Text this: Fragment-based visual tracking with adaptive appearance model