A neural classifier for anomaly detection in magnetic motion capture

Abstract

Over recent years, the fall in cost, and increased availability of motion capture equipment has led to an increase in non-specialist companies being able to use motion capture data to guide animation sequences for computer games and other applications.[1] A bottleneck in the animation production process is in the clean-up of capture sessions to remove and/or correct anomalous (unusable) frames and noise. In this paper an investigation is carried out into whether the 2-layer SOM network previously designed [5] and trained on one capture session, can be used to create a neural classifier to be used to classify another separate capture session. © IFIP International Federation for Information Processing 2006.