Variable structure multiple model for articulated human motion tracking from monocular video sequences.