Current Trends and Challenges in Egocentric Video Analysis For Learning from Demonstration

Egocentric video analysis has gained significant attention in recent years because of its ability to capture visual information from a first-person perspective that closely mimics human perception.This paradigm plays a crucial role in learning from demonstrations, wherein machines learn tasks by observing human actions.This paper presents a conceptual and experimental study on egocentric video-based learning, focusing on shape and alphabet recognition in real-world first-person videos.Two egocentric videos were used: the first demonstrated the sequential drawing of geometric shapes, while the second illustrated the drawing of alphabets inside the same shapes.The proposed approach enables the system to first recognize the shapes and subsequently detect the letters contained within them.This paper discusses current trends and underlying concepts, challenges such as camera motion and data quality, and provides experimental observations.The results highlight the feasibility of egocentric video analysis for structured visual learning tasks and identify key research challenges for future studies.