A real-time motion capture framework for synchronized neural decoding

Neural decoding is an active research area concerned with how sensory and other information is represented in the brain by networks of neurons. An important step in neural decoding research is to collect the subject´s motion and neural activities synchronously, which requires a real-time motion capture system with high accuracy. In this paper, we propose a practical motion capture framework with the capability of processing motion data and output character animation in real-time. We use a two-stage coarse-to-fine method to preprocess the raw motion capture data. We employ Kalman filter to coarsely estimate the positions of missing markers and filter out the possible noisy markers. The positions of the missing markers are refined with the relationship between the current frame and similar frames in motion templates. We operate the motion data in PCA space to reduce computational complexity. We present the results for our approach as applied to capturing human hand motions, which demonstrates the accuracy and usefulness of our real-time motion capture framework.