| Title |
Efficient Real-Time System Design for Optical Flow Based on Lucas-Kanade Algorithm |
| Authors |
김민수(Min-Su Kim) ; 박수민(Su-Min Park) ; 조우성(Woo-Sung Cho) ; 박태근(Tae-Geun Park) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.9.1591 |
| Keywords |
Optical Flow; Realtime System; Lucas-Kanade algorithm; VLSI design |
| Abstract |
In this paper, a real-time optical flow system based on the Lucas-Kanade algorithm was designed. In order to improve hardware complexity and design efficiency, optimal bit allocation was performed through error rate analysis. Also real-time processing performance was achieved through parallel/pipeline structure and overlapping scheduling. By optimizing the memory cycle so that even and odd data can be read and written at the same time, data utilization was maximized. In the Lucas-Kanade algorithm, the output data of the pre-calculated intermediate kernel is frequently reused. So data were updated by applying optimal overlapping scheduling using minimal memory and the required amount of computation and memory space were reduced. The proposed architecture was designed and synthesized with a DB HiTek 110nm standard cell library with an error rate of 3.70% for Yosemite Sequence and a maximum operating frequency of 268MHz (264.6Mpixel/s), using 71.59K gates and 15.43KB of memory. |