462KUTSOV, TRIFONOV
2
α0 = α, the probability of missing the object is β0 = 1 –Φ(q0 – q1 – α), where the detection parameter is q0 = (V0, V0).S
As an example, we consider the case of detection ofa uniformly colored object in the form of an ellipse withthe half-axes a and b observed against a homogeneousbackground, i.e., s(r) = s0 and ν(r) = ν0. Assume thatthe motion direction is known and that the object movesalong the half-axis of length a. In this case, signal func-tion (3.2) has the form
(V,V0)=S(τ)S
23/22
2[2+(1–τ)–31–τ]2
-arcsinτ+--------------------------------------------------------------, 1–-π3πτ(3.5)2
= z 0≤τ<1,
4 ---------,τ≥1, 3πτ (V0, V0) = 2πabT(s0 –where τ = |V – V0|T/2a and z2 = S
ν0)2/N0.
If the prior probabilities of the presence p1 andabsence p0 of the object image in the observed realiza-tion, as well as the prior probability density wpr(V) ofthe values of the mean velocity of the object, areknown, then the detection quality can be characterizedby the mean probability of error [12]
Pe*=p0α*+p1β*,(3.6)
where α is the probability of an error of the rst kind(false alarm) determined by formula (3.4),
β*=
is the unconditional probability of an error of the sec-ond kind (missing the object) (3.4) moving with thevelocity V0, while the quasi-plausible detector is aimedat detecting an object moving with the velocity V .Assume that, in the synthesis of the quasi-plausibledetector, the object is supposed to be immovable, i.e.,V = 0, while, in a real situation, the object moves withthe velocity V0 uniformly distributed in the priordomain W with the prior probability density
1/Vmax,V∈W
wpr(V)=
0,V W.
(3.7)
Substituting (3.4) and (3.7) into (3.6), we nd that, forV = 0, the mean error probability is
hz
Pe*=p01–Φ --+--
z2 2+p1-------πm
πm/4
–πm/4
∫
hz
Φ--+--–zS(τ)dτ,z2
(3.8)
where m = VmaxT/πa is the reduced length of the priorinterval of possible values of the velocity values[12, 13].
To determine the detection threshold h, we can usethe criterion of the minimum of the mean error proba-bility Pe* [11, 12], i.e., to nd the threshold from thecondition
h=hopt*=argminPe*.
(3.9)
Thus, the threshold hopt* depends on m, z, and p0.
When V = V0 (the optimal detector), the mean errorprobability is [12]
hzhz
Pe0=p01–Φ --+-- +p1Φ --–-- ,
z2 z2
h=hopt0=argminPe0.
(3.10)
W
∫β*(V*,V)w
pr(V0)dV0
Pe10–110–210–310–4
and the threshold h is determined from the condition
(3.11)
Therefore, hopt0 depends on z and p0.
Figure 1 shows the effect of the OSSh on the meanerror probability Pe* when using the quasi-plausibledetector with optimized threshold (3.9) computed byformula (3.8) for p0 = 0.5. The dashed line, dotted line,and the dashed–dotted line correspond to m = 0.1, 1,and 10, respectively. For comparison, the solid line pre-sents the dependence of Pe0 on z computed by (3.10) forthe optimized threshold h determined by (3.11).
As can be seen, even for relatively small values m,the loss in the value of the mean error probabilityobtained because of lack of knowledge about themotion velocity is quite signi cant. To improve the ef -ciency of detection, it is possible to evaluate the motion
123
Fig. 1.
456
7z
JOURNAL OF COMPUTER AND SYSTEMS SCIENCES INTERNATIONAL Vol. 45 No. 3 2006
NorthChina Univ.ofTechnology,Beljing 100041China) Abstract:Thefordetection....beusedto打ackmoving object, fordigitalimageprocessing discussed. character,...
Detection of a moving ob... 10页 5财富值 A Moving Object Detectio......In the image model Eq. 1, radial photometric distortions of perspective ...
Moving Object Detection and Tracking in Video Based on Higher Order ...to the wavelet transform coecients of the motion compensated image frames....
1, NO. 1, MARCH 1999 65 Detection of Moving Cast Shadows for Object ...In the image model (1), radial photometric distortions of perspective ...
motion and geometry based obstacle detection in image sequences of traffic ...Up to now the classi cation whether a (relatively) moving object is a ...
REAL-TIME MOVING OBJECT DETECTION IN VIDEO SEQUENCES USING SPATIO-TEMPORAL ...foreground objects from the background a pure background image has to be ...
of a Moving Object from Range Video Farzad ...in the range image, and a rigid transform is ...A blob detection algorithm based on the target ...
some extensions to the method for moving object detection presented in [4]...the imaging surface) from spatio-temporal patterns of image intensity [1]....
Detection of a moving ob... 10页 5财富值 Object detection and tra......Their effectiveness in image processing still remains to confirm. The discrete...
be detected as foreground object and plauge the moving objects segmentation. ...shadow detection algorithm can greatly benefit the practical image analysis ...
热门文档