The designed blob fusion algorithm has the objective of correcting
possible deficiencies in foreground objects detection produced
by segmentation errors, as incomplete or fragmented person figures
in the segmented images. In some situations, the video frames
have excessive noise or data loss due to the video transmission
from the shopping malls to the control center (see Section 4 and
Fig. 6 to understand these problems derived from the data communication
model and the usage of IP cameras). These difficulties
cause poor segmentation results in some cases, as is presented in
Fig. 3, where a frame with data loss and strong pixelation gives
several foreground blobs where there is only one person, which
is corrected with our blob fusion method. This approach is also
supported by Brutzer et al. (2011), which opens a discussion about
the usage of post-processing techniques for correcting background
subtraction errors.