DOI QR코드

DOI QR Code

A Threshold Controller for FAST Hardware Accelerator

FAST 하드웨어 가속기를 위한 임계값 제어기

  • Kim, Taek-Kyu (Research Reactor System Design Department, Korea Atomic Energy Research Institute) ;
  • Suh, Yong-Suk (Research Reactor System Design Department, Korea Atomic Energy Research Institute)
  • 김택규 (한국원자력연구원 연구로계통설계부) ;
  • 서용석 (한국원자력연구원 연구로계통설계부)
  • Received : 2014.08.19
  • Accepted : 2014.10.27
  • Published : 2014.11.25

Abstract

Various researches are performed to extract significant features from continuous images. The FAST algorithm has the simple structure for arithmetic operation and it is easy to extraction the features in real time. For this reason, the FPGA based hardware accelerator is implemented and widely applied for the FAST algorithm. The hardware accelerator needs the threshold to extract the features from images. The threshold is influenced not only the number of extracted features but also the total execution time. Therefore, the way of threshold control is important to stabilize the total execution time and to extract features as much as possible. In order to control the threshold, this paper proposes the PI controller. The function and performance for the proposed PI controller are verified by using test images and the PI control logic is designed based on Xilinx Vertex IV FPGA. The proposed scheme can be implemented by adding 47 Flip Flops, 146 LUTs, and 91 Slices to the FAST hardware accelerator. This proposed approach only occupies 2.1% of Flip Flop, 4.4% of LUTs, and 4.5% of Slices and can be regarded as a small portion of hardware cost.

카메라와 같이 연속적인 영상을 제공하는 환경에서 특징 점들을 추출하기 위해 다양한 알고리즘들이 연구되고 있다. 특히, FAST (Feature from Accelerated Segment Test) 알고리즘은 연산 구조가 간단하고 실시간 특징 점 추출이 용이하여 FPGA 기반 하드웨어 가속기로 구현되어 사용되고 있다. 사용된 FAST 하드웨어 가속기는 특징 점을 추출하기 위해 임계값을 필요로 한다. 임계값은 영상에서 추출되는 특징 점의 기준이 되는 값으로, 값의 크기에 따라 추출되는 특징 점의 개수가 정해질 뿐만 아니라 전체 수행시간에도 영향을 주기 때문에, 일정한 수행시간 동안에 많은 특징 점들을 추출하기 위해서는 적절한 임계값 제어 방법이 요구된다. 본 논문에서는 임계값 제어를 위해 PI 제어기를 제안한다. 제안한 PI 제어기는 시험 영상들을 통해 기능 및 성능을 검증하였고, Xilinx Vertex IV FPGA 기반의 로직으로 구현 비용을 계산하였다. 제안한 PI 제어기는 47개의 Flip Flops, 146개의 LUTs, 그리고 91개의 Slices을 사용해, FAST 하드웨어 가속기 2.1%의 Flip Flop, 4.4%의 LUTs, 그리고 4.6%의 Slice에 해당하는 적은 비용으로 구현되었다.

Keywords

References

  1. L. Teixeria, W. Celes, and M. Gattass, "Accelerated Corner-detector Algorithms," in BMVC08, 2008. [Online]. Available: http://www. comp.leeds.ac.uk/bmvc2008/proceedings/paper/45.pdf
  2. D. Bouris, A. Nikitakis, and J. Waters, "Fast and Efficient FPGA-Based Feature Detection Employing the SURF Algorithm," 18th IEEE Annual International Symposium on Digital Object Identifier, pp. 3-10, 2010.
  3. S. Smith and J. Brady, "SUSAN-A New Approach to Low-level Image Processing," International Journal of Computer Vision, vol. 23, pp. 45-48, 1997. https://doi.org/10.1023/A:1007963824710
  4. Edward Rosten et al, "Faster and better: A Machine Learning Approach to Corner Detection," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 32, no. 1, pp. 105-119, 2010. https://doi.org/10.1109/TPAMI.2008.275
  5. T.-K. Kim, "An Embedded FAST Hardware Accelerator for Image Feature Detection," Journal of IEEK, vol. 49, SP no. 2, pp. 28-34, 2012.
  6. T.-K. Kim, Y.-S. Suh, and Y.-K. Kim, "Implementation of Feature Point Extraction Accelerator Based on FAST," The IEEK Summer Conference, vol. 36, no. 1, pp. 609-612, 2013.
  7. Norman S. Nise, "Control Systems Engeering," Sixth Edition, John Wiley & Sons, 2011.
  8. E. Rosten and T. Drummond, "Fusing Points and Lines for High Performance Tracking," Proc. 10th IEEE International Conference Computer Vision, vol. 2, pp. 1508-1515, 2005.