Point trajectory-based inference of what the videographer wanted to capture