Skeleton Split Strategies for Spatial Temporal Graph Convolution Networks

Action recognition has been recognized as an activity in which individuals’ behaviour can be observed. Assembling profiles of regular activities such daily living support identifying trends the data during critical events. A skeleton representation human body proven to effective for this task. The skeletons are presented graphs form-like. However, topology a graph is not structured like Euclidean-based data. Therefore, new set methods perform convolution operation upon proposed. Our proposal based on Spatial Temporal-Graph Convolutional Network (ST-GCN) framework. In study, we proposed improved label mapping ST-GCN We introduce three split techniques (full distance split, connection and index split) alternative approach operation. experiments study have trained using two benchmark datasets: NTU-RGB + D Kinetics evaluate performance. results indicate that our outperform previous partition strategies more stable training without edge importance weighting additional parameter. provide realistic solution real-time applications centred systems indoor environments.