Spatial and Rotation Invariant 3D Gesture Recognition Based on Sparse Representation

Advances in motion tracking technology, especially for commodity hardware, still require robust 3D gesture recognition in order to fully exploit the benefits of natural user interfaces. In this paper, we introduce a novel 3D gesture recognition algorithm based on the sparse representation of 3D human motion. The sparse representation of human motion provides a set of features that can be used to efficiently classify gestures in real-time. Compared to existing gesture recognition systems, sparse representation, the proposed approach enables full spatial and rotation invariance and provides high tolerance to noise. Moreover, the proposed classification scheme takes into account the inter-user variability which increases gesture classification accuracy in user-independent scenarios. We validated our approach with existing motion databases for gestural interaction and performed a user evaluation with naive subjects to show its robustness to arbitrarily defined gestures. The results showed that our classification scheme has high classification accuracy for user-independent scenarios even with users who have different handedness. We believe that sparse representation of human motion will pave the way for a new generation of 3D gesture recognition systems in order to fully open the potential of natural user interfaces.