Browsing by Author "Fernandes, Guilherme Poeta"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Video Action Recognition for Indoor SportsPublication . Fernandes, Guilherme Poeta; Silva, Bruno Miguel Correia da; Proença, Hugo Pedro Martins CarriçoThe increasing demand for Artificial Intelligence (AI) technologies in sport has emerged as a major note in recent years. Besides helping referees at crucial decision times, AI has become a game changer, giving coaches advanced analytical capabilities that dramatically improve their teams’ training strategies, performance and athleticism during games through AI integration of computer vision into sports applications. Through wearable technology, this transformation has been accelerated by sensors and positioning devices that provide valuable information, but overcome their real-time limitations through the ability of video analytics to extract data needs to be increased. Television (TV) AI technology in the sports industry has greatly enhanced sports coverage, transforming the viewing experience for the fans at home. This shift was driven by the evolution of human action recognition in computer vision, which made action classification and spatio-temporal action placement an important area that has been extensively studied through motion recognition. While video analysis promises new applications in sports, it also presents challenges in a variety of settings, including changes in speed, camera placement, background noise, and field conditions during recording. Independent action detection is based on various factors such as human position and interaction with objects. These difficulties can be overcomed by designing segments that adequately describe motion and developing classifications based on these selected features A good solution to video recognition problems lies in the use of DNN. Unlike traditional machine learning, DNN are able to learn features representing specific behaviors from AI systems without requiring extensive manual processes In this regard, this Dissertation focused on research and techniques in the field of computer vision for indoor sports, namely Basketball. In this work, we aim to understand the efficiency of CNN compared to other traditional methods and develop an action recognition algorithm for Basketball games. This system provides valuable data for coaches to monitor and improve their team’s performance. By bridging the gap between technology and sports, this Dissertation seeks to improve Basketball research, applying deep learning techniques to data sets to find reliable solutions.