Section: New Results

A New Hybrid Architecture for Human Activity Recognition from RGB-D videos

Participants : Srijan Das, Monique Thonnat, Kaustubh Sakhalkar, Michal Koperski, Francois Brémond, Gianpiero Francesca.

Keywords: Visual cues, Data fusion, RGB-D videos

Activity Recognition from RGB-D videos is still an open problem due to the presence of large varieties of actions. We have proposed a new architecture by mixing a high level handcrafted strategy and machine learning techniques. In order to address the problem of large variety of actions, we proposed a novel two level fusion strategy to combine motion, appearance and 3D pose information. For 3D pose information, we use the work published in AVSS 18 (described above). As similar actions are common in daily living activities, we also proposed a mechanism for similar action discrimination using dedicated SVMs. We validated our approach on four public datasets, CAD-60, CAD-120, MSRDailyActivity3D, and NTU-RGB+D improving the state-of-the-art results on them. The proposed architecture has been published in the industrial session of MMM 2019 [41].