3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU). e

Similar documents
Ⅱ. Embedded GPU 모바일 프로세서의 발전방향은 저전력 고성능 컴퓨팅이다. 이 러한 목표를 달성하기 위해서 모바일 프로세서 기술은 멀티코 어 형태로 발전해 가고 있다. 예를 들어 NVIDIA의 최신 응용프 로세서인 Tegra3의 경우 쿼드코어 ARM Corte

09권오설_ok.hwp

2 : 3 (Myeongah Cho et al.: Three-Dimensional Rotation Angle Preprocessing and Weighted Blending for Fast Panoramic Image Method) (Special Paper) 23 2

(JBE Vol. 24, No. 2, March 2019) (Special Paper) 24 2, (JBE Vol. 24, No. 2, March 2019) ISSN

(JBE Vol. 23, No. 6, November 2018) (Regular Paper) 23 6, (JBE Vol. 23, No. 6, November 2018) ISSN 2

(JBE Vol. 21, No. 1, January 2016) (Regular Paper) 21 1, (JBE Vol. 21, No. 1, January 2016) ISSN 228

... K-vision Fig.. K-vision camera tracking screen Drummond [3] 3. 3 (lines), (edge) 3. (target). (homography perspective transform) [4]. (drifting).

19_9_767.hwp

6.24-9년 6월

26 이경승(394~400).hwp

(JBE Vol. 20, No. 5, September 2015) (Special Paper) 20 5, (JBE Vol. 20, No. 5, September 2015) ISS

example code are examined in this stage The low pressure pressurizer reactor trip module of the Plant Protection System was programmed as subject for

°í¼®ÁÖ Ãâ·Â

02( ) SAV12-19.hwp

08김현휘_ok.hwp

06( ) CPLV 페이지로수정.hwp

05( ) CPLV12-04.hwp

(JBE Vol. 20, No. 6, November 2015) (Regular Paper) 20 6, (JBE Vol. 20, No. 6, November 2015) ISSN

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE. vol. 29, no. 10, Oct ,,. 0.5 %.., cm mm FR4 (ε r =4.4)

(JBE Vol. 23, No. 5, September 2018) (Regular Paper) 23 5, (JBE Vol. 23, No. 5, September 2018) ISSN

1. 서 론

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Feb.; 29(2), IS

<35335FBCDBC7D1C1A42DB8E2B8AEBDBAC5CDC0C720C0FCB1E2C0FB20C6AFBCBA20BAD0BCAE2E687770>

인문사회과학기술융합학회

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Jul.; 27(7),

REP - CP - 016, N OVEMBER 사진 요약 25 가지 색상 Surf 를 이용한 사진 요약과 사진 배치 알고리즘 Photo Summarization - Representative Photo Selection based on 25 Color Hi

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE. vol. 29, no. 6, Jun Rate). STAP(Space-Time Adaptive Processing)., -

03홍성욱.hwp

(JBE Vol. 24, No. 1, January 2019) (Regular Paper) 24 1, (JBE Vol. 24, No. 1, January 2019) ISSN 2287

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Nov.; 25(11),

(JBE Vol. 7, No. 4, July 0)., [].,,. [4,5,6] [7,8,9]., (bilateral filter, BF) [4,5]. BF., BF,. (joint bilateral filter, JBF) [7,8]. JBF,., BF., JBF,.

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Jun.; 27(6),

DBPIA-NURIMEDIA

À±½Â¿í Ãâ·Â

2 : (JEM) QTBT (Yong-Uk Yoon et al.: A Fast Decision Method of Quadtree plus Binary Tree (QTBT) Depth in JEM) (Special Paper) 22 5, (JBE Vol. 2

04 최진규.hwp

(JBE Vol. 22, No. 2, March 2017) (Regular Paper) 22 2, (JBE Vol. 22, No. 2, March 2017) ISSN

디지털포렌식학회 논문양식

06_ÀÌÀçÈÆ¿Ü0926

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Nov.; 26(11),

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Jan.; 26(1),

High Resolution Disparity Map Generation Using TOF Depth Camera In this paper, we propose a high-resolution disparity map generation method using a lo

