Extração de características de narrativas audiovisuais a partir de elementos visuais e suas relações

Abstract

Videos shared on the internet represent more than 80% of the traffic worldwide. YouTube is considered the second most used platform for video sharing. The optimization of the processes related to content analysis regarding these audiovisual structures is treated as a solution to make these processes more effective. For these reasons, this work presents a tool, developed in Python, which aims to automate the process of extracting visual features from videos shared on YouTube. For this, an architecture with the main processes executed in the proposed tool, which applies two pre-trained deep-lerning models, one for object detection and the other for action detection, was validated. Based on the result obtained, this approach can be integrated into the analysis process of audiovisual narratives, since the characteristics extracted by the tool are related to the main elements that compose the visual resources of the narrative.