Enhanced adaptation of MP-GCN for recognizing playground activities using only 2D skeletons and geometric cues. It builds spatio-temporal graphs that capture interactions among people and playground structures, supported by a streamlined pipeline for pose detection, tracking, graph construction, data stratification, and multi-stream training
machine-learning ai deep-learning human-activity-recognition graph-convolutional-networks skeleton-based-action-recognition graph-neural-networks multi-person-tracking group-activity-recognition mpgcn playground-activity-recognition
-
Updated
Dec 7, 2025 - Jupyter Notebook