東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

A perspective systems approach to pa...

Loucks, Edward Philip.

FindBook

Google Book

Amazon

博客來

A perspective systems approach to parameter identification problems in machine vision.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	A perspective systems approach to parameter identification problems in machine vision./
作者:	Loucks, Edward Philip.
面頁冊數:	135 p.
附註:	Source: Dissertation Abstracts International, Volume: 56-03, Section: B, page: 1680.
Contained By:	Dissertation Abstracts International56-03B.
標題:	Engineering, System Science. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=9523630

A perspective systems approach to parameter identification problems in machine vision.
Loucks, Edward Philip.

A perspective systems approach to parameter identification problems in machine vision. - 135 p.

Source: Dissertation Abstracts International, Volume: 56-03, Section: B, page: 1680.

Thesis (D.Sc.)--Washington University, 1994.

The main topic of this dissertation is the identification of parameters describing the motion and shape of an object via observations from a single CCD camera. A solution to this problem is achieved using a perspective systems approach. The 3-D motion of the object produces an apparent motion of the intensity pattern on the image plane of the camera called the "optical flow." We find that the optical flow can be expressed as a dynamical system parameterized by certain "essential parameters." It is shown that estimates of these essential parameters can be calculated directly from image data. For images with smoothly varying intensity patterns, estimates are obtained from the image intensities. For images with sharply varying intensity patterns, we first identify and track features on the object such as corners or edges. The information gained from feature tracking is used to calculate estimates of the essential parameters. The essential parameters, in turn, are written as the perspective output of a system describing the dynamics of the planar surface, called the "shape dynamics." By "perspective output," we mean that the output is observed up to a homogeneous line. We call a linear system with such an output a "perspective system." The parameters of the shape dynamics are precisely the motion parameters we seek. Thus the motion identification problem becomes a parameter identification problem for a perspective system. A complete solution to this problem is obtained for the case of rigid body motion and for the previously unstudied case of affine motion. These motions are studied under both perspective and orthographic projection and in both continuous and discrete time settings. Previous methods for estimating motion have generally used information in local time only, i.e. optical flow data. In the discrete time setting, only a few frames (two or three) have been used. The most important contribution of this dissertation is the use of perspective systems that allows the inclusion of image data over longer intervals of time. It is shown that, in this way, it is possible to obtain improved estimates of the motion parameters than was previously reported. Furthermore, by using such an approach, techniques such as recursive filtering via nonlinear observers can be implemented that were not possible with previous methods.Subjects--Topical Terms:

1018128
Engineering, System Science.

A perspective systems approach to parameter identification problems in machine vision.
LDR:03238nmm 2200277 4500 001 1844745
005 20051017075510.5
008 130614s1994 eng d
035 $a (UnM)AAI9523630
035 $a AAI9523630
040 $a UnM $c UnM
100 1 $a Loucks, Edward Philip. $3 1932931
245 1 2 $a A perspective systems approach to parameter identification problems in machine vision.
300 $a 135 p.
500 $a Source: Dissertation Abstracts International, Volume: 56-03, Section: B, page: 1680.
500 $a Director: B. K. Ghosh.
502 $a Thesis (D.Sc.)--Washington University, 1994.
520 $a The main topic of this dissertation is the identification of parameters describing the motion and shape of an object via observations from a single CCD camera. A solution to this problem is achieved using a perspective systems approach. The 3-D motion of the object produces an apparent motion of the intensity pattern on the image plane of the camera called the "optical flow." We find that the optical flow can be expressed as a dynamical system parameterized by certain "essential parameters." It is shown that estimates of these essential parameters can be calculated directly from image data. For images with smoothly varying intensity patterns, estimates are obtained from the image intensities. For images with sharply varying intensity patterns, we first identify and track features on the object such as corners or edges. The information gained from feature tracking is used to calculate estimates of the essential parameters. The essential parameters, in turn, are written as the perspective output of a system describing the dynamics of the planar surface, called the "shape dynamics." By "perspective output," we mean that the output is observed up to a homogeneous line. We call a linear system with such an output a "perspective system." The parameters of the shape dynamics are precisely the motion parameters we seek. Thus the motion identification problem becomes a parameter identification problem for a perspective system. A complete solution to this problem is obtained for the case of rigid body motion and for the previously unstudied case of affine motion. These motions are studied under both perspective and orthographic projection and in both continuous and discrete time settings. Previous methods for estimating motion have generally used information in local time only, i.e. optical flow data. In the discrete time setting, only a few frames (two or three) have been used. The most important contribution of this dissertation is the use of perspective systems that allows the inclusion of image data over longer intervals of time. It is shown that, in this way, it is possible to obtain improved estimates of the motion parameters than was previously reported. Furthermore, by using such an approach, techniques such as recursive filtering via nonlinear observers can be implemented that were not possible with previous methods.
590 $a School code: 0252.
650 4 $a Engineering, System Science. $3 1018128
650 4 $a Mathematics. $3 515831
650 4 $a Computer Science. $3 626642
690 $a 0790
690 $a 0405
690 $a 0984
710 2 0 $a Washington University. $3 1250147
773 0 $t Dissertation Abstracts International $g 56-03B.
790 1 0 $a Ghosh, B. K., $e advisor
790 $a 0252
791 $a D.Sc.
792 $a 1994
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=9523630