STAB-GCN: A Spatio-Temporal Attention-Based Graph Convolutional Network for Group Activity Recognition

Group activity recognition is a central theme in many domains, such as sports video analysis, CCTV surveillance, tactics, and social scenario understanding. However, there are still challenges embedding actors’ relations multi-person due to occlusion, movement, light. Current studies mainly focus on collective individual local features from the spatial temporal perspectives, which results inefficiency, low robustness, portability. To this end, Spatio-Temporal Attention-Based Graph Convolution Network (STAB-GCN) model proposed effectively embed deep complex between actors. Specifically, we leverage attention mechanism attentively explore spatio-temporal latent This approach captures contextual information improves group embedding. Then, feed actor relation graphs built videos into our STAB-GCN for further inference, selectively attends relevant while ignoring those irrelevant extraction task. We perform experiments three available datasets, acquiring better performance than state-of-the-art methods. The verify validity of highlight obstructive impacts attention-based graph recognition.