Multi-Module Human Motion Analysis from a Monocular Video