Abstract

The Sports Action Recognition (SAR) domain plays a significant role in research, supporting applications ranging from coaching assistance and athlete performance analysis to real-time commercial entertainment. Although large-scale and small-scale datasets exist, their direct application to specific sports such as cricket remains challenging due to limited granularity and domain relevance. Existing Cricket Action Analysis (CAA) datasets are constrained by small scale, limited modalities, and narrow focus, primarily on batting actions. To address these limitations, we introduce the Cricket Excited Actions (CEA) dataset, developed in collaboration with professional cricket players. The dataset captures challenging multi-person actions in realistic cricket scenarios. Activity classes including Clean Bowled, Six, Four, and Catches follow official cricket standards and represent crucial match moments. Through precise annotation and empirical evaluation using state-of-the-art action recognition architectures, this work provides a comprehensive benchmark that advances research in cricket action analysis and supports the broader sports analytics community.

Additional Comments