Digital Signal Processing Reference
In-Depth Information
Fig. 5.16
Video tracking of overlapping objects:
First & Second row
: Three people sequence;
Third & Fourth row
: IU sequence;
top
:mask;
bottom
: superimposed mask
From the comparison of Fig.
5.15
b,c, the tracking errors in Fig.
5.15
b without
motion occlusion analysis have been successfully handled using the proposed track-
ing strategy towards modeling the motion occlusion as layer transition, which can
achieve more precise representation of individual object by eliminating the lost re-
gions or mislabeling of the new uncovered parts. Object tracking results on every
11th frame on
Three-People
and
IU
sequences are shown in Fig.
5.16
. The selected
frames show the average performance of video tracking, which contain typical track-
ing problems such as objects' partial occlusion, separation, and appearance of new
part. 2D regions of each object are tracked consistently and correctly across the
video sequences with various dynamics in the scene. The satisfactory segmentation
and tracking results presented in Fig.
5.16
demonstrate the efficiency and robustness
of the proposed algorithm in subjective performance.
