CN103514594A - Method for detecting acceleration corner and mobile device for executing method - Google Patents
Method for detecting acceleration corner and mobile device for executing method Download PDFInfo
- Publication number
- CN103514594A CN103514594A CN201210219887.4A CN201210219887A CN103514594A CN 103514594 A CN103514594 A CN 103514594A CN 201210219887 A CN201210219887 A CN 201210219887A CN 103514594 A CN103514594 A CN 103514594A
- Authority
- CN
- China
- Prior art keywords
- corner detection
- dimensional image
- mobile device
- drawing chip
- corner
- 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
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000001133 acceleration Effects 0.000 title abstract description 6
- 238000001514 detection method Methods 0.000 claims abstract description 51
- 238000012545 processing Methods 0.000 claims description 25
- 238000004040 coloring Methods 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 10
- 238000002474 experimental method Methods 0.000 description 10
- 230000000052 comparative effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 2
- 230000003190 augmentative effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
A method for detecting acceleration corner points and a mobile device for executing the method are provided, wherein the method for detecting the acceleration corner points is executed on the mobile device by utilizing a drawing chip and comprises the following steps: the drawing chip reads a two-dimensional image and carries out corner detection on the two-dimensional image in a multi-core parallel operation mode according to a second program code written by an open graphic library coloring language so as to find out corners in the two-dimensional image. The invention utilizes the characteristic of multi-core parallel operation of the drawing chip to execute the corner detection algorithm in a multi-core way to simultaneously detect a plurality of image points so as to carry out corner detection, and can accelerate the speed of image identification of the mobile device.
Description
Technical field
The present invention relates to a kind of method of accelerating Corner Detection, particularly relate to a kind of mobile device that utilizes the method for drawing chip acceleration Corner Detection and carry out the method on mobile device.
Background technology
In image recognition, Corner Detection is very important technology, and is widely used in the related application of processing in various images.Corner Detection is used in such as 3D and rebuilds in the application such as (3D reconstruction), object identification (object recognition), image mosaic (image mosaicing), motive objects tracking (motion tracking).And universal along with mobile device, the technology of image recognition is also often used on mobile device platform.
Since iPhone listing in 2008, intelligent mobile device becomes the product that people are indispensable already.The life of pressing close to people due to mobile device, and there is the characteristic of carrying, therefore, also diversification more of the application of image recognition on mobile device, for example utilize recognition of face release mobile phone, AR augmented reality (Augmented Reality) game etc., and current mobile device is all with central processing unit (CPU, Central Processing Unit) carry out a large amount of computings of image recognition processing, but because the speed of the CPU of mobile device is so fast unlike the speed of the CPU of PC, therefore the speed of mobile device recognition image be subject to the restriction of its CPU speed.
And drawing chip (GPU) in the display card of PC is mainly as 3D rendering drawing computing, and its arithmetic speed CPU is more a lot of than fast.
So, in order to solve the limited problem of CPU calculation function, at present existing dealer uses drawing chip (Graphics Processing Unit) (GPU, required a large amount of computings to be processed while graphics processing units) assisting image recognition, for example, in well-known open computer vision routine library (OpenCV) latest edition, the program of part provides the function of carrying out computation assistance with drawing chip, wherein topmost method is to use GPU general-purpose computations (GPGPU, General-purpose computing on graphics processing units) CUDA version (unified calculation device architecture) (Compute Unified Device Architecture), assist a large amount of computings in image recognition.So-called CUDA version, refers to and allows developer with high level language as language such as C or C#, control the version of drawing chip.
Yet, these methods of using GPGPU to accelerate are only applicable to up-to-date (future) three-dimensional (3D) drawing chip, for example NVIDIA GeForce GTX is serial, and can only be applied at present on PC (PC) platform, does not support mobile device.Therefore, existing mobile device still can only carry out image recognition processing by CPU, thereby cannot promote the speed of image recognition.Therefore, how lifting mobile device, at the execution speed of the Corner Detection of image recognition, just becomes the theme of the present invention's research.