RRH Class-J 5G [2].,. LTE 3G [3]. RRH, W-CDMA(Wideband Code Division Multiple Access), 3G, LTE. RRH RF, RF. 1 RRH, CPRI(Common Public Radio Interface)

45-51 ¹Ú¼ø¸¸

DBPIA-NURIMEDIA

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Dec.; 27(12),

8-VSB (Vestigial Sideband Modulation)., (Carrier Phase Offset, CPO) (Timing Frequency Offset),. VSB, 8-PAM(pulse amplitude modulation,, ) DC 1.25V, [2

<353420B1C7B9CCB6F52DC1F5B0ADC7F6BDC7C0BB20C0CCBFEBC7D120BEC6B5BFB1B3C0B0C7C1B7CEB1D7B7A52E687770>

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Sep.; 30(9),

10 이지훈KICS hwp

1 : 360 VR (Da-yoon Nam et al.: Color and Illumination Compensation Algorithm for 360 VR Panorama Image) (Special Paper) 24 1, (JBE Vol. 24, No

<30312DC1A4BAB8C5EBBDC5C7E0C1A4B9D7C1A4C3A52DC1A4BFB5C3B62E687770>

학습영역의 Taxonomy에 기초한 CD-ROM Title의 효과분석

(JBE Vol. 21, No. 1, January 2016) (Special Paper) 21 1, (JBE Vol. 21, No. 1, January 2016) ISSN 228

untitled

2 : (Juhyeok Mun et al.: Visual Object Tracking by Using Multiple Random Walkers) (Special Paper) 21 6, (JBE Vol. 21, No. 6, November 2016) ht

½Éº´È¿ Ãâ·Â

???? 1

<31325FB1E8B0E6BCBA2E687770>

04 김영규.hwp

08이규형_ok.hwp

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Mar.; 25(3),

±è¼ºÃ¶ Ãâ·Â-1

Software Requirrment Analysis를 위한 정보 검색 기술의 응용

V28.

1 : (Sunmin Lee et al.: Design and Implementation of Indoor Location Recognition System based on Fingerprint and Random Forest)., [1][2]. GPS(Global P

<30362E20C6EDC1FD2DB0EDBFB5B4EBB4D420BCF6C1A42E687770>

DBPIA-NURIMEDIA

[ReadyToCameral]RUF¹öÆÛ(CSTA02-29).hwp

DBPIA-NURIMEDIA

<313120C0AFC0FCC0DA5FBECBB0EDB8AEC1F2C0BB5FC0CCBFEBC7D15FB1E8C0BAC5C25FBCF6C1A42E687770>

2 : (Jaeyoung Kim et al.: A Statistical Approach for Improving the Embedding Capacity of Block Matching based Image Steganography) (Regular Paper) 22

DBPIA-NURIMEDIA

DBPIA-NURIMEDIA

Æ÷Àå½Ã¼³94š

정보기술응용학회 발표

10(3)-09.fm

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Mar.; 28(3),

, N-. N- DLNA(Digital Living Network Alliance).,. DLNA DLNA. DLNA,, UPnP, IPv4, HTTP DLNA. DLNA, DLNA [1]. DLNA DLNA DLNA., [2]. DLNA UPnP. DLNA DLNA.

09박종진_ok.hwp

Windows Embedded Compact 2013 [그림 1]은 Windows CE 로 알려진 Microsoft의 Windows Embedded Compact OS의 history를 보여주고 있다. [표 1] 은 각 Windows CE 버전들의 주요 특징들을 담고

SchoolNet튜토리얼.PDF

07.045~051(D04_신상욱).fm

Analysis of objective and error source of ski technical championship Jin Su Seok 1, Seoung ki Kang 1 *, Jae Hyung Lee 1, & Won Il Son 2 1 yong in Univ

untitled

±èÇö¿í Ãâ·Â

3 : (Won Jang et al.: Musical Instrument Conversion based Music Ensemble Application Development for Smartphone) (Special Paper) 22 2, (JBE Vol

