Title :
Evaluation, evolution, and optimal manipulation of uncertainties in computer vision
Author :
Zhou, Ling-Xiang ; Gu, Wei-Kang
Author_Institution :
Dept. of Electr. Eng., Zhejiang Univ., Hangzhou, China
Abstract :
Spatial quantization error plays an important role in computer vision. In order to manipulate image information optimally, the vision system must quantitatively take the measurement uncertainties into account. In this paper a normal probability error model of point features in a general motion vision system is presented, where the number of cameras, the camera attitude and the initial pixel localization errors are all arbitrary. The paper provides a novel representation, which describes uncertainties of the feature coordinates as well as the inverse of the covariance matrix. 2D image point uncertainties can also be represented by an extension of this unified 3D form. Explicit formulas and experimental results for both error generation and evolution are presented
Keywords :
cameras; computer vision; covariance matrices; errors; matrix inversion; motion estimation; probability; uncertainty handling; 2D image point uncertainties; 3D form; camera attitude; cameras; computer vision; covariance matrix inverse; error evolution; error generation; feature coordinates; image information; measurement uncertainty; motion vision system; normal probability error model; pixel localization errors; spatial quantization error; Cameras; Computer errors; Computer vision; Covariance matrix; Image reconstruction; Layout; Machine vision; Measurement uncertainty; Quantization; Sensor systems; Stereo vision;
Conference_Titel :
Intelligent Processing Systems, 1997. ICIPS '97. 1997 IEEE International Conference on
Conference_Location :
Beijing
Print_ISBN :
0-7803-4253-4
DOI :
10.1109/ICIPS.1997.669266
