Year: 2025
Type: conference-paper
Abstract: In this study, selective intersection flow (SIF), a lightweight optical flow algorithm, was used to enhance efficiency and accuracy by filtering out non-contributive pixels. SIF, derived from the differential category of algorithms, is used to compute optical flow by analyzing intersections of equations from selected pixels rather than solving for all pixels. It replaces warping with a minimal computational procedure for initial flow estimate and employs a sliding window for optimized single-core performance. SIF runs 1.7–1.8 times faster and achieves 1.2–1.4 higher accuracy than the single iteration of the Lucas–Kanade method, showing promise for real-time micro drone navigation.
Authors: Che Liu, Chen-Fu Yeh, Chung-Chuan Lo
Institution: National Tsing Hua University
Language: English
Topic: Advanced Vision and Imaging
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4414409875