DRONE VIDEO PROCESSING BY FRAGMENT ANALYSIS
Sergii Mashtalir, Dmytro P. Lendel · System technologies · 2025
This study focused on analyzing drone video to address the inherent complexity of processing videos captured with a moving camera. Each frame is divided into smaller fragments through a grid-based segmentation technique, enabling a localized and detailed motion analysis. Singular Value Decomposition (SVD) is applied to compute the Ki Fan norm for each fragment, enabling the detection of dynamic changes between consecutive frames. The fragment-level analysis allows the algorithm to robustly identify regions of interest and dif-ferentiate between global motion (camera movement) and local motion (object movement) despite challenges introduced by camera instability. The motion state is classified into four categories: stable camera with no object movement, stable camera with object movement, moving camera with no object movement, and moving camera with object movement. This fragment-based methodology enhances precision in dynamic scene analysis, offering a scal-able and efficient solution for applications such as video stabilization, object tracking, and real-time motion detection in complex environments.