A Spatial and Temporal Features Mixture Model with Body Parts for Video-based Person Re-Identification