Summary of the invention
Therefore, the object of the present invention is to provide a kind of method of utilizing drawing chip to accelerate Corner Detection on mobile device.So the present invention accelerates the method for Corner Detection, comprises:
(A) this drawing chip reads a two dimensional image.
(C) this drawing chip, according to a kind of the second procedure code of writing to open graphic package storehouse shading language, carries out Corner Detection in the mode of multinuclear parallel calculation to this two dimensional image, to find out angle point wherein.
Preferably, this two dimensional image that step (A) reads is a color 2 D image, and the method also comprises the step (B) being executed between step (A) and step (C):
(B) this drawing chip, according to a kind of the first procedure code of writing to open graphic package storehouse shading language, carries out GTG processing in the mode of multinuclear parallel calculation to this color 2 D image, to produce a GTG two dimensional image; And
Step (C) is that this GTG two dimensional image is carried out to Corner Detection.
Preferably, this two dimensional image that step (A) reads is a GTG two dimensional image, and step (C) is that this GTG two dimensional image is carried out to Corner Detection.
Preferably, the Corner Detection of step (C) is to adopt FAST Corner Detection algorithm.
Preferably, this mobile device is a mobile phone or personal digital assistants.
Another object of the present invention is to provide a kind of mobile device, comprise a drawing chip, this drawing chip is carried out as utilized drawing chip to accelerate the method for Corner Detection on being set forth in mobile device.
Beneficial effect of the present invention is: the present invention utilizes the characteristic of drawing chip multinuclear parallel calculation, and multinuclear ground is carried out Corner Detection algorithm and detected a plurality of picture point to carry out Corner Detection simultaneously, can accelerate the speed that mobile device carries out image recognition.
Accompanying drawing explanation
Fig. 1 is a calcspar, shows that carry out the present invention utilizes drawing chip to accelerate the mobile device of the method for Corner Detection on mobile device;
Fig. 2 is a process flow diagram, shows that the present invention utilizes drawing chip to accelerate a preferred embodiment of the method for Corner Detection on mobile device;
Fig. 3 is an architecture signal comparison diagram, shows the architecture comparison of central processing unit and drawing chip; And
Fig. 4 is a schematic diagram, shows the FAST Corner Detection algorithm that this preferred embodiment is carried out.
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in detail:
Consult Fig. 4, the present invention shown in figure utilizes drawing chip to accelerate the preferred embodiment of the method for Corner Detection on mobile device, it is the characteristic of utilizing drawing chip multinuclear parallel calculation, accelerate the Corner Detection of image recognition, and in order to calculate fast, the present embodiment is selected feature (Features from Accelerated Segment Test, the FAST) algorithm of the block test of acceleration.
Consult Fig. 1, carry out the present invention and on mobile device, utilize drawing chip to accelerate the preferred embodiment of mobile device of the method for Corner Detection, comprise the drawing chip 11 and the central processing unit 12 that are connected.
Consult Fig. 2, the method for the present embodiment comprises three steps, and wherein main two step S2 and S3, be respectively that GTG is processed and FAST computing, and these two steps are all to use drawing chip 11 to carry out parallel calculation
Step S1-this drawing chip 11 reads a color 2 D image.Specifically, be that this central processing unit 12 first reads after this color 2 D image in the present embodiment, set this drawing chip 11 and prepare, then by this color 2 D image, give this drawing chip 11 and process.
Step S2-this drawing chip 11 according to a kind of to open graphic package storehouse shading language (OpenGL (Open Graphics Library) Shading Language, GLSL) the first procedure code of writing, mode with multinuclear parallel calculation is carried out GTG processing to this color 2 D image, to produce a GTG two dimensional image.
Step S3-this drawing chip 11, according to a kind of the second procedure code of writing to open graphic package storehouse shading language, carries out Corner Detection in the mode of multinuclear parallel calculation to this GTG two dimensional image, to find out angle point wherein.The Corner Detection of the present embodiment is to adopt FAST Corner Detection algorithm.
