Deep Video Stream Information Analysis and Retrieval: Challenges and Opportunities
Nikolaos Passalis, Maria Tzelepi, P. Charitidis, Stavros Doropoulos, Stavros D. Vologiannidis, Anastasios Tefas · 2022
Deep Learning (DL) provided powerful tools for various visual information analysis and retrieval tasks, outperforming previously used methods. However, despite the potential of such approaches for various tasks, applying them in video stream applications, such as media monitoring or surveillance, where a large number of streams should be processed in parallel, is not trivial and comes with several challenges. This paper aims to provide a brief overview of the current state-of-the-art in DL tools that can be used for deep video stream information analysis and retrieval. Apart from a review of the current literature, we also include experimental results discussing deployment challenges, ranging from speed to energy consumption, demonstrating the capabilities of readily available commodity hardware in processing video streams for selected DL models.