267results about How to "Takes up little storage space" patented technology

The embodiment of the invention discloses a gait recognition method and a corresponding pedometer. The gait recognition method comprises the following steps: acquiring acceleration signals during wrist movement, wherein the acceleration signals comprise a first sensitive shaft signal and a second sensitive shaft signal; selecting one of the first sensitive shaft signal and the second sensitive shaft signal as the current gait recognition signal according to the moving state of the previous gait; determining whether the current gait is complete or not by judging the conversion between a wave peak section and a wave trough section of the current gait recognition signal based on the dynamic threshold value preset for the previous gait. The gait recognition method is capable of effectively recognizing running and other walking modes, thus improving the step-counting accuracy.

The invention relates to a rotational inertia on-line identification method for an alternating current (AC) permanent magnet synchronous motor servo system and belongs to the technical field of rotational inertia identification for motors. The method aims to solve the problems that an on-line inertia identification method based on the traditional least square method is high in requirements for storage space and arithmetic speed of digital processing chips. The method adopts a load torque identification part to identify a constant load disturbance torque value TL; a system dynamic state judging part is adopted for judging whether rotation speed of a motor is at a dynamic change stage, and when the rotation speed of the motor is at the dynamic change stage and the rate of change of the rotation speed of the motor is higher than a critical change rate for identification of rotational inertia of the motor, a rotational inertia estimation value is obtained through calculation by utilization of a least square estimation method based on recursion; and when the rotational inertia on-line identification process meets requirements of a preset output rotational inertia estimation value and meets a requirement for precision, the rotational inertia estimation value is output and the rotational inertia on-line identification process is completed. The method is used for rotational inertia on-line identification for the motor servo system.

The invention discloses a fast multi-label picture retrieval system and a realization method. The method comprises the following steps: deploying an RPN (Region Proposal Network) for extracting region proposals in a convolutional neural network, extracting region proposal information of pictures, and performing ROI pooling calculation on the region proposal information; after pooling, building a multi-label classification loss function through a fully connected layer according to multi-label information to train the convolutional neural network, and building a weighted three-dimensional loss function to train the convolutional neural network; extracting the hash code of each picture from a picture candidate set through the convolutional neural network after multi-task learning, saving the hash codes to a database, and comparing the hash codes with the hash codes in the database, thus completing picture retrieval. The whole network is trained through multi-task learning of classification and hashing, and therefore, the accuracy of retrieval is ensured. Moreover, the similarity is measured using Hamming distance in the process of retrieval, and the efficiency of retrieval is improved greatly.

The invention discloses data redundancy and file operation methods based on a Hadoop distributed file system (HDFS). The file operation method comprises the processes of file writing, file reading, file adding and file deletion. The file writing process comprises the following steps of: giving a file writing request by a client; allocating main nodes; segmenting a file; generating and storing basic blocks; generating and storing coding blocks; and reporting the main nodes. The file reading process comprises the following steps of: giving a file reading request by the client; selecting the main nodes; sending information data; according to the information data, restoring an original file; and reporting the main nodes. The file adding process comprises the following steps of: giving a file adding request by the client; querying file information; allocating the main nodes; segmenting an added file; generating and storing the basic blocks; generating and storing the coding blocks; and reporting the added file. The file deletion process comprises the following steps of: giving a file deletion request by the client; processing the file name of the file to be deleted; deleting a hidden file; deleting isolated block metadata; and deleting any block. The invention has the advantages of high read-write performance and efficiency and reliability in storage, and can be applied to storage and management of the HDFS under the access of a great number of clients; and storage resources are saved.

According to a ground feature contour extraction method based on a ground laser point cloud provided by the invention, surface-based space measurement breaking through a traditional single-point measurement mode is introduced into urban measurement application, and the urban measurement efficiency is improved. As the most important space entity of a city, the three-dimensional information of ground objects, especially buildings, is the important basic geographic information of the city and has important application value. The whole scanning process is high in data acquisition efficiency, ground control points do not need to be specially arranged, scanning measurement is slightly influenced by the external environment, and the measurement precision is high. Not only is automation degree ofbuilding space data acquisition improved, but also scanned point cloud can express real three-dimensional information and detail structure of a building in detail. Accurate contour information is acquired by means of the point cloud data, and a three-dimensional model of the building is reconstructed. The method has the advantages of high interaction performance, high intelligent degree, high expandability, high extraction speed, high contour extraction precision and the like.

The invention discloses a physical method based on environment mapping for simulating human skin subsurface scattering. The method comprises steps that, a legendre formula is utilized to project and disintegrate an illumination integration equation in a to-be-simulated scene to acquire a luminance value of each pixel; the luminance value of each pixel is multiplied with an overflow color value to acquire a scattering value of each pixel at a back light face. The physical method based on environment mapping for simulating human skin subsurface scattering solves problems of large data processing workload, large occupation storage space and low processing efficiency existing in the prior art and has advantages of small data processing workload, small occupation storage space and high processing efficiency.

The invention provides a training method for a vehicle identification model, comprising the following steps of: collecting image samples, transforming the collected image samples, representing a transformed image sample by a feature matrix with constant size but transformable features, extracting partial features in the feature matrix with constant size but transformable features of each image sample to compose a feature space, performing sparse coding on the feature space to obtain a sparse coding feature basis matrix, computing the maximum multi-scale feature vector of each image sample, and determining the parameter of a linear classifier. By the training method for the vehicle identification model, the complexity, occupied memory space and computing time of an algorithm can be reduced, so that the vehicle identification is implemented fast; and meanwhile, the vehicle identification precision is improved. The invention further provides a vehicle identification method and apparatus based on the training method for the vehicle identification model.

The invention discloses a query and memory space cleaning method based on a cloud server and a query and memory space cleaning system thereof, wherein the memory space cleaning method based on the cloud server comprises the following steps of: scanning the memory space by a mobile terminal and taking a path of a currently scanned catalogue as to-be-queried information and inquiring it in a local database; confirming that information matched with the to-be-queried information is absent from the local database in the mobile terminal, and uploading the to-be-queried information to the server; collecting garbage from corresponding catalogues by the mobile terminal according to the cleaning strategies of the content returned by the server. Through application of the method and the system, the query efficiency can be improved so as to save the memory space and further improve the cleaning efficiency.

The invention relates to an information extraction method, comprising the following steps: a spider software obtains the website information related to the theme of search and carries out word segmentation on the website information to obtain the word segmentation result information including a plurality of words and/or phrases; weighting operation is carried out on the words and/or phrases respectively based on the same semantic attribute parameter in the semantic corpus to obtain the semantic attribute parameter of the website information; and the website information is stored in the extraction result queue if the semantic attribute parameter of the website information is in the preset range of the semantic attribute parameter. The embodiment of the invention ensures high degree of correlation between the result obtained by extraction and the theme of search, improves the working efficiency of the spider software and minishes the storage space occupied by the spider software.

The invention discloses a topological map generation method based on a visual fusion landmarks, and the method comprises the steps: inputting an RGB image, carrying out the semantic segmentation of the RGB image based on a convolutional neural network, and extracting a point with an output value greater than a preset value as a feature point according to a feature map of a semantic segmentation intermediate layer; obtaining a depth map of a scene, based on the depth information of the depth map and the two-dimensional coordinate information of the feature points on the image, acquiring the three-dimensional coordinates of the feature points and the pose information of the robot according to a camera model to establish a three-dimensional map of the scene, performing texture segmentation based on deep learning on the input image, and acquiring codes of each pixel point in a texture feature space; obtaining a membership degree distribution function of the point cloud relative to a predetermined landmark by adopting a landmark level data fusion method based on fuzzy mathematics, and obtaining a three-dimensional semantic map fused with the landmark according to the semantic membershipdegree distribution function and the three-dimensional map; and constructing a topological map based on the three-dimensional semantic map of the fused landmark to generate a semantic topological mapof the fused landmark.

The invention discloses an original and accompanying singing switching method and apparatus, and belongs to the technical field of networks. The method comprises: when receiving an accompaniment play request of a specified song, playing an accompaniment audio file through a first audio track based on a double-audio-track file package of the specified song, wherein the double-audio-track file package comprises the accompaniment audio file and a human voice audio file; and when detecting a first specified operation, playing the human voice audio file through a second audio track, wherein the first specified operation is used for switching from a current accompanying singing mode to an original singing mode. According to the original and accompanying singing switching method and apparatus, the original and accompanying singing switching is realized by controlling the playing of the human voice audio file, and the loading speed is increased by taking the double-audio-track file package generated by the human voice audio file and the accompanying audio file as a play file; and meanwhile, the original and accompanying singing switching method and apparatus also have the characteristics of small occupied storage space and favorability for downloading.

The disclosed method comprises: 1) defining the whole texture direction as the horizontal and vertical frame line in target area by some user texture frame; 2) pre-treating the frame line, constructing the horizontal and vertical texture middle mesh, and uniting all the middle meshes to generate mesh texture; 3) taking interactive regulation to reflect the texture variation led by rumple and torsion on clothes surface; 4) calculating the texture coordinate of every pixel in target area for texture mapping. This invention avoids modeling, needs small calculation and storage quantity, and has well use value.

The invention discloses a magnetic levitation train relative positioning sensor signal processing method and device. The method includes the following steps that S100, a real-time original output signal, disturbed by a joint, of a relative positioning sensor is obtained as a to-be-processed input signal; S200, the input signal is processed through a slip form differentiator, and a tracking signaland a differential signal of the input signal are obtained; and S300, a magnetic pole phase angle signal used for magnetic levitation train motor traction is generated through the tracking signal andthe differential signal of the processed input signal. The method and device have the beneficial effects that distorted signal processing capacity is high, the tracking accuracy is high, the processing speed is high, the calculated amount is small, and the occupied storage space is small. The signal generated when a magnetic levitation train passes a stator rail especially the relative positioningsensor on the portion of the rail joint is processed so that accurate position information of the magnetic levitation train can be supplied to a magnetic levitation train traction system to guide andguarantee safety operation of the magnetic levitation train.

An embedded system language teaching device with an ability of pronunciation quality evaluation comprises a hardware platform and a software system. The software system comprises a system program, an arithmetic program and an application program, and operates on a voice processing chip of the hardware platform. The arithmetic program comprises a coder-decoder program of voice and a pronunciation quality evaluation arithmetic program. The idiographic program module comprises audio data play of teaching content, taping, replaying tape, audio speed adjustment and pronunciation quality evaluation. The embedded system language teaching device with the ability of the pronunciation quality evaluation resolves the disadvantages of the prior teaching device, thus not only having the ability to implement the prior teaching device functions such as learning words and sentences, reading, exercise and test and so on, but also integrating the pronunciation quality evaluation function of learners and interactive teaching thought.

The invention discloses a data simplified duplicate removal and mirror image remote backup protection method and system. The method comprises the steps of firstly segmenting to-be-stored data files into multiple data blocks; secondly performing duplicate checking on the segmented data blocks to delete duplicated data blocks; and thirdly backing up the processed data in a data storage pool, and backing up the storage pool in another mirror image storage pool. Compared with the prior art, the data simplified duplicate removal and mirror image remote backup protection method and system has the advantages that unique data is effectively protected after duplicated data in a production server is deleted; the method and the system are suitable for backup in medium and small-sized enterprises; thestorage space occupied for duplicated data transmission can be effectively reduced; the storage cost is reduced; the practicality is high; the application range is wide; and the popularization is easy.

The invention belongs to the technical field of DCS graphical algorithm configuration, In order to solve the technical problem of low system performance in the technical scheme of off-line simulationby calling GDB/MI interface, the invention provides an off-line simulation method and device of DCS graphical algorithm configuration, which realizes data access and algorithm control off-line simulation function through self-encapsulating simulation tool plug-in. The off-line simulation method comprises the following steps: (1) compiling a graphics algorithm corresponding to the graphics algorithm configuration into an algorithm program and an algorithm offset program; 2, load that algorithm program and the algorithm offset program into the memory by using an offline simulation plug-in, and realize simulation control and variable data operation by using the algorithm program and the algorithm offset program.

The invention provides a video image 3D noise reduction method, a computer device and a computer readable storage medium. The method comprises the steps of obtaining a current video frame and a previous video noise reduction frame; Performing time domain filtering according to the current video frame and the previous video noise reduction frame to obtain a time domain filtering weight of each pixel point of the current video frame; Performing spatial domain filtering based on direction statistics on the current video frame to obtain a spatial domain filtering result of each pixel point of thecurrent video frame; And carrying out weighted fusion on the spatial domain filtering result and the previous video noise reduction frame according to the time domain filtering weight to obtain a video noise reduction frame of the current video frame. The computer device is provided with a processor, and when the processor executes a program, the video image 3D noise reduction method can be achieved. A computer program is stored in the computer readable storage medium and used for achieving the video image 3D noise reduction method. By applying the method, common Gaussian noise and compressionnoise of the video are remarkably reduced, and meanwhile few hardware resources such as storage space and bandwidth are occupied.

The invention provides a synchronous transmission control method and device for a digital retina video stream and a feature stream, electronic equipment, network video monitoring equipment and a computer readable medium. The method comprises the following steps: acquiring a monitoring video stream; generating a concentrated video stream corresponding to the monitoring video stream by adopting a video concentration technology; extracting feature information in the monitoring video stream to form a feature stream; and respectively uploading the concentrated video stream and the feature stream toa cloud server. Because the concentrated video stream has the advantages of small occupied storage space, high fidelity factor and the like, compared with a mode of uploading all monitoring video streams to a cloud server in the prior art, the method has the advantages of small occupied storage space, low network transmission load, high transmission efficiency and the like; meanwhile, the featurestream is extracted according to the monitoring video stream and uploaded to the cloud server, thereby facilitating the quick recognition of the positioning target and the mining of the movement track of the positioning target, and effectively improving the retrieval efficiency of the monitoring video.

The invention relates to a face identification method based on JSM (joint sparsity models) and sparsity preserving projection. According to the method, a transformation base formed by all training images substitutes a common random matrix in the JSM algorithm; public parts and privacy parts of each type of training face images are extracted by utilizing the JSM algorithm; images in a face database are classified as per people; all images of one person are grouped into the same class; the public parts indicate common face characteristics of each type of face images; the privacy parts indicate detail changes on expressions of faces, illumination and the like; images reconstructed by utilizing sparse public and privacy parts are approximate to original training images; and a dimensionality reduction matrix is obtained through solving the most optimized problem with the smallest reconstruction error. The dimensionality reduction matrix obtained finally performs dimensionality reduction treatment on testing images, the public parts and the privacy parts of each type of training images subjected to dimensionality reduction are used for reconstructing the testing images, and the testing images belong to the class of training images with the smallest reconstruction error.

The invention relates to a character two-dimension code encoding and decoding method for a short message, which is used for solving the problems that the storage space occupation of a two-dimension code based on an image is large and the transmission efficiency is low, and has the advantages of capability of storing bar codes by characters, efficient transmission, and simple and efficient encoding and decoding algorithms. A character two-dimension code is a rectangular-shaped character two-dimension matrix module and the matrix module is composed of view finding sign units, a version number unit, a data specification number unit and a data region sign unit, wherein three view finding sign units are arranged and respectively located at the upper left corner, a upper right corner and the lower right corner of the matrix module; the version number unit is located in back of the view finding sign unit which is located at the upper left corner of the matrix module; the data specification number unit is located in back of the version number unit; and the data region sign unit is located in back of the data specification number unit. According to the character two-dimension code encoding and decoding method for the short message, character two-dimension bar codes are widely applied to electronic commerce.