Game Development Reference
In-Depth Information
include a backgroundmodeling and updatingmodule, a background-basedmotion
compensation module, a background buffer and related control flows (marked by
red color).
2. Low-complexity background modeling: A segment-and-weight based running
average (SWRA) is utilized to approximately calculate the background by assign-
ing a larger weight on the frequent values in the averaging process. Technologi-
cally, SWRA divides the pixels at a position in the training frames into temporal
segments with their own mean values and weights, and then calculates the running
and weighted average result on the mean values of the segments. In the process,
pixels in the same segment have the same background/foreground property and
the long segments have larger weights. Experimental results show that SWRA
can achieve good performance yet without suffering a large memory cost and
high computational complexity (Huang and Liao
2001
).
3. G-picture based background prediction: To realize better prediction efficiency of
the background pixels in the current frame, the G-picture can be quantified with
a much smaller QP and be encoded as a nondisplay frame. Moreover, IEEE 1857
surveillance groups also adopt the S-picture as a special P-picture which can be
only predicted from the reconstructed G-picture. Therefore, each P frame can be
predicted not only from two recent frames, but also from one recent frame and
the G-picture.
4. Optional difference coding for mixed macro-blocks (MBs): IEEE 1857 surveil-
lance groups can utilize the traditional entropy-based coding tool for each MB
in a P-picture. In addition, it can also optionally use the background difference
prediction strategy, namely, to predict the difference between the current MB and
its corresponding background MB using the different data between the recent
reference frame and the G-picture. In such a way, MBs with both foreground and
background can be encoded more efficiently.
5. Improved motion vector prediction: An algorithm is specially designed to derive
the predicted motion vector (PMV) when using the not-display G-picture as direct
or indirect prediction reference. That is, if one neighbor of the current MB utilizes
or does not utilize the not-display G-picture as reference, the contribution of this
neighboring block in deriving the final PMV of the current MB should be set to 0.
In addition, when the colocated block in the backward reference frame utilizes the
not-display G-picture as reference, the spatial derivation instead of the temporal
derivation of PMV should be adopted. In this way, dividing-zero error can also
be avoided.
Overall speaking, compared with other video coding standards such as
AVC/H.264, HEVC/H.265, the most important contribution of IEEE 1857 surveil-
lance groups is to seamlessly integrate the model-based coding (including back-
ground modeling and foreground coding) with the traditional hybrid video coding
framework. As far as we concerned, the IEEE 1857 is the first video coding standard
specially designed for surveillance videos. These model-based coding techniques in
IEEE 1857 surveillance groups can be embedded into the cutting-edge HEVC/H.265
framework, so as to optimize its encoder for further increasing the coding efficiency
