Posture Recognition for Human Interface in Virtual Reality

This paper introduces a method for improving the accuracy of human posture recognition using non-wearable sensors. We tackle the challenges associated with motion recognition when using a low-precision depth camera. In our study, we define pose recognition as a classification task within a multi-dimensional space. We propose a spatial modeling approach that utilizes a lazy learning system, enabling robust posture recognition despite low-quality motion data or significant variations in user actions. The input motion data is transformed into informative feature vectors, normalized to the user's initial pose. Experimental results show high performance and accuracy, even in the presence of unstable or erroneous motion input.