Application of a multimodel data processing pipeline for solving the video lecture scene segmentation problem

Subject of research: the set of methods and models of machine learning and artificial intelligence in the field of computer vision that ensure the analysis and processing of video lectures as multimodal data sources. Purpose of research: to develop and substantiate an approach to solving the non-trivial problem of segmenting a video lecture into scenes, each corresponding to a unique frame that reflects textual content at a specific stage of material presentation, with the aim of subsequent processing of the obtained results using optical character recognition methods. Research methods: the primary research method is the decomposition of a video lecture as a multimodal object into individual modalities, followed by data analysis based on a multimodel data processing pipeline. Objects of research: within the scope of this paper, the objects of the study are video lectures of three main types, serving as sources of multimodal data for analysis and processing. Research findings: a multimodel data processing pipeline has been developed and substantiated for each of the identified types of video lectures. An evaluation of the applied models was conducted, and their influence on the formation of the final processing results was analyzed.