DOI QR코드

DOI QR Code

SVM 기반 사물 인식을 위한 고성능 벡터 내적 연산 회로의 MPW 칩 구현 및 검증

MPW Chip Implementation and Verification of High-performance Vector Inner Product Calculation Circuit for SVM-based Object Recognition

  • 신재호 (한국외국어대학교 전자공학과) ;
  • 김수진 (한국외국어대학교 전자공학과) ;
  • 조경순 (한국외국어대학교 전자공학과)
  • Shin, Jaeho (Department of Electronics Engineering, Hankuk University of Foreign Studies) ;
  • Kim, Soojin (Department of Electronics Engineering, Hankuk University of Foreign Studies) ;
  • Cho, Kyeongsoon (Department of Electronics Engineering, Hankuk University of Foreign Studies)
  • 투고 : 2013.08.05
  • 발행 : 2013.11.25

초록

본 논문은 SVM 알고리즘 기반의 실시간 사물 인식을 위한 고성능 벡터 내적 연산 회로를 제안한다. SVM 알고리즘은 다른 사물 인식 알고리즘에 비해 인식률이 높지만 연산량이 많다. 벡터 내적 연산은 SVM 알고리즘 연산의 주요 연산으로 사용되므로 실시간 사물 인식을 위해서는 고성능 벡터 내적 연산 회로의 구현이 필수적이다. 제안하는 회로는 연산 속도를 높이기 위해 6단 파이프라인 구조를 적용하였으며 SVM 기반 실시간 사물 인식을 가능하게 한다. 제안하는 회로는 Verilog HDL을 사용하여 RTL로 구현하였으며 실리콘 검증을 위해 TSMC 180nm 표준 셀 라이브러리를 이용하여 MPW 칩으로 제작하였다. 테스트 보드와 검증 애플리케이션 소프트웨어를 개발하고 이를 사용하여 MPW 칩의 동작을 확인하였다.

This paper proposes a high-performance vector inner product calculation circuit for real-time object recognition based on SVM algorithm. SVM algorithm shows a higher detection rate than other object recognition algorithms. However, it requires a huge amount of computational efforts. Since vector inner product calculation is one of the major operations of SVM algorithm, it is important to implement a high-performance vector inner product calculation circuit for real-time object recognition capability. The proposed circuit adopts the pipeline architecture with six stages to increase the operating speed and makes it possible to recognize objects in real time based on SVM. The proposed circuit was described in Verilog HDL at RTL. For silicon verification, an MPW chip was fabricated using TSMC 180nm standard cell library. The operation of the implemented MPW chip was verified on the test board with test application software developed for the chip verification.

키워드

참고문헌

  1. V. N. Vapnik, Statistical Learning Theory, John Wiley & Sons, 1998.
  2. C. Kyrkou and T. Theocharides, "A Parallel Hardware Architecture for Real-Time Object Detection with Support Vector Machines," IEEE Trans. Computers., vol. 66, pp. 831-842, June 2012.
  3. M. Hiromoto and R. Miyamoto, "Hardware Architecture for High-accuracy Real-time Pedestrian Detection with CoHOG Features," IEEE 12th International Conf. Computer Vision., pp. 894-899, Kyoto, Japan, October 2009.
  4. A. Ghio, "A Support Vector Machine based Pedestrian Recognition System on Resourcelimited Hardware Architectures," Microelectronics and Electronics Conf., pp. 161-163, July 2007.
  5. M. Papadonikolakis, "Novel Cascade FPGA Accelerator or Support Vector Machines Classification," IEEE Trans. on Neural Networks and Learning System., vol. 23, pp. 1040-1052, July 2012. https://doi.org/10.1109/TNNLS.2012.2196446