The present embodiment is used drawing chip 11 to have arithmetic speed faster than use central processing unit 12.This is that the hardware architecture due to drawing chip similarly is the central processing unit of a multinuclear, and its core can be up to hundreds of, therefore, if drawing chip has 100 cores, just its arithmetic speed can be than fast upper 100 times of the central processing unit of monokaryon so.Central processing unit is as shown in Figure 3 four cores, and drawing chip has 14 multiprocessors, and each multiprocessor has the thirty-two nucleus heart, therefore has 448 cores, so arithmetic speed is 101 twelvefolds of central processing unit.
The FAST Corner Detection algorithm of carrying out at step S3 is proposed by Edward. Luo Siteng (Edward Rosten) doctor, its feature is as its name " fast ", be mainly used in the application that needs high speed processing, as the instant angle point that extracts from film, experimental data from the present embodiment, in more and more higher now high-res film, immediately obtain angle point, FAST Corner Detection algorithm can be said and have overwhelming superiority.
FAST Corner Detection algorithm, as shown in Fig. 4 and formula (1), be take tested point P as example, I
pfor the brightness value of tested point P, t is threshold value (threshold value), and X is the 1st to 16 comparison point, and { 1...16}, each is around central point I for X ∈
p16 comparison point brightness of point are designated as I
p->Xso, shown in following formula, have three kinds of states, the first I
p-> Xbe less than or equal to I
p-t, represents I
p-> Xcompared with I
psecretly, the second I
p-> Xbetween I
p-t and I
pbetween+t, represent I
p->Xwith I
psimilar, the third I
p-> Xbe more than or equal to I
p+ t, represents I
p-> Xcompared with I
pbright.FAST Corner Detection algorithm is got first and is judged with the third state whether tested point P is angle point, if there are continuous 9 to 12 I
p-> Xpoint compares I
pdark or bright, judge that tested point P is as angle point.
It should be noted that because Corner Detection is to calculate according to brightness value, therefore, as long as can access the means of brightness value, just can carry out the Corner Detection of the present embodiment.
Compared to acceleration robust feature (the Speeded Up Robost Features that has outstanding Corner Detection ability, SURF) algorithm, can process the image that yardstick is constant, FAST Corner Detection algorithm is applied to needing, in the instant application of processing, being also more suitable for the realization on mobile device 1 with the speed advantage of its processing.
And above-mentioned open graphic package storehouse shading language is the language that drawing Application Program Interface (API)-opening graphic package storehouse mobile device 1 version (Open GL ES 2.0) of using on mobile device 1 provides, it is one and take C language as basic high-order shading language, it provides developer directly to control drawing pipeline is more, and modern 3D draws to draw and approaches the appearance that the surprising effect of outdoor scene all gives the credit to this specification.
And in step S2, because can being all converted to the color value of each passage of three primary colors (RGB) 0 to 255 in 0 to 1.0 scope, processes drawing chip 11, and cause error slightly, make some color value in drawing chip 11, have 1 error, as color value 127 is transformed into the interior meeting of drawing chip 11, be 128, but this error is not large to affecting experimental result, does not affect the data of actual measurement.Following illustrative experiment result.
The experiment picture that the present embodiment is used is Lai Natu (Lena), and its resolution is respectively 512x512,1024x1024,1536x1536 and 2048x2048.The platform of testing is the mobile phone iPhone 3GS of Apple and iPhone 4, and its hardware specification is as shown in table 1.
And under memory limitations for fear of mobile device 1, Open GL ES standard picture resolution is 2048x2048 to the maximum, that is the present embodiment experimental resolution is only taken to the reason of 2048x2048.
Table 1, iPhone 3GS and iPhone 4 hardware comparison sheets
Experimentally divide three parts, first relatively carries out the execution speed of FAST algorithm (FAST-GPU) on iPhone 3GS and iPhone 4 platforms with drawing chip 11.Second portion is relatively used drawing chip 11 to carry out FAST algorithm on iPhone 4, (FAST-NonGPU, that is use central processing unit 12) carries out FAST algorithm and the execution speed of carrying out SURF algorithm not use drawing chip 11.Third part is tested the execution speed of three kinds of Corner Detection with five image experiments with picture.
