Robust Spatio-Temporal Features for Human Interaction Recognition Via Artificial Neural Network

Human Interaction Recognition plays a key role in identification of usual and unusual human behaviors and facilitates public dealings, violence detection, robots perception, and virtual entertainments. This paper presents a novel human interaction recognition (HIR) system to recognize human interactions in continuous image sequences. The proposed technology segments full body silhouettes and identifies key body points to extract robust spatio-temporal features having distinct characteristics for each interaction. Our descriptors focus on local descriptions, capture intensity variations, point-to-point distances and time based relations. The system is trained through artificial neural network to recognize six basic interactions taken from UT-Interaction dataset.