Recognizing Context for Privacy Preserving of First Person Vision Image Sequences