<comparative example 1>experiment on iPhone 3GS and iPhone 4
First relatively on iPhone 3GS and iPhone 4, use drawing chip 11 to carry out the speed of FAST algorithm, resolution is respectively the Lai Natu of 512x512,1024x1024,1536x1536 and 2048x2048, the unit of execution time is microsecond (μ s), and experimental result is as shown in table 2.
Table 2, iPhone 3GS and iPhone 4 execution results
By what test, found that FAST-GPU part velocity contrast on two platforms is seldom consistent, by aforesaid table 1 hardware comparison sheet, due to what use, be same model drawing chip 11, can find out that its processing power depends primarily on drawing chip 11 rather than central processing unit 12.
<comparative example 2>three Corner Detection of comparison on iPhone 4
The Lai Natu with experiment one is still used in the experiment of the second part, compares specifically the execution speed of FAST-GPU, FAST-NonGPU and tri-algorithms of SURF.
Its execution result is as shown in table 3, take 512x512 as example, approximately than FAST fast 3.58 times of FAST GPU, than the fast 59.6 times of left and right of SURF, and 2048x2048, than FAST fast 24 times of FAST GPU, more reach 468 times of left and right than SURF gap, by data, also can find out that larger its execution speed gap of resolution is also larger
Table 3, iPhone 4 execution time list
| Picture size | FAS T-GPU | FAS T-NonGPU | SURF |
| 512x512 | 64 | 229 | 3816 |
| 1024x1024 | 81 | 848 | 15128 |
| 1536x1536 | 114 | 1877 | 36301 |
| 2048x2048 | 146 | 3509 | 68271 |
<comparative example 3>different picture experiments
It is conventional that experiment three part is got five images experiments, and the figure that picture material is different carrys out the execution speed of three kinds of algorithms of comparison.Experimental data partly compares the mean value of five figure, and its result is as shown in table 4.
Table 4, the average execution time list of comparative group five pictures
| Picture size | FAST-GPU | FAST-NonGPU | SURF |
| 512x512 | 47~64 | 205 | 3240 |
| 1024x1024 | 63~81 | 787 | 13964 |
| 1536x1536 | 97~114 | 1756 | 31020 |
| 2048x2048 | 130~147 | 3121 | 63120 |
The result of experiment is compared with aforementioned Lai Natu, the picture using there is no too big-difference to the execution speed of drawing chip 11 computings, FAST-NonGPU part is only slow a little, and SURF gap is larger, but the difference that affect execution speed is still that the resolution of picture is big or small.
In sum, the present invention utilizes the characteristic of drawing chip multinuclear parallel calculation, multinuclear ground is carried out FAST Corner Detection algorithm and is detected a plurality of picture point to carry out Corner Detection simultaneously, can accelerate the speed that mobile device carries out image recognition, and be really presented in experimental result, therefore really can reach the present invention's object.
Should be understood that those skilled in the art can make various changes or modifications the present invention after having read above-mentioned instruction content of the present invention, these equivalent form of values fall within the application's appended claims limited range equally.
