Multimodal Deep Learning Approaches in Sports Analytics: A Survey

In recent years, the rapid evolution of consumer electronics has driven significant advancements in smart products. This progress has enhanced data collection technologies and enabled multiple modalities for acquiring information. Consequently, sports research benefits from higher data quality and more consistent long-term support. Multimodal sports data research collects information from diverse modalities, both visual and non-visual, and integrates them through alignment and fusion techniques to produce datasets suitable for downstream applications. This survey reviews the prevalent applications of multimodal sports data, focusing on its use in prediction, identification, prevention, and diagnosis. Furthermore, the survey discusses challenges in alignment and fusion, explores the future role of consumer electronics in sports, and highlights emerging perspectives on multimodal sports data development in the context of Artificial Intelligence Generated Content (AIGC).