Interactive Activity Learning from Trajectories with Qualitative Spatio-Temporal Relation

Automatically analyzing interactions from video has gained much attention in recent years. Here a novel method has been proposed for analyzing interactions between two agents based on the trajectories. Previous works related to this topic are methods based on features, since they only extract features from objects. A method based on qualitative spatio-temporal relations is adopted which utilizes knowledge of the model (qualitative spatio-temporal relation calculi) instead of the original trajectory information. Based on the previous qualitative spatio-temporal relation works, such as Qualitative trajectory calculus (QTC), some new calculi are now proposed for long term and complex interactions. By the experiments, the results showed that our proposed calculi are very useful for representing interactions and improved the interaction learning more effectively.