Skeleton-based relational reasoning for group activity analysis

Research on group activity recognition mostly leans the standard two-stream approach (RGB and Optical Flow) as their input features. Few have explored explicit pose information, with none using it directly to reason about persons interactions. In this paper, we leverage skeleton information learn interactions between individuals straight from it. With our proposed method GIRN, multiple relationship types are inferred independent modules, that describe relations body joints pair-by-pair. Additionally relations, also experiment previously unexplored relevant objects (e.g. volleyball). The distinct then merged through an attention mechanism, gives more importance those for distinguishing activity. We evaluate in Volleyball dataset, obtaining competitive results state-of-the-art. Our experiments demonstrate potential of skeleton-based approaches modeling multi-person