Viewpoint invariant sign language recognition

摘要

Viewpoint invariance is a grand challenge for sign language recognition. In this paper, we propose a novel viewpoint invariant method for sign language recognition. The recognition task is converted to a verification task under the proposed method. This conversion is based on the geometric constraint that the fundamental matrix associated with two views SHOULD BE UNIQUE when the observation and template signs can be considered as obtained synchronously under a virtual stereo vision and vice versa. The Dempster–Shafer theory is applied to improve the robustness of the geometry model. Our experiment demonstrates the efficiency of the proposed method. Furthermore, the proposed method can be extended to other recognition tasks, such as gait recognition and lip-reading recognition.