Abstract

We present a system for the temporal detection of social interactions. Many of the works until now have succeeded in recognising activities from clipped videos in datasets, but for robotic applications, it is important to be able to move to more realistic data. For this reason, it is important to be able to detect temporally the intervals of time in which humans are performing an individual activity or a social one. Recognition of the human activities is a key feature for analysing the human behaviour. In particular, recognition of social activities could be useful to trigger human-robot interactions or to detect situations of potential danger. Based on that, this research has three goals: (1) define a new set of descriptors able to represent the phenomena; (2) develop a computational model able to discern the intervals in which a pair of people are interacting or performing individual activities; (3) provide a public dataset with RGB-D videos where social interactions and individual activities happen in a continuous stream. Results show that using the proposed approach allows to reach a good performance in the temporal segmentation of social activities.