Game Development Reference
In-Depth Information
both foreground and background are encoded more efficiently. It should be noted that
background difference prediction cannot be applied together with multihypothesis
prediction tools to reduce the worst decoding bandwidth.
8.4.3 Performance of AVS/AVS2 Surveillance Coding
We employ another eight CIF&SD videos and two 1,600
1,200 HD videos from
Hisense Co. Ltd to evaluate the BHO. Figure 8.12 shows all these videos. As we
can see, these ten surveillance videos cover different monitoring scenes, including
bright and dusky lightness (BR/DU), large and small foreground (LF/SF), fast and
slow motion (FM/SM). These videos have been either utilized in the standardization
process of AVS2 (Marpe et al. 2006 ) or obtained from the PKU-SVD-A dataset
(Zhang et al. 2014b ).
On the AVS2 common test conditions with five SD and HD surveillance videos,
the video coding performance under the surveillance tools is compared with the
performance without these tools. Results in Table 8.8 show, these tools achieves 52.6,
46.4, and 28.5% bit savings on random access, random access without B frames and
low-delay coding.
×
Fig. 8.12 All surveillance videos used in the experiments
 
Search WWH ::




Custom Search