetition frequency P[i] of the radar transmitting signal according to a target echo signal; renewing the maximum fuzzy degree f[i] corresponding to the apparent distance threshold s[i] to obtain the renewed maximum fuzzy degree f[xi] corresponding to the apparent distance threshold s[i]; obtaining the qth element r[iq] of the distance unit sequence number set (please see the specifications), wherein the qth element r[iq] of the distance unit sequence number set (please see the specifications) is equal to the sum of x[i] and the product of q and q[i], and q is equal to 0, 1, 2...f[xi]; renewing the numerical value (please see the specifications) in the ith row and the R[i]th line of the finding table as 1, wherein R[i] is equal to r[i0], r[i1], r[i2]...r[if[xi]]; making j sequentially be 1, 2...N, and obtaining the distance A between a target and the pulse Doppler radar if the sum y[j] of all the numerical values in the jth line in the finding table is larger than or equal to L.

The invention discloses a method for measuring and calculating the human-robot minimum distance in a collaborative environment. According to the relative position relation and the rotating axes of theparts of a robot, a generalized motion model of the robot is established. Meanwhile, a motion node iteration data set of the robot with a minimum human-robot distance is constructed. Meanwhile, the image data of the human-robot cooperation space is collected through a 3D visual sensor, so that the human body information in the collaborative environment is tracked and identified. The skeleton nodedata of a human body are extracted, and a human skeleton node iteration data set with the minimum human-robot distance is constructed. According to the obtained iteration data set, the minimum distance between the human body and the robot in the collaborative environment, and the space coordinates of corresponding points are calculated in the iterative manner. According to the invention, based onthe method for measuring and calculating the human-robot minimum distance in the collaborative environment, an accurate and easy-to-achieve human-robot distance calculation model is built in real time. After that, a human-robot minimum distance and a corresponding point in the collaborative environment are calculated in real time. The safety of the robot can be improved. The human-robot minimum distance in the environment can be efficiently tracked in real time.