A Survey on Expressive Action Quality Assessment Driven by Aesthetics and Biomechanics

The evolution of artificial intelligence and computer vision technologies has driven Action Analysis toward more refined areas. As a unique form of human movement for artistic creation, emotional storytelling, and competitive expression through body language, Expressive Action Analysis holds significant research value. It plays a crucial role in advancing the digital transformation of the cultural and artistic industries, optimizing competitive sports' scientific training systems, and ensuring competition fairness. Furthermore, it differs significantly from functional actions in terms of both the essence of movement and evaluation dimensions. This paper systematically summarizes the four core challenges in Expressive Action Quality Assessment: the lack of mathematical modeling frameworks, the inconsistent evaluation standards caused by diverse artistic styles, the limitations of traditional methods in capturing spatiotemporal features, and the absence of technical frameworks for dynamic modeling in complex interactive scenarios. To address these issues, this paper proposes four innovative research directions: a multimodal feature fusion architecture, generalizable representation learning across styles, spatiotemporal joint dynamic modeling strategies, and modeling constrained by human-object interaction. We also argue for the necessity of constructing an aesthetics-biomechanics dual-driven evaluation theoretical framework and explore the significance of developing an explainable evaluation model enhanced by contextual prior knowledge. Finally, the paper envisions the future development of this field, focusing on constructing standardized benchmark databases, designing multidimensional quantitative evaluation systems, and promoting interdisciplinary integration, real-time computing optimization, and collaborative innovation in personalized evaluation models.