Presentation 2012-05-29
FPGA implementation of a video-based real-time pupil detection method
Yuma HATANAKA, Keisuke DOHI, Kazuhiro NEGI, Yuichiro SHIBATA, Kiyoshi OGURI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, we show an FPGA implementation of the Starbust algorithm without external memory, which detects an ellipse from a given input image for eye tracking. The first part of this algorithm, in which feature points for a potential contours are detected, is implemented in a deep-pipelined stream-oriented architecture. For the second part, where many ellipses are repeatedly fit and validated with the RANSAC algorithm, we compare three different architectures based on Cramer's rule, Gauss-Jordan elimination and LU decomposition. Comparison criteria include resource usage, throughput and energy consumption. As a result, our system achieved a real-time throughput of 62.5 frames per second with power consumption of approximately 3 watts. In addition, the three implementations showed different characteristics in terms of the performance, resource utilization and energy consumption, suggesting that the optimal algorithm needs to be chosen depending on the amount of resources on FPGAs and required performance.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) FPGA / Starburst / RANSAC / ellipse estimation
Paper # RECONF2012-3
Date of Issue

Conference Information
Committee AI
Conference Date 2012/5/22(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Artificial Intelligence and Knowledge-Based Processing (AI)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) FPGA implementation of a video-based real-time pupil detection method
Sub Title (in English)
Keyword(1) FPGA
Keyword(2) Starburst
Keyword(3) RANSAC
Keyword(4) ellipse estimation
1st Author's Name Yuma HATANAKA
1st Author's Affiliation Department of Computer and Information Sciences, Graduate School of Engineering, Nagasaki University()
2nd Author's Name Keisuke DOHI
2nd Author's Affiliation Department of Computer and Information Sciences, Graduate School of Engineering, Nagasaki University
3rd Author's Name Kazuhiro NEGI
3rd Author's Affiliation Department of Computer and Information Sciences, Graduate School of Engineering, Nagasaki University
4th Author's Name Yuichiro SHIBATA
4th Author's Affiliation Department of Computer and Information Sciences, Graduate School of Engineering, Nagasaki University
5th Author's Name Kiyoshi OGURI
5th Author's Affiliation Department of Computer and Information Sciences, Graduate School of Engineering, Nagasaki University
Date 2012-05-29
Paper # RECONF2012-3
Volume (vol) vol.112
Number (no) 70
Page pp.pp.-
#Pages 6
Date of Issue