By incorporating hybrid tracking and image matching,
the system provides a much better user experience
by showing the labeled results moving together
with the previously recognized objects in each frame,
even though the expensive image recognition is only
performed for selected frames, as shown in Fig. 4.