文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > Detection of a moving object in the image
侵权投诉

Detection of a moving object in the image

ISSN 1064-2307, Journal of Computer and Systems Sciences International, 2006, Vol. 45, No. 3, pp. 459–468. © Pleiades Publishing, Inc, 2006.

Original Russian Text © R.V. Kutsov, A.P. Trifonov, 2006, published in Izvestiya Akademii Nauk. Teoriya i Sistemy Upravleniya, 2006, No. 3, pp. 129–138.

Detection of a moving object in the image

Detection of a Moving Object in the Image

R. V. Kutsov and A. P. Trifonov

Voronezh State University, Universitetskaya pl. 1, Voronezh, 394893 Russia

Received May 31, 2005

Abstract—Quasi-plausible, maximum-likelihood, and Bayes algorithms for detecting in the image a spatiallengthy object moving with unknown velocity for an applicative model of the interaction of a useful image andthe background are synthesized. The ef ciency of the algorithms and the limitations of the applicability of theasymptotic expressions for the detection features are established by computer statistical simulation.DOI: 10.1134/S1064230706030129

INTRODUCTION

The detection and recognition of objects with thehelp of their images is a topical problem in solvingautomatic monitoring and control problems. One prom-ising direction in the development of these systems isthe use of computer vision techniques applied in navi-gation and guidance systems of aircraft [1]; in aircraft,satellites, and other systems of remote state control ofguided areas; in natural objects, the surrounding envi-ronment, and intruding objects [2]; in medical andindustrial diagnosis systems, etc.

The operation of automatic systems under real con-ditions is accompanied by noise of random characterand different physical natures. The most widespreadmodel of noise uctuation is additive white Gaussiannoise.

The resolution of observation systems, which hasconsiderably improved, has stimulated lately the theoryof the detection of objects based on their images withaccount of background shading. The problems ofdetecting spatial lengthy objects were considered in [3–9] and other works. In [4, 5], it was shown that the useof the additive model of interaction of spatial objectsand the background may result in inauthentic results. In[4–6], based on the applicative model taking intoaccount the effects of shading of a part of the back-ground by an object, the characteristics of the detectionof an immovable spatial object were obtained. In [7],the statistical characteristics of a suboptimal interframedetector of patterns moving with a known velocity froma given position were evaluated under the assumptionthat the speci ed patterns in consecutive frames are notoccluded. In [8], the potentialities of detecting a spatialobject moving with a known velocity were investigated.However, situations when the observer does not knowthe value of the object velocity occur frequently.

The objective of this paper is to synthesize and ana-lyze quasi-plausible, maximum-likelihood, and Bayesalgorithms for detecting spatial objects moving with anunknown velocity based on their images.

1. STATEMENT OF THE PROBLEM

Assume that, in a two-dimensional region on theinterval [0, T], the realization of a Gaussian random eld x(r, t), where r = (x, y) is the radius–vector of thepoint in the plane belonging to and t is the time, isavailable for observation. Suppose that [8], for thehypothesis H1, the eld x(r, t) contains an image of theobject moving from a certain position in a given direc-tion with a velocity V0 s(r – V0t), immovable back-ground ν(r), and additive Gaussian spatial temporalwhite noise n(r, t) with zero expectation and a correla-tion function Kn(r1, t1; r2, t2) = n(r1, t1)n(r2, t2) =N0δ(r1 – r2)δ(t1 – t2)/2, where N0 is the one-sided spec-tral density of the white noise. Under the hypothesis H0,the eld x(r, t) contains the background ν(r) and addi-tive temporal spatial noise n(r, t).

According to the applicative model [4–6], whichtakes into account the effect of shading a part of thebackground by the object, we suppose that the objectimage covers the part s of the region , and the back-ground is formed by the remaining part of the observa-tion region. When there is no object, the background isthe whole observation region. We assume that the reso-lution ability of the observation system is suf cientlyhigh so that the dimensions of the inhomogeneities ofthe object and background are high compared with thedimensions of a pixel in the observation region. Then,on a time interval [0, T], the realization of the image inthe image plane

x(r,t)=γ0[s(r–V0t)–ν(r)]Is(r–V0t)

+ν(r)+n(r,t)

(1.1)

is available for observation, where γ0 = 0, when thehypothesis H0, γ0 = 1 and H1 hold, and Is(r) = 1 for r ∈ s and Is(r) = 0 for r s is an indicator function thatdescribes the object shape. Based on the data observed,it is necessary to make a decision on whether or not theobject is located in the observation region.

459

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email