Motion vector extraction method for computer vision systems employing lossy compression

Purpose of research. In the modern digital era, the role of video cameras as high‑quality sources of primary data is increasing. However, raw video data by themselves carry low informational value without subsequent analysis. The key tasks that allow extracting semantic information from a video sequence are object localization and motion detection. The relevance of this task is determined by its critical importance for a wide range of applied and research disciplines. Despite its long history, detecting moving objects remains a relevant scientific problem due to the following challenges: variability of lighting conditions, dynamic background, occlusion effects, and the need to operate in real time. The aim of this work is to reduce computational load when solving object motion analysis tasks in real time by developing and testing a method for extracting motion vectors from compressed video streams. Methods. To achieve this goal, the framework of motion vectors was used as the basis for compensating temporal redundancy, along with computer vision algorithms and motion compensation algorithms in video data. Results. A software module has been developed that allows extracting motion vectors directly from a video stream. An experimental evaluation of the proposed method’s effectiveness was carried out, demonstrating its efficiency in various applied fields, including video surveillance, agriculture, and robotics, with a significant reduction in computational costs. Conclusion. The experimental evaluations have shown that using motion vectors already contained in compressed video data allows effectively solving motion analysis tasks without the need to recalculate them. This is especially relevant for systems with limited computational resources.