A Neural Classifier for Anomaly Detection in Magnetic Motion Capture

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.