Kinect Gesture Recognition: SVM vs. RVM

Human gesture recognition has been an active and challenging problem, especially when motion capture devices become more popular. Various studies have shown that support vector machines (SVMs) with Gaussian kernels are among the most prominent models for an accurate gesture classification. We demonstrate in this paper that the relevance vector machines (RVMs) could also achieve the state-of-the-art predictive performance. Moreover, RVMs run much faster than SVMs in testing phase. Intensive experiments on the Microsoft´s MSRC-12 Kinect gesture data set also pointed out that prediction behaviors of SVMs and RVMs are very similar in terms of parameter sensitivity, accuracy in leave-subject-out test, and gesture discrimination.