Claims (6)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210219887.4A CN103514594A (en) | 2012-06-28 | 2012-06-28 | Method for detecting acceleration corner and mobile device for executing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210219887.4A CN103514594A (en) | 2012-06-28 | 2012-06-28 | Method for detecting acceleration corner and mobile device for executing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103514594A true CN103514594A (en) | 2014-01-15 |
Family
ID=49897272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210219887.4A Pending CN103514594A (en) | 2012-06-28 | 2012-06-28 | Method for detecting acceleration corner and mobile device for executing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103514594A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11062165B2 (en) | 2013-05-23 | 2021-07-13 | Movidius Limited | Corner detection |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030098875A1 (en) * | 2001-11-29 | 2003-05-29 | Yoshiyuki Kurokawa | Display device and display system using the same |
| CN101305620A (en) * | 2005-11-09 | 2008-11-12 | 辉达公司 | Graphics processing unit use and apparatus |
| US7522167B1 (en) * | 2004-12-16 | 2009-04-21 | Nvidia Corporation | Coherence of displayed images for split-frame rendering in multi-processor graphics system |
| US20110085139A1 (en) * | 2009-10-08 | 2011-04-14 | Tobii Technology Ab | Eye-tracking using a gpu |
-
2012
- 2012-06-28 CN CN201210219887.4A patent/CN103514594A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030098875A1 (en) * | 2001-11-29 | 2003-05-29 | Yoshiyuki Kurokawa | Display device and display system using the same |
| US7522167B1 (en) * | 2004-12-16 | 2009-04-21 | Nvidia Corporation | Coherence of displayed images for split-frame rendering in multi-processor graphics system |
| CN101305620A (en) * | 2005-11-09 | 2008-11-12 | 辉达公司 | Graphics processing unit use and apparatus |
| US20110085139A1 (en) * | 2009-10-08 | 2011-04-14 | Tobii Technology Ab | Eye-tracking using a gpu |
Non-Patent Citations (2)
| Title |
|---|
| 朱遵尚 等: "基于GPU的实时亚像素Harris角点检测", 《计算机工程》, vol. 36, no. 12, 30 June 2010 (2010-06-30) * |
| 燕鹏 等: "基于FAST改进的快速角点探测方法", 《红外与激光工程》, vol. 38, no. 6, 31 December 2009 (2009-12-31) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11062165B2 (en) | 2013-05-23 | 2021-07-13 | Movidius Limited | Corner detection |
| US11605212B2 (en) | 2013-05-23 | 2023-03-14 | Movidius Limited | Corner detection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11074717B2 (en) | Detecting and estimating the pose of an object using a neural network model | |
| El-Gayar et al. | A comparative study of image low level feature extraction algorithms | |
| US11734890B2 (en) | Three-dimensional model recovery from two-dimensional images | |
| EP3723043A1 (en) | Segmentation using an unsupervised neural network training technique | |
| Wang et al. | Workload analysis and efficient OpenCL-based implementation of SIFT algorithm on a smartphone | |
| CN110751620B (en) | Method for estimating volume and weight, electronic device, and computer-readable storage medium | |
| CN110136082A (en) | Occlusion culling method, apparatus and computer equipment | |
| CN107016716B (en) | Graphics processing apparatus and method of determining level of detail | |
| CN114241151B (en) | Three-dimensional model simplification method, three-dimensional model simplification device, computer equipment and computer storage medium | |
| CN104732490A (en) | CUDA-based quick bilateral filtering method | |
| CN114445318A (en) | A defect detection method, device, electronic device and storage medium | |
| WO2017070923A1 (en) | Human face recognition method and apparatus | |
| CN108074237A (en) | Approach for detecting image sharpness, device, storage medium and electronic equipment | |
| CN107392085A (en) | The method for visualizing convolutional neural networks | |
| CN111563458A (en) | Target detection and positioning method based on YOLOv3 and OpenCV | |
| US20220036106A1 (en) | Method and apparatus for data calculation in neural network model, and image processing method and apparatus | |
| CN116824055B (en) | Picture-based rapid three-dimensional interactive modeling method | |
| CN117407223A (en) | User interface defect detection method and device | |
| US20250336084A1 (en) | Modeling planar surfaces using direct plane fitting | |
| CN110660048A (en) | Leather surface defect detection algorithm based on shape characteristics | |
| CN103514594A (en) | Method for detecting acceleration corner and mobile device for executing method | |
| CN118365951A (en) | Ship identification model establishment and ship identification method, device and electronic equipment | |
| CN117689621A (en) | Measurement method, measurement device, electronic equipment and computer readable storage medium | |
| Luo et al. | Accelerating Harris algorithm with GPU for corner detection | |
| CN117830247A (en) | Fabric defect detection method, device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140115 |