NOP query: um novo método orientado a notificações para processamento de buscas em vídeos em tempo real

Abstract

With the rapid growth of video data, the search for events in videos is becoming increasingly relevant, and many challenges related to computational cost and latency arise, especially for real-time applications. In this context, important works have highlighted many issues when dealing with large video datasets or streaming data, including high processing cost, complex imperative programming, high latency with query processing and a lot of complexity to detect and retrieve spatiotemporal events in videos. Additionally, there is a lack of a video standard query language and an efficient search engine, which requires a pragmatic solution, as was the case with ANSI SQL for relational databases. From the epistemological and ontological point of view, the origin of the problem would be beyond the current approaches, reaching its usual programming paradigms and related idiosyncrasies, such as the tendency to misuse processing capacity and program part coupling. The so-called “paradigm unfitness” would be a relevant and even primary cause of software development problems that hinder breakthrough solutions in the video query system as much as other domains involving distributed and high-demanding processing. Considering the need for new and efficient solutions, the so-called NotificationOriented Paradigm (NOP) could be a suitable alternative. NOP introduced a new way of thinking and developing software in which small collaborative entities perform fact execution and logical decision processing based on precise notifications. Inspired by the current gaps in the field of video querying and the advantages of NOP and existing development tools, this thesis proposes a new query-processing method that handles multiple video events and reacts with low latency when matches occur. A distinct conformation of the state-of-the-art NOP Framework is proposed and applied to query the system domain to avoid intensive database operations such as storage, retrieval, and triggers. In order to perform query searches, a new video querying language is proposed, developed, and used. In addition, a new dataset and two case studies with several experiments are proposed and developed to show the feasibility of the proposed NOP-based solution for this domain. Experiments on the labeled dataset demonstrated the suitability of the proposed method for low-latency processing with polynomial complexity. The results are better than the state of the art, which works at exponential cost.