International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
In (Anandan, 1989) SSD-function is sequentially optimized by
the Laplacian pyramid. Minimum is found for all levels of the
pyramid to begin with the highest level (the smallest image) and
dropping to the lowest level (the whole image). Speed vector is
being obtained more accurate on each level. In (Singh, 1992)
minimum of SSD-function is found through iteration process.
But correlation approach is not robust too because it strongly de-
pends on invariability of scene brightness. In (Heeger, 1988) fre-
quency approach is proposed. This approach is based on “power”
function, evaluated as the Gabor filter (Gabor, 1946) with fre-
quencies Lz, Ly, w:
R(u,v) = nd —4n?o2070, (uLz vL, + w) }
(uozok)? + (voyok)? + (020y)?
where 95, 0, 04, — standard Gabor filter derivatives.
Speed vector (u, v)? is found during minimization of function
12
it M JF, (u,v)
flu, v) = 2 mi ne ue)
with respect to u and v, where m; — measured power value, R; —
predicted power value, m; and R; are average power values.
3 REGRESSION PSEUDOSPECTRUMS
In this section we introduce the notion of multiple-regression
pseudospectrums.
Let again /(k) be an input image pixel matrix with width w and
height h on frame number k, I(k) € R"*^,. It is assumed that
I(%) is a grayscale image, so 0 < I(k),, «€ 255 Vr =
l...w, y = 1...h Let us call Mn(k) an regression accu-
mulator of n frames with parameter à, calculated on frame k. It
will be a matrix Mn (k) € R**"(Box et al., 1994):
M,(k + 1) = aM,(k) + (1 — a)I(k). (1)
You can calculate the accumulator value M,(k) on frame k by
adding each older member in series (1):
e
E
Ma(k) — (1-0) M. a^ 7I. Q)
iz0
ll
Let us assume that [(k) is an element of the image matrix I (k)
and mn(k) is an element of the accumulator matrix M,, (k) with
the same coordinates, as [(k). Let us suppose that on an initially
zero input of accumulator (2) since some moment ko (without
loss of generality, let ko — 0), during enough long time some
signal with intensity / is being given:
k—1
ma(K) 7 I(1— 0) 7 a*1 zi - o^). (3)
i=0
Now it's quite simple to find such o, so that m,, (k) would surely
exceed [3 share of signal [ after n frames:
Mn{n) = I(1 — a”) = Bl.
Hence
Un = Yi = B. (4)
560
Thus, o is such time averaging parameter, at which the accu-
mulator sum will be equal to m,(n) = SI through n frames. At
the same time n here can be called 5 memory length or, simply,
length of accumulator memory with the corresponding averaging
parameter o, — V1- f.
Given o; can be found as (4), the whole accumulator sum in one
pixel at variable frame k can be found as
ma(k) 2 (1 — o5) 2 1(1— (1 — 8)//^). (5)
The m,(k) graphs for different an,n = 4, 8, 16, 32 values are
shown in Figure 1, supposed 4 = 0.5,/ = 100, ko = 10.
HS
3
Lo
3 » signal!
I
aeu
dei
quia
5i i
ü m 20
x
frame number
Figure 1: Accumulated pixel mn (k) for different an.
Thus, time averaging parameter œn, defined in (4), is in fact the
satiety parameter of the filter response function. It allows to judge
after which time (in frames) n accumulated sum will be equal to
Bl.
According to (5), œn possesses the multiplicity property:
Un = One: (6)
Indeed, 0? ; — ( nl B) = V1- 8 = an. Let us call
Dn,s(k) = mn(k) — Mn-s(k)
a difference between the responses of accumulators with multi-
ple smoothing parameter n and n - s. By (6) and assuming that
some signal with intensity / is being given from time ko = 0, this
difference will possess a very interesting property:
Dy s(n:5s) 2 ma(n- s) — mas(n-5) — (7)
-i(1-a-5*) -i(1-a- 9*9) =
-i0-9(-0-873)-11-9)Y-g.
Consider the behaviour of derivatives D, (k) function. Let s —
2 and 8 — 0.5. Then, according to (7), difference between accu-
mulator with memory of 2n frames and accumulator with mem-
ory of n frames will be equal to
Da a(2n) z: 0.951. (8)
Figure 2 shows differences D,, s between accumulators with vari-
able n and s = 2,1 = 100.
Int
prs
Figure
As you (
tors 4 - /
the signa
over, it i:
can be re
Thus, fii
memory
wavelet
dospectr
lators (7
sequence
pseudosy
gate botl
as "mear
If the m:
consister
N, but fc
signal m
signal ha
was othe
Similarly
signal. C
reached,
Expectec
example,
value of
until ma»
maximur
Now con
old of th
ages. Fig
trum for
sequence
Apparent
compone
rection fi
of the frc
extremun
magnitud
drop esti
D16,2(k)
dospectru
consider
Consider
nent on ]