Discriminative human action recognition using HOI descriptor and key poses

摘要

Action recognition on large categories of unconstrained videos is still a challenging one due to appearance variations, recording settings and inter-personal differences. A highly discriminating descriptor is introduced which afford context information to recognition human actions in a video. We focus to discriminate the actions having identical spatio-temporal relations by human-object interaction analysis and with similar motion pattern by pose estimation. Robust solutions obtained in this process find immense applications in domains such as visual surveillance, video retrieval and human-computer interaction. We validate our approach on two human activity datasets and show performance improvements over competing state-of-the-art representations.