Situated Instructions and Guidance For Self-training and Self-coaching in Sports

Recent advancements in Virtual Reality (VR) and Augmented Reality (AR) have made remote education immersive and 3D, gaining traction in various fields like medicine, entertainment, education, and engineering. Remote sports training has also gained attention, leading to the development of applications for analyzing movements and improving skills through 3D visualization. However, automatic guidance systems are not well-researched and limitations of existing motion capture setups have been highlighted. In this thesis, we propose to use a combination of video analysis, AR, and activity recognition for remote sports training. Our approach includes two capture modes (egocentric and exocentric) for indoor and outdoor activities and utilizes computer vision-based methods for motion estimation instead of relying on wearable sensors that need expensive facilities. We explore different visualization modes and plan to develop a deep-learning model to provide automatic guidance during training sessions.