02손예진_ok.hwp

14.이동천교수님수정

<343620B3EBB1A4C7F62DBDBAB8B6C6AEC6F9BFEB20C2F7BCB1C0CCC5BBB0E6BAB820BED6C7C3B8AEC4C9C0CCBCC720B0B3B9DF2E687770>

1. KT 올레스퀘어 미디어파사드 콘텐츠 개발.hwp

Journal of Educational Innovation Research 2018, Vol. 28, No. 1, pp DOI: A study on Characte

01이국세_ok.hwp

특집-5

박선영무선충전-내지

(JBE Vol. 23, No. 1, January 2018). (VR),. IT (Facebook) (Oculus) VR Gear IT [1].,.,,,,..,,.. ( ) 3,,..,,. [2].,,,.,,. HMD,. HMD,,. TV.....,,,,, 3 3,,

3. 클라우드 컴퓨팅 상호 운용성 기반의 서비스 평가 방법론 개발.hwp

<31372DB9DABAB4C8A32E687770>

14.531~539(08-037).fm

08~15_º¸°ÇÀÇ·áºÐ¾ßODAÆò°¡

Transcription:

(JBE Vol. 19, No. 3, May 2014) (Special Paper) 19 3, 2014 5 (JBE Vol. 19, No. 3, May 2014) http://dx.doi.org/10.5909/jbe.2014.19.3.316 ISSN 2287-9137 (Online) ISSN 1226-7953 (Print) OpenCL Embedded GPU a), a), a), a) Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU Seung Heon Kang a), Seung-Jae Lee a), Man Hee Lee a), and In Kyu Park a) embedded GPU (SIFT, SURF), GPU. GPGPU(general purpose computation on GPU) embedded GPU OpenCL. GPU OpenGL Shading Language(GLSL) OpenCL., OpenCL GLSL embedded CPU 3~4., GPU. Abstract In this paper, we parallelize the popular feature detection algorithms, i.e. SIFT and SURF, and its application to fast panoramic image generation on the latest embedded GPU. Parallelized algorithms are implemented using recently developed OpenCL as the embedded GPGPU software platform. We compare the implementation efficiency and speed performance of conventional OpenGL Shading Language and OpenCL. Experimental result shows that implementation on OpenCL has comparable performance with GLSL. Compared with the performance on the embedded CPU in the same application processor, the embedded GPU runs 3~4 times faster. As an example of using feature extraction, panorama image synthesis is performed on embedded GPU by applying image matching using detected features. Keyword : Embedded GPGPU, OpenCL, OpenGL, SIFT, SURF, panorama image a) (Department of Information and Communication) Corresponding Author : (In Kyu Park) E-mail: pik@inha.ac.kr Tel: +82-32-860-9190.[10041664, Shader GPU ].. Manuscript received March 17, 2014 Revised May 2, 2014 Accepted May 8, 2014

3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU). embedded GPU(graphic processing unit) 3,. GPGPU(general purpose computation on GPU),. Embedded GPU GPU. Qualcomm, Imagination Technologies, ARM, NVIDIA embedded GPU ( 4~16),., GPGPU SIMD (single instruction multiple data), GPGPU [1][2][3]., (feature extraction) (feature descriptor) SIFT(scale-invariant feature transform) [4] SURF(speeded up robust features) [5] OpenCL embedded GPU. embedded CPU. OpenGL Shading Language (GLSL) [6] GPU OpenCL [7]., embedded GPU. [8]. OpenCL 3 GPU, GPU NVIDIA CUDA. OpenCL, embedded GPU OpenCL. embedded GPU OpenCL. Embedded GPU,. SIFT, SURF, embedded GPU OpenCL embedded GPU.. 2 GPGPU. 3 SIFT SURF. 4 SIFT. 5. 6.. GPGPU OpenCL CPU, GPU, DSP [7]. OpenCL embedded GPU GPGPU. GLSL GPGPU OpenGL GPGPU.. OpenCL GPGPU

(JBE Vol. 19, No. 3, May 2014) image2d_t CPU,. OpenCL. CPU. OpenCL OpenGL... Embedded GPU SIFT SURF, OpenCL OpenGL embedded GPU, GPGPU. 1. SIFT SIFT, [4]. SIFT 1. 1) RGB 2) 3) Difference of Gaussian (DoG) 4) (keypoint localization) 5) (orientation) 6) (descriptor) OpenCL. RGB. (per-pixel),. (octave). (local memory). 4. Difference of Gaussian. CPU work-item., non-maximum suppression (NMS)..,.. 1. SIFT Fig 1. Structure of SIFT algorithm

3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU) 1. SIFT OpenCL Table 1. OpenCL implementation structure of SIFT feature extraction algorithm Work-Item (Work-Group ) RGB2Gray 1 Per-pixel 16 16 Pyramid 1 Per-pixel using local memory 16 8 DoG 1 Per-pixel 8 8 Keypoint Localization 1 Per-pixel 16 16 Orientation Computation 1 Per-pixel 16 16 Descriptor Generation 1 Per-feature using local memory 16 1 2. SURF SURF SIFT [5]. SURF 2. 1) RGB 2) (integral image) 3) (box) Hessian determinant 4) Non-maximum suppression (NMS) 5) Harr response 6) OpenCL. RGB.,., workgroup work-item (reduction)., work-item 64 1,. work-item. GLSL 255. OpenCL image2d_t.. Hessian (determinant), 4 4 Hessian 10. 2. SURF Fig 2. Structure of SURF algorithm

(JBE Vol. 19, No. 3, May 2014), work-item. Hessian. NMS NMS. CPU GPU. work-item. GPU., OpenCL GLSL (21 2 ).. GPU GPU embedded GPU GPGPU. GLSL OpenCL, (51 23 ). OpenCL GLSL GPGPU., OpenCL. 3. GPU Embedded GPU GPGPU. GPU GPGPU, GPU embedded GPU GPGPU [1][2]. (1), (2) Loop unrolling for, (3), (4) Work-item work-group work-item. work-item work-group.,,.. SIFT (homography) (warping) [9]. 2. SURF OpenCL Table 2. OpenCL implementation structure of SURF feature extraction algorithm Work-Item (Work-Group ) RGB2Gray 1 Per-pixel 16 16 Integral Image 2 Reduction using local memory 64 1, 1 64 Hessian Determinant 10 Per-pixel 8 8, 4 16, 16 2, 16 1 NMS 8 Per-pixel 16 8, 32 4, 32 2, 16 2 Orientation Computation 1 Per-feature using local memory 16 1 Descriptor Generation 1 Per-feature using local memory 16 1

3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU) 3. Fig 3. Block diagram of panorama image generation 3. 1) SIFT 2) 3) RANSAC 4) (inlier)/ (outlier) 5) 3), 4) 6) Embedded GPU OpenCL. III SIFT. SIFT. work-item, 128 SIFT.. RANSAC (random sample consensus) [10],. 256, work-item.,.. embedded GPU. 3. OpenCL Table 3. OpenCL implementation structure of panorama image generation Work-Item (Work-Group ) SIFT 1 Per-feature using local memory 16 1 2 Per-feature 16 16 (RANSAC) Homography 1 Per-sample 16 1 (RANSAC) Inlier/Outlier 1 Per-sample 16 1

322 방송공학회논문지 제19권 제3호, 2014년 5월 (JBE Vol. 19, No. 3, May 2014) 전자의 Galaxy S4 LTE-A이며, 이는 Qualcomm의 Snapdragon 800 응용 프로세서를 장착하였다. Snapdragon 800 응용프로세서는 Krait공정의 쿼드코어 2.3GHz CPU와 Adreno 330 쿼드코어 450MHz GPU를 탑재하고 있다. 스 마트폰의 운영체제는 Android 4.2이고, 실험 영상의 해상 도는 1280 720이다. 스마트폰에서의 인터페이스 프로그램 을 이용하여 파일로 저장된 영상에 대한 처리와 카메라에 서의 라이브 입력 비디오를 캡처한 영상에 대한 처리를 선 택적으로 수행할 수 있다. 1. SIFT (a) (b) SIFT 실험영상 알고리즘 결과 알고리즘 그림 4(a)의 실험 영상에 대하여 OpenCL을 이용하여 구 현한 SIFT 알고리즘의 세부 단계 별 수행속도는 표 4에 제 시한 바와 같다. 본 실험 영상에 대하여 추출된 특징점의 개수는 354개이다. 피라미드 영상 생성은 지역메모리를 사 용함으로 속도 개선을 기대할 수 있으나, 각 피라미드별 화 소당 차이를 구하는 양이 많기 때문에 수행 시간이 길어질 수밖에 없다. 또한 특징점 방향계산과 특징점 특징기술자 생성에서는 특징점의 개수만큼의 병렬처리를 수행하기 때 문에 영상의 복잡도와 특징 추출 문턱치 값에 따라 이 두 단계의 수행시간은 변경될 수 있다. 다양한 실험 영상에 대 한 SIFT 특징점 추출 결과는 그림 5(a)에 추가적으로 제시 하였다. Embedded GPU 환경에서의 SIFT 알고리즘 구현 은 두 언어 모두 병렬 알고리즘의 구조가 유사하기 때문에 GLSL(301.4ms/frame)과 OpenCL(331.5ms/frame)에서 평 균적으로 대등한 성능을 보인다. 같은 알고리즘이 구현되 어있는 Android용 OpenCV에서의 수행 결과는 1227ms/ frame이다. 따라서 GPU에서의 SIFT 특징 추출 수행 속도 표 4. 그림 4의 실험 영상에 대한 OpenCL 기반 SIFT 알고리즘 단계별 수행 속도 Table 4. Step-by-step execution time of SIFT implementation on OpenCL SIFT 알고리즘 결과 그림 4. 수행시간 측정에 사용한 실험 영상과 SIFT, SURF 알고리즘 수행 결과 (c) SURF Fig 4. Test image used in execution time measurement and its SIFT, SURF results 알고리즘 세부 단계 수행시간 (ms) RGB2Gray 9.14 Pyramid 93.71 DoG 15.79 Keypoint Localization 91.54 Orientation Computation 50.32 Descriptor Generation 62.13 합계 322.63

3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU) CPU 3.7. 2. SURF 4(a) OpenCL SURF 5. 296. 5 NMS.,. 5 4 Hessian. 8,. SURF embedded GPU, OpenGL OpenCL 192.4ms/frame 272.1ms/frame., OpenCV CPU SURF 981.3ms/frame GPU CPU 3.6. SIFT SURF, Hessian 5. 4 OpenCL SURF Table 5. Step-by-step execution time of SURF implementation on OpenCL SURF (ms) RGB2Gray 6.25 Integral Image 57.36 Hessian Determinant 88.28 NMS 23.49 Orientation Computation 45.13 Descriptor Generation 30.12 250.63. Embedded GPU, SURF OpenCL SIFT. 3. SIFT. 5(a) SIFT. 5(b). 5(c) RANSAC. 5(d). 157, 127, SIFT 81.5ms. 6.. RANSAC 6. 5 Table 6. Execution time of panorama image generation step SIFT (ms) 270.42 ( ) 242.64 ( ) 46.98 (RANSAC) Homography 8.28 (RANSAC) Inlier/Outlier 7.29 10.32 585.93

324 방송공학회논문지 제19권 제3호, 2014년 5월 (JBE Vol. 19, No. 3, May 2014) 에서는 반복 횟수만큼 스레드가 생성됨으로 반복 횟수가 늘어나는 만큼 확률적으로 정확도가 높아질 수도 있으나 (a) (b) 그림 5. 파노라마 영상 합성 중간 결과 두 장의 입력 영상 초기 특징점 정합 결과 (c) RANSAC (d) 이에 따라 수행시간은 늘어나게 된다. 본 논문에서는 실험 적으로 RANSAC의 반복 회수를 256회로 결정하였다. 또 단계 후 최적의 특징점 네 쌍 합성된 파노라마 영상 Fig 5. Intermediate result of panorama image synthesis

