Authors:
Yanan Miao
1
;
Huan Ma
1
;
Xiaoming Tao
2
and
Jia Cui
1
Affiliations:
1
National Computer Network Emergency Response Technical Team/Coordination Center of China, China
;
2
Tsinghua University, China
Keyword(s):
Landmarks Localization, 3D Shape Estimation, Pose Estimation, Cascaded Regression.
Related
Ontology
Subjects/Areas/Topics:
Applications
;
Computer Vision, Visualization and Computer Graphics
;
Geometry and Modeling
;
Image-Based Modