Human gesture recognition using orientation segmentation feature on random rorest

In the field of gesture recognition, one of the major challenges lies in that different user may sign different style of gesture. Traditional exemplar-based methods are vulnerable to gesture scaling and hand location translating. To overcome such disadvantage, we propose an efficient and inexpensive solution for classifying hand gestures by defining an invariant feature and applying it on random forest. One of the prominent characteristics of gesture is the underlying sequence structure, which can be greatly distinguished from other gestures. Hence, direction of gesture sequence segments has been established as simple comparison features for training random forest classifier, and then predicting gestures at sign piece level. The property of this feature determines that our recognition method can invariant to gesture scaling and hand location translating. It is free to act gesture at any angular field of view and not subject to different acting style of signer. The results show that the performance of proposed method outweighs other state-of-art methods for gesture recognition.