강승헌 외 인 및 를 이용한 영상 특집 추출 및 파노라마 영상 생성의 병렬화 3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU) 325 (a) (b) 그림 6. Embedded GPU(Adreno 330)및 OpenCL을 이용한 특징점 추출 결과 (a) SIFT, (b) SURF Fig 6. Feature extraction results using OpenCL on the embedded GPU (Adreno 330). They are captured on the smartphone s screen. (a) SIFT. (b) SURF 한 정합 성능 향상을 위하여 영상 간 중첩되는 영역은 좌우 50% 일 경우로 제한하여 실험을 진행하였다. 다양한 영상 에 대한 파노라마 영상 생성 실험 결과를 그림 7에 제시하 였다.

326 방송공학회논문지 제19권 제3호, 2014년 5월 (JBE Vol. 19, No. 3, May 2014) (a) (b) (a) 세장의 입력 영상 (실내) 의 영상을 이용하여 합성된 파노라마 영상 (c) 세장의 입력 영상 (야외) 의 영상을 이용하여 합성된 파노라마 영상 그림 7. Embedded GPU(Adreno 330)및 OpenCL을 이용한 파노라마 영상 생성 결과 (d) (c) Fig 7. Result of panorama generation using OpenCL on the embedded GPU (Adreno 330)

3 : OpenCL Embedded GPU (Seung Heon Kang et al. : Parallelization of Feature Detection and Panorama Image Generation using OpenCL and Embedded GPU). embedded GPU OpenCL SIFT SURF,. embedded CPU GLSL, CPU 3.7 OpenCL GLSL. OpenCL OpenGL. embedded GPU. (References) [1] I. K. Park, N. Singhal, M. H. Lee, S. Cho, and C. Kim, Design and performance evaluation of image processing algorithms on GPUs, IEEE Trans. on Parallel and Distributed Systems, vol. 22, no. 1, pp. 91 104, January 2011. [2] N. Singhal, J. W. Yoo, H. Y. Choi, and I. K. Park, Implementation and optimization of image processing algorithms on embedded GPU, IEICE Trans. on Information and Systems, E95-D(5):1475 1484, May 2012 [3] J. Lee, S. H. Kang, M. H. Lee, S. Li, H. Kim, I. K. Park,, Real-time parallel image processing library using mobile GPU, Journal of KIISE: Computing Practices and Letters, vol. 20, no. 2, pp. 96-100, February 2014. [4] D. Lowe, Distinctive image features from scale-invariant keypoints, International Journal of Computer Vision, vol. 60, no. 2, pp. 91 110, November 2004. [5] H. Bay, A. Ess, T. Tuytelaars, and L. V. Gool, Speeded-up robust features, Computer Vision and Image Understanding, vol. 110, no. 3, pp. 346 359, June 2008. [6] The Khronos Group, OpenGL Shading Language, https://www.opengl. org/ documentation/glsl/. [7] The Khronos Group, Open Computing Language (OpenCL), http://www.khronos.org/opencl/. [8] S. H. Kang, S.-J. Lee, and I. K. Park, Parallelization and optimization of feature dtection algorithms on embedded GPU, Proc. International Workshop on Advanced Image Technology, pp. 164-167, January 2014. [9] R. I. Hartley and A. Zisserman, Multiple View Geometry in Computer Vision, Second Edition, Cambridge University Press, 2004. [10] M. A. Fischler and R. C. Bolles, Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography, Communications of the ACM, vol. 24, no. 6, pp. 381-395, June 1981. - 2012 2 : - 2012 3 ~ : - :, GPGPU - 2013 2 : - 2013 3 ~ : - :, GPGPU

(JBE Vol. 19, No. 3, May 2014) - 2006 2 : - 2007 3 ~ 2008 2 : (ETRI) - 2008 8 : - 2011 1 ~ 2011 8 : Harvard University - 2008 9 ~ : - :, 3, GPGPU - 1995 2 : - 1997 2 : - 2001 8 : - 2001 9 ~ 2004 3 : - 2007 1 ~ 2008 2 : Mitsubishi Electric Research Laboratories (MERL) - 2004 3 ~ : - : ( 3, 3, computational photography), GPGPU