Dynamic hand gesture recognition based on 3D pattern assembled trajectories

Said Yacine Boulahia (1, 2) , Eric Anquetil (1, 2) , Franck Multon (3, 4) , Richard Kulpa (5, 3, 4)

1 IntuiDoc - intuitive user interaction for document
2 INSA Rennes - Institut National des Sciences Appliquées - Rennes
3 M2S - Laboratoire Mouvement Sport Santé
4 MIMETIC - Analysis-Synthesis Approach for Virtual Human Simulation
5 CERV - Centre Européen de Réalité Virtuelle

Over the past few years, advances in commercial 3D sensors have substantially promoted the research of dynamic hand gesture recognition. On a other side, whole body gestures recognition has also attracted increasing attention since the emergence of Kinect like sensors. One may notice that both research topics deal with human-made motions and are likely to face similar challenges. In this paper, our aim is thus to evaluate the applicability of an action recognition feature-set to model dynamic hand gestures using skeleton data. Furthermore, existing datasets are often composed of pre-segmented gestures that are performed with a single hand only. We collected therefore a more challenging dataset, which contains unsegmented streams of 13 hand gesture classes, performed with either a single hand or two hands. Our approach is first evaluated on an existing dataset, namely DHG dataset, and then using our collected dataset. Better results compared to previous approaches are reported.

Mots clés: Leap Motion dataset; Skeleton joint data; 3D dynamic hand gesture recognition; Real-time recognition

Domaines: Multimédia [cs.MM]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Apprentissage [cs.LG]; Interface homme-machine [cs.HC]; Synthèse d'image et réalité virtuelle [cs.GR]

Said Yacine Boulahia, Eric Anquetil, Franck Multon, Richard Kulpa. Dynamic hand gesture recognition based on 3D pattern assembled trajectories. IPTA 2017 - 7th IEEE International Conference on Image Processing Theory, Tools and Applications, Nov 2017, Montreal, Canada. pp.1-6. ⟨hal-01666377⟩