CN103345382A - CPU+GPU group nuclear supercomputer system and SIFT feature matching parallel computing method - Google Patents
CPU+GPU group nuclear supercomputer system and SIFT feature matching parallel computing method Download PDFInfo
- Publication number
- CN103345382A CN103345382A CN2013102962389A CN201310296238A CN103345382A CN 103345382 A CN103345382 A CN 103345382A CN 2013102962389 A CN2013102962389 A CN 2013102962389A CN 201310296238 A CN201310296238 A CN 201310296238A CN 103345382 A CN103345382 A CN 103345382A
- Authority
- CN
- China
- Prior art keywords
- gpu
- cpu
- data
- key point
- sift
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a CPU+GPU group nuclear supercomputer system and an SIFT feature matching parallel computing method. A CPU is used for interpreting instructions of a computer, processing data in computer software and carrying out high-performance computing. A north bridge chip unit is used for providing support for the type and basic frequency of the CPU, the type and maximum capacity of an internal storage, ISA/PCI/AGP slots, ECC error correction and the like. A system storage unit is used for storing information of the data. A graphics adaptor unit is used for controlling the output of computer graphs. A GPU module is installed on the graphics adaptor unit and used for carrying out a large amount of simple parallel computing and drawing the data into graphs. A GPU storage module is used for storing collected data. Through the parallel analysis, much computing is divided and performed in the CPU and the GPU, the respective computing advantage is played, the speed of the SIFT feature matching GPU parallel algorithm is improved by nearly 30 times compared with the speed of a CPU series algorithm, the time of data processing is shortened greatly, the real-time performance is improved, and extracting and matching of remote sensing image feature points are achieved.
Description
Technical field
The invention belongs to the space remote sensing photography, relate in particular to a kind of CPU+GPU group and examine super calculation system and SIFT characteristic matching parallel calculating method.
Background technology
Current, the development of photogrammetric and computer vision, characteristics of image coupling become the important foundation that the basic problem of area researches such as Photogrammetry and Remote Sensing, resource analysis, 3-D reconstruction, computer vision and pattern-recognition and object identification, tracking etc. are used all the more.For several remote sensing images of Same Scene, may there be multiple difference between them: different resolution, gray scale attribute, position (translation and rotation), engineer's scale, nonlinear deformation etc.Traditional feature point detection algorithm is difficult for designing the unique point that a large amount of templates are mated the texture exquisiteness as the feature point detection algorithm based on template matches; Feature point detection arithmetic accuracy based on rim detection is not high; The feature point detection algorithm that changes based on brightness is subjected to noise, and the influence of illumination is very big.Also there is a common problem in existing image matching method based on feature at present: the unchangeability of their unique points that adopts is generally relatively poor, does not particularly have usually unchangeability affine or the transmission projection conversion.Therefore, the coupling with remote sensing image of larger difference and feature distortion remains a difficult point problem.
(Scale Invariant Feature Transform, SIFT) algorithm is a kind of based on the theoretical efficient algorithm that extracts image local feature of metric space to the conversion of yardstick invariant features.By at difference of Gaussian metric space (Difference Of Gaussian, DOG) seek extreme point as key point, the SIFT feature all remains unchanged to yardstick convergent-divergent, rotation, illumination variation, affined transformation, viewpoint change, noise all there is to a certain degree stability, but the SIFT characteristic matching is handled the problem that data volume is big, computation complexity is high and operand is big that faces, in the application scenario that real-time is had relatively high expectations, application is restricted.
Summary of the invention
The object of the present invention is to provide a kind of SIFT characteristic matching software systems towards the CPU+GPU framework, be intended to solve the SIFT characteristic matching and handle the problem that data volume is big, computation complexity is high and operand is big that faces.
The present invention is achieved in that a kind of CPU+GPU group examines super calculation system, and this structure comprises CPU element, north bridge chips unit, system storage unit, graphics adapter unit:
CPU element is used for the data of interpretive machine instruction and process computer software as the subsystem of primary processor, carries out high performance calculating;
The north bridge chips unit links to each other with CPU element by Front Side Bus, is used for providing to supports such as the type of the type of CPU and dominant frequency, internal memory and max cap., ISA/PCI/AGP slot, ECC error correction;
The system storage unit links to each other with the north bridge chips unit by memory bus, is used for the information of storage data;
The graphics adapter unit links to each other with the north bridge chips unit by the PCI-Express bus, is used for the output of control computer graphical;
The GPU module is installed on the graphics adapter unit, is the core of graphics adapter, is used for carrying out a large amount of simple parallel computations and data being drawn as figure;
The GPU memory module links to each other with the GPU module by dma operation, is used for the data that store collected arrives.
Further, CPU is four nuclears;
Further, CPU+GPU group examines in the super calculation system and comprises two graphics adapter;
Further, software systems must operate on the PC with NVIDIA display adapter that GeForce8 series is above, Quadro is serial of supporting CUDA.
The present invention is achieved in that a kind of SIFT characteristic matching parallel calculating method, and the detailed process of this method is as follows:
Step 1: read the input image to main memory, be passed to the gaussian kernel data of different scale in the constant storage of GPU by main memory;
Step 2: utilize the set of keypoints information of in GPU, calculating, start gradient magnitude and direction that kernel calculates near pixel key point;
Step 3: the SIFT descriptor that calculates 128 dimensions;
Step 4: carry out the SIFT characteristic matching, determine the match point position.
Further, in GPU, carry out the substep continuous filtering and accelerate the pyramidal structure of Gauss's metric space, and gaussian pyramid is stored in the global storage.To CPU, two Gauss's images of adjacent yardstick subtract each other and obtain DOG pyramid multiscale space and represent the gaussian pyramid image by retaking of a year or grade then.After being uploaded to candidate feature point set information among the GPU, just can in GPU, accurately locate all candidate feature points of difference of Gaussian pyramid.Calculate the key point principal curvatures of image intensity on every side, by one 2 * 2 Hessian matrix computations eigenwert ratio, detect the key point principal curvatures and whether surpass preset threshold, after removing unnecessary point, determine the position of set of keypoints and precise marking key point, yardstick.Key point position, yardstick will recover in GPU;
Further, utilize the Gauss's weighting function be stored in the global storage, the gradient magnitude of each pixel in the key point neighborhood window is carried out Gauss's weighting and added up and set up direction histogram, detect histogrammic peak value, determine key point principal direction;
Further, 16 * 16 image data pieces centered by key point are according to the process of yardstick, position and the directional structure vectorical structure SIFT descriptor of key point, realize then can bringing into play cpu logic branch, advantage that judgement is strong at CPU;
Further, determine that the match point position reads in shared storage according to the natural order of original point with dimension data, the emphasis of optimization is that distance is calculated.Square accumulate mode that is calculated with number of dimensions purpose intermediate result at synchronization that guarantees each dimension difference is efficient.
SIFT characteristic matching software systems towards the CPU+GPU framework provided by the invention based on the SIFT characteristic matching algorithm of GPU by the concurrency analysis, many computed segmentation are calculated respectively between CPU and GPU, bring into play calculating advantage separately, demonstrated fully the great ability of CPU+GPU Heterogeneous Computing.Make SIFT characteristic matching GPU parallel algorithm improve nearly 30 times than CPU serial algorithm speed, significantly shorten the time that data are handled, real-time is greatly improved, and realizes remote sensing image feature point extraction and coupling.
Description of drawings
Fig. 1 is that CPU+GPU group provided by the invention examines the super structural drawing of calculating system logic;
1, CPU element; 2, north bridge chips unit; 3, system storage unit; 4, graphics adapter Unit 1,4-1, GPUl module, 4-2, GPUl memory module; 5, graphics adapter Unit 2,5-1, GPU2 module, 5-2, GPU2 memory module
Fig. 2 is the process flow diagram of SIFT characteristic matching parallel calculating method provided by the invention;
Embodiment
The necessary technology scheme:
The present invention is achieved in that l by reference to the accompanying drawings, and a kind of CPU+GPU group examines super calculation system, and this structure comprises CPU element, north bridge chips unit, system storage unit, figure orchestration unit:
CPU element is used for the data of interpretive machine instruction and process computer software as the subsystem of primary processor, carries out high performance calculating;
The north bridge chips unit links to each other with CPU element by Front Side Bus, is used for providing to supports such as the type of the type of CPU and dominant frequency, internal memory and max cap., ISA/PCI/AGP slot, ECC error correction;
The system storage unit links to each other with the north bridge chips unit by memory bus, is used for the information of storage data;
The graphics adapter unit links to each other with the north bridge chips unit by the PCI-Express bus, is used for the output of control computer graphical;
The GPU module is installed on the graphics adapter unit, is the core of graphics adapter, is used for carrying out a large amount of simple parallel computations and data being drawn as figure;
The GPU memory module links to each other with the GPU module by dma operation, is used for the data that store collected arrives.
Further, CPU is four nuclears;
Further, CPU+GPU group examines in the super calculation system and comprises two graphics adapter;
Further, software systems must operate on the PC with NVIDIA display adapter that GeForce8 series is above, Quadro is serial of supporting CUDA.
The present invention is achieved in that by reference to the accompanying drawings 2, a kind of SIFT characteristic matching parallel calculating method, and the detailed process of this method is as follows:
S201: read the input image to main memory, be passed to the gaussian kernel data of different scale in the constant storage of GPU by main memory;
S202: utilize the set of keypoints information of in GPU, calculating, start gradient magnitude and direction that kernel calculates near pixel key point;
S203: the SIFT descriptor that calculates 128 dimensions;
S204: carry out the SIFT characteristic matching, determine the match point position.
Further, in GPU, carry out the substep continuous filtering and accelerate the pyramidal structure of Gauss's metric space, and gaussian pyramid is stored in the global storage.To CPU, two Gauss's images of adjacent yardstick subtract each other and obtain DOG pyramid multiscale space and represent the gaussian pyramid image by retaking of a year or grade then.After being uploaded to candidate feature point set information among the GPU, just can in GPU, accurately locate all candidate feature points of difference of Gaussian pyramid.Calculate the key point principal curvatures of image intensity on every side, by one 2 * 2 Hessian matrix computations eigenwert ratio, detect the key point principal curvatures and whether surpass preset threshold, after removing unnecessary point, determine the position of set of keypoints and precise marking key point, yardstick.Key point position, yardstick will recover in GPU;
Further, utilize the Gauss's weighting function be stored in the global storage, the gradient magnitude of each pixel in the key point neighborhood window is carried out Gauss's weighting and added up and set up direction histogram, detect histogrammic peak value, determine key point principal direction;
Further, 16 * 16 image data pieces centered by key point are according to the process of yardstick, position and the directional structure vectorical structure SIFT descriptor of key point, realize then can bringing into play cpu logic branch, advantage that judgement is strong at CPU;
Further, determine that the match point position reads in shared storage according to the natural order of original point with dimension data, the emphasis of optimization is that distance is calculated.Square accumulate mode that is calculated with number of dimensions purpose intermediate result at synchronization that guarantees each dimension difference is efficient.
A kind of SIFT characteristic matching software systems towards the CPU+GPU framework are passed through the parallelization data structure in analysis and the problem definition, calculation task is described to the mapping mechanism of CUDA (Compute Unified Device Architecture), problem or algorithm are divided into a plurality of subtasks, and the subtask of dividing provided rational dispatching algorithm, make GPU and CPU bring into play characteristics separately, thereby obtain the GPU general-purpose computations ability of greater efficiency.The detailed process of a kind of SIFT characteristic matching parallel computation is at first to read the input image to main memory, is passed to the gaussian kernel data of different scale in the constant storage of GPU by main memory; The set of keypoints information that utilization is calculated in GPU starts gradient magnitude and direction that kernel calculates near pixel key point; Calculate the SIFT descriptor of 128 dimensions; Carry out the SIFT characteristic matching at last, determine the match point position.
The above only is preferred embodiment of the present invention, not in order to limiting the present invention, all any modifications of doing within the spirit and principles in the present invention, is equal to and replaces and improvement etc., all should be included within protection scope of the present invention.
Claims (9)
1. a CPU+GPU group examines super calculation system, it is characterized in that, this CPU+GPU group examines super calculation system and comprises CPU element, north bridge chips unit, system storage unit, graphics adapter unit:
CPU element is used for the data that interpretive machine refers to the present and process computer software as the subsystem of primary processor, calculates;
The north bridge chips unit links to each other with CPU element by Front Side Bus, is used for providing to the type of the type of CPU and dominant frequency, internal memory and max cap., ISA/PCI/AGP slot, ECC error correction support;
The system storage unit links to each other with the north bridge chips unit by memory bus, is used for the information of storage data;
The graphics adapter unit links to each other with the north bridge chips unit by the PCI-Express bus, is used for the output of control computer graphical;
The GPU module is installed on the graphics adapter unit, is the core of graphics adapter, is used for carrying out a large amount of simple parallel computations and data being drawn as figure;
The GPU memory module links to each other with the GPU module by dma operation, is used for the data that store collected arrives.
2. CPU+GPU group as claimed in claim 1 examines super calculation system, it is characterized in that, CPU is four nuclears.
3. CPU+GPU group as claimed in claim 1 examines super calculation system, it is characterized in that, CPU+GPU group examines in the super calculation system and comprises two graphics adapter.
4. CPU+GPU group as claimed in claim 1 examines super calculation system, it is characterized in that, software systems must operate on the PC with NVIDIA display adapter that GeForce8 series is above, Quadro is serial of supporting CUDA.
5. SIFT characteristic matching parallel calculating method is characterized in that the detailed process of this method is as follows:
Step 1: read the input image to main memory, be passed to the gaussian kernel data of different scale in the constant storage of GPU by main memory;
Step 2: utilize the set of keypoints information of in GPU, calculating, start gradient magnitude and direction that kernel calculates near pixel key point;
Step 3: the SIFT descriptor that calculates 128 dimensions;
Step 4: carry out the SIFT characteristic matching, determine the match point position.
6. SIFT characteristic matching parallel calculating method as claimed in claim 5 is characterized in that, carries out the substep continuous filtering and accelerate the pyramidal structure of Gauss's metric space in GPU, and gaussian pyramid is stored in the global storage; To CPU, two Gauss's images of adjacent yardstick subtract each other and obtain DOG pyramid multiscale space and represent the gaussian pyramid image by retaking of a year or grade then; After being uploaded to candidate feature point set information among the GPU, just can in GPU, accurately locate all candidate feature points of difference of Gaussian pyramid, calculate the key point principal curvatures of image intensity on every side, Hessian matrix computations eigenwert ratio by one 2 * 2, detect the key point principal curvatures and whether surpass preset threshold, after removing unnecessary point, determine the position of set of keypoints and precise marking key point, yardstick, key point position, yardstick will recover in GPU.
7. SIFT characteristic matching parallel calculating method as claimed in claim 5, it is characterized in that, utilize the Gauss's weighting function that has been stored in the global storage, the gradient magnitude of each pixel in the key point neighborhood window is carried out Gauss's weighting and added up and set up direction histogram, detect histogrammic peak value, determine key point principal direction.
8. SIFT characteristic matching parallel calculating method as claimed in claim 5 is characterized in that, 16 * 16 image data pieces centered by key point are realized at CPU according to the process of yardstick, position and the directional structure vectorical structure SIFT descriptor of key point.
9. SIFT characteristic matching parallel calculating method as claimed in claim 5, it is characterized in that, determine that the match point position reads in shared storage according to the natural order of original point with dimension data, the emphasis of optimizing is that distance is calculated, and guarantees that square accumulate mode that is calculated with number of dimensions purpose intermediate result at synchronization of each dimension difference is efficient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013102962389A CN103345382A (en) | 2013-07-15 | 2013-07-15 | CPU+GPU group nuclear supercomputer system and SIFT feature matching parallel computing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013102962389A CN103345382A (en) | 2013-07-15 | 2013-07-15 | CPU+GPU group nuclear supercomputer system and SIFT feature matching parallel computing method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103345382A true CN103345382A (en) | 2013-10-09 |
Family
ID=49280180
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2013102962389A Pending CN103345382A (en) | 2013-07-15 | 2013-07-15 | CPU+GPU group nuclear supercomputer system and SIFT feature matching parallel computing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103345382A (en) |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103577160A (en) * | 2013-10-17 | 2014-02-12 | 江苏科技大学 | Characteristic extraction parallel-processing method for big data |
CN103956991A (en) * | 2014-04-10 | 2014-07-30 | 北京遥测技术研究所 | FIR filter parallel realization method based on CPU/GPU heterogeneous platform |
CN104121942A (en) * | 2014-07-08 | 2014-10-29 | 哈尔滨工业大学 | Automobile instrument automatic detection device based on graphic processing unit (GPU) and open CV image processing |
CN104268169A (en) * | 2014-09-11 | 2015-01-07 | 浙江中测新图地理信息技术有限公司 | Remote sensing image data rapidly processing method based on PS (Photoshop) |
CN104732221A (en) * | 2015-03-30 | 2015-06-24 | 郑州师范学院 | SIFT feature matching method based on OpenCL parallel acceleration |
CN105550085A (en) * | 2015-12-10 | 2016-05-04 | 浪潮电子信息产业股份有限公司 | RDMA (Remote Direct Memory Access) testing method based on GPUDerect (Graphics Processing Unit Derict) |
CN105892931A (en) * | 2014-05-16 | 2016-08-24 | 上海京知信息科技有限公司 | heterogeneous CPU-GPU system configuration based on intelligent flash cache |
CN106502956A (en) * | 2016-10-28 | 2017-03-15 | 张军 | A kind of operating system prototype of multinuclear isomery CPU GPU system frameworks |
CN106845490A (en) * | 2016-12-30 | 2017-06-13 | 国网河南省电力公司检修公司 | Based on the super-pressure converting equipment infrared image localization method that GPU accelerates |
CN107358238A (en) * | 2017-03-01 | 2017-11-17 | 郑州云海信息技术有限公司 | A kind of method and system for extracting image feature information |
CN107657599A (en) * | 2017-08-07 | 2018-02-02 | 北京航空航天大学 | Remote sensing image fusion system in parallel implementation method based on combination grain division and dynamic load balance |
CN108275530A (en) * | 2018-01-18 | 2018-07-13 | 柯钢 | A kind of elevator safety method for early warning based on machine learning |
US10269148B2 (en) | 2017-05-01 | 2019-04-23 | Lockheed Martin Corporation | Real-time image undistortion for incremental 3D reconstruction |
US10269147B2 (en) | 2017-05-01 | 2019-04-23 | Lockheed Martin Corporation | Real-time camera position estimation with drift mitigation in incremental structure from motion |
CN109801207A (en) * | 2019-01-08 | 2019-05-24 | 桂林电子科技大学 | The image feature high speed detection and matching system of CPU-FPGA collaboration |
CN109829864A (en) * | 2019-01-30 | 2019-05-31 | 北京达佳互联信息技术有限公司 | Image processing method, device, equipment and storage medium |
CN110414534A (en) * | 2019-07-01 | 2019-11-05 | 深圳前海达闼云端智能科技有限公司 | Image feature extraction method and device, storage medium and electronic equipment |
CN111024078A (en) * | 2019-11-05 | 2020-04-17 | 广东工业大学 | Unmanned aerial vehicle vision SLAM method based on GPU acceleration |
CN111445503A (en) * | 2020-03-25 | 2020-07-24 | 桂林电子科技大学 | Pyramid mutual information image registration method based on parallel programming model on GPU cluster |
CN111898589A (en) * | 2020-08-26 | 2020-11-06 | 中国水利水电科学研究院 | Unmanned aerial vehicle image rapid registration method based on GPU + feature recognition |
CN111913798A (en) * | 2020-07-09 | 2020-11-10 | 太原理工大学 | Fast non-overlapping template matching calculation method based on CUDA |
CN113486729A (en) * | 2021-06-15 | 2021-10-08 | 北京道达天际科技有限公司 | Unmanned aerial vehicle image feature point extraction method based on GPU |
CN117437451A (en) * | 2023-12-21 | 2024-01-23 | 芯瞳半导体技术(山东)有限公司 | Image matching method, device, equipment and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101681192A (en) * | 2007-05-09 | 2010-03-24 | 松下电器产业株式会社 | Signal processor and signal processing system |
US20120066664A1 (en) * | 2010-09-13 | 2012-03-15 | Massively Parallel Technologies, Inc. | Software design and automatic coding for parallel computing |
-
2013
- 2013-07-15 CN CN2013102962389A patent/CN103345382A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101681192A (en) * | 2007-05-09 | 2010-03-24 | 松下电器产业株式会社 | Signal processor and signal processing system |
US20120066664A1 (en) * | 2010-09-13 | 2012-03-15 | Massively Parallel Technologies, Inc. | Software design and automatic coding for parallel computing |
Non-Patent Citations (1)
Title |
---|
肖汉: "基于CPU+GPU的影像匹配高效能异构并行计算研究", 《中国博士学位论文全文数据库(电子期刊)》, no. 04, 15 April 2012 (2012-04-15) * |
Cited By (33)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103577160A (en) * | 2013-10-17 | 2014-02-12 | 江苏科技大学 | Characteristic extraction parallel-processing method for big data |
CN103956991A (en) * | 2014-04-10 | 2014-07-30 | 北京遥测技术研究所 | FIR filter parallel realization method based on CPU/GPU heterogeneous platform |
CN103956991B (en) * | 2014-04-10 | 2017-09-29 | 北京遥测技术研究所 | A kind of FIR filtering Parallel Realization methods based on CPU/GPU heterogeneous platforms |
CN105892931A (en) * | 2014-05-16 | 2016-08-24 | 上海京知信息科技有限公司 | heterogeneous CPU-GPU system configuration based on intelligent flash cache |
CN104121942A (en) * | 2014-07-08 | 2014-10-29 | 哈尔滨工业大学 | Automobile instrument automatic detection device based on graphic processing unit (GPU) and open CV image processing |
CN104268169B (en) * | 2014-09-11 | 2017-06-09 | 浙江中测新图地理信息技术有限公司 | A kind of remote sensing image data immediate processing method based on PS softwares |
CN104268169A (en) * | 2014-09-11 | 2015-01-07 | 浙江中测新图地理信息技术有限公司 | Remote sensing image data rapidly processing method based on PS (Photoshop) |
CN104732221A (en) * | 2015-03-30 | 2015-06-24 | 郑州师范学院 | SIFT feature matching method based on OpenCL parallel acceleration |
CN105550085A (en) * | 2015-12-10 | 2016-05-04 | 浪潮电子信息产业股份有限公司 | RDMA (Remote Direct Memory Access) testing method based on GPUDerect (Graphics Processing Unit Derict) |
CN106502956A (en) * | 2016-10-28 | 2017-03-15 | 张军 | A kind of operating system prototype of multinuclear isomery CPU GPU system frameworks |
CN106845490A (en) * | 2016-12-30 | 2017-06-13 | 国网河南省电力公司检修公司 | Based on the super-pressure converting equipment infrared image localization method that GPU accelerates |
CN107358238A (en) * | 2017-03-01 | 2017-11-17 | 郑州云海信息技术有限公司 | A kind of method and system for extracting image feature information |
US10269148B2 (en) | 2017-05-01 | 2019-04-23 | Lockheed Martin Corporation | Real-time image undistortion for incremental 3D reconstruction |
US10269147B2 (en) | 2017-05-01 | 2019-04-23 | Lockheed Martin Corporation | Real-time camera position estimation with drift mitigation in incremental structure from motion |
CN107657599A (en) * | 2017-08-07 | 2018-02-02 | 北京航空航天大学 | Remote sensing image fusion system in parallel implementation method based on combination grain division and dynamic load balance |
CN107657599B (en) * | 2017-08-07 | 2020-11-03 | 北京航空航天大学 | Parallel implementation method of remote sensing image fusion system based on mixed granularity division and dynamic load distribution |
CN108275530A (en) * | 2018-01-18 | 2018-07-13 | 柯钢 | A kind of elevator safety method for early warning based on machine learning |
CN109801207A (en) * | 2019-01-08 | 2019-05-24 | 桂林电子科技大学 | The image feature high speed detection and matching system of CPU-FPGA collaboration |
CN109829864A (en) * | 2019-01-30 | 2019-05-31 | 北京达佳互联信息技术有限公司 | Image processing method, device, equipment and storage medium |
CN109829864B (en) * | 2019-01-30 | 2021-05-18 | 北京达佳互联信息技术有限公司 | Image processing method, device, equipment and storage medium |
CN110414534B (en) * | 2019-07-01 | 2021-12-03 | 达闼机器人有限公司 | Image feature extraction method and device, storage medium and electronic equipment |
CN110414534A (en) * | 2019-07-01 | 2019-11-05 | 深圳前海达闼云端智能科技有限公司 | Image feature extraction method and device, storage medium and electronic equipment |
CN111024078B (en) * | 2019-11-05 | 2021-03-16 | 广东工业大学 | Unmanned aerial vehicle vision SLAM method based on GPU acceleration |
CN111024078A (en) * | 2019-11-05 | 2020-04-17 | 广东工业大学 | Unmanned aerial vehicle vision SLAM method based on GPU acceleration |
CN111445503A (en) * | 2020-03-25 | 2020-07-24 | 桂林电子科技大学 | Pyramid mutual information image registration method based on parallel programming model on GPU cluster |
CN111445503B (en) * | 2020-03-25 | 2023-04-25 | 桂林电子科技大学 | Pyramid mutual information image registration method based on parallel programming model on GPU cluster |
CN111913798A (en) * | 2020-07-09 | 2020-11-10 | 太原理工大学 | Fast non-overlapping template matching calculation method based on CUDA |
CN111913798B (en) * | 2020-07-09 | 2024-02-09 | 太原理工大学 | CUDA-based rapid non-overlapping template matching calculation method |
CN111898589A (en) * | 2020-08-26 | 2020-11-06 | 中国水利水电科学研究院 | Unmanned aerial vehicle image rapid registration method based on GPU + feature recognition |
CN111898589B (en) * | 2020-08-26 | 2023-11-14 | 中国水利水电科学研究院 | Unmanned aerial vehicle image rapid registration method based on GPU+feature recognition |
CN113486729A (en) * | 2021-06-15 | 2021-10-08 | 北京道达天际科技有限公司 | Unmanned aerial vehicle image feature point extraction method based on GPU |
CN117437451A (en) * | 2023-12-21 | 2024-01-23 | 芯瞳半导体技术(山东)有限公司 | Image matching method, device, equipment and storage medium |
CN117437451B (en) * | 2023-12-21 | 2024-04-16 | 芯瞳半导体技术(山东)有限公司 | Image matching method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103345382A (en) | CPU+GPU group nuclear supercomputer system and SIFT feature matching parallel computing method | |
CN108228798B (en) | Method and device for determining matching relation between point cloud data | |
CN111311650B (en) | Point cloud data registration method, device and storage medium | |
CN110097582B (en) | Point cloud optimal registration and real-time display system and working method | |
CN102880509B (en) | Compute unified device architecture (CUDA) based grid digital elevation model (DEM) neighborhood analysis system and method | |
CN107358238A (en) | A kind of method and system for extracting image feature information | |
CN110751676A (en) | Heterogeneous computing system and method based on target detection and readable storage medium | |
CN102968785A (en) | High-speed parallel image matching method based on multi-core digital signal processor (DSP) | |
CN110837839B (en) | High-precision unmanned aerial vehicle orthographic image manufacturing and data acquisition method | |
CN108716916B (en) | Distributed parallel star point centroid extraction method based on superblock and FPGA implementation device | |
CN101968895A (en) | Two-dimensional image conversion system and method | |
JP6128617B2 (en) | Image recognition apparatus and program | |
CN109522906A (en) | The quick SIFT feature extracting method of low complex degree based on FPGA | |
CN115330940A (en) | Three-dimensional reconstruction method, device, equipment and medium | |
CN111161348B (en) | Object pose estimation method, device and equipment based on monocular camera | |
Li et al. | A new automatic real-time crop row recognition based on SoC-FPGA | |
CN103761499A (en) | Barcode identification method based on multi-core DSP | |
Pohl et al. | An efficient FPGA-based hardware framework for natural feature extraction and related computer vision tasks | |
KR20110139864A (en) | System and method for face detection | |
CN115294277B (en) | Three-dimensional reconstruction method and device of object, electronic equipment and storage medium | |
Zhang et al. | Efficient large-scale oblique image matching based on cascade hashing and match data scheduling | |
Han et al. | A MapReduce approach for SIFT feature extraction | |
Askari et al. | Parallel gpu implementation of hough transform for circles | |
CN113822838A (en) | Base recognition apparatus and base recognition method | |
CN106326846B (en) | The forest plant parallel extraction method of unmanned plane image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20131009 |
|
RJ01 | Rejection of invention patent application after publication |