Game Development Reference
In-Depth Information
Fig. 8.7 The calculation of
buffered AVG and W
relies on the float precision calculation of proportion and variance, so both memory
and time are saved; ComparedwithMean Shift, SWRAdoes not need to allocate large
memory to buffer multiple training frames, so memory will be significantly reduced;
This method is also different from nonparametric methods like codebook, although
the codebook does not need to buffer multiple training frames, the management of
the multiple codewords is very time consuming and memory sparing.
Table 8.6 shows the background modeling time. The RA spares the least mod-
eling time and MS spares the largest. Moreover, it shows that SWRA spares much
less time than MS, GMM-1 and GMM-5 on all the sequences, only about 25% of
the computing time spared by GMM-5. Moreover, SWRA is not sensitive to video
content, which is quite different from GMM-X and MS.
Another comparison for the memory cost of RA, MS, GMM-1, GMM-5, and pro-
posed SWRA is shown in Table 8.7 . In further for SWRA, the required memorized
data for each pixel include: one current pixel with type of char and one float-precision
mean value; two float-precision mean values avg'/avg and their corresponding char-
type weights. In summary, the total memory cost is no more than 14 bytes for each
pixel. Results show that SWRA helps to achieve better performance than other mod-
eling algorithm on average. Moreover, compared to the state-of-the-art GMM-5,
SWRA only consumes 10% of the memory cost and spares 25% of GMM-5's mod-
eling time.
Table 8.6 Background modeling-based FET-X versus T-AVC on PSNR gain (dB) and modeling
time (second) on surveillance videos
FET-
Crossroad-sd
Overbridge-sd
Bank
Office
SD average
GMM-1
6.6
6.6
6.6
6.7
6.6
RA
24.4
24.5
24.4
24.5
24.5
MS
43.3
41.8
41
46.4
43.1
GMM-5
242.8
236.2
235.9
252.8
241.9
SWRA
10.7
10.4
10.3
10.5
10.5
FET-
Crossroad-cif
Overbridge-cif
Snowgate
Snowroad
CIF average
GMM-1
1.6
1.7
1.7
1.7
1.7
RA
5.9
5.9
5.8
5.8
5.9
MS
11.5
11.3
10.4
10.3
10.8
GMM-5
61.8
61.2
57
56.4
59.1
SWRA
2.2
2.3
2.3
2.3
2.3
 
Search WWH ::




Custom Search