Creating an Original Technique for Identifying Scenes in Multimedia Streams

In the course of practical operation in automated systems that work with multiple video streams, it often becomes necessary to identify certain objects or scenes in video frames. In this regard, this work is aimed at analyzing and improving computer vision methods that provide identification of specified images in video streams. The original method for identifying images in video streams presented below implements the calculation of descriptors and comparison of similarity measures with the search object. The presented technique is characterized by high performance and can significantly reduce the time of analyzing a video stream. A feature of the technique is to narrow the search range by clarifying the intersection of identifiers. The developed method has shown high efficiency in relation to a large part of modern analogs. In the process of creating methodology, it was compared with such identifying scene methods as, for example, the “SURF” and “Kudan” methods. The comparison results allow us to conclude that the presented method is approximately 7–23% more effective than that of analogues.