| Title |
Design of Real-time Optical Flow Estimation Circuit for HD Images |
| DOI |
https://doi.org/10.5573/ieie.2024.61.9.23 |
| Keywords |
Optical flow; DIS-FAST algorithm; Motion analysis; HD image; Real-time processing |
| Abstract |
In recent times, autonomous driving technology has been rapidly advancing, necessitating accurate motion analysis for detecting and tracking pedestrians and objects in the process of images. Among various methods for motion analysis, the optical flow estimation, which allows for semantic interpretation, is now widely utilized. This paper proposes an architecture for a high-performance optical flow estimation circuit capable of processing HD images in real time. The algorithm implemented in the circuit is the DIS-FAST optical flow estimation algorithm, significantly reducing time complexity through inverse search. To reduce the memory space in the circuit, only the image data of particular scale in an image pyramid are stored in internal memory, while the image data of other scales are obtained through computation when necessary, minimizing the need for accessing external memory and thereby enhancing the processing speed. The synthesized circuit using 65nm standard cell library includes 313,408 gates and its maximum operating frequency is 333MHz. Hence, HD images are processed at the speed of 33 FPS, satisfying the performance required for real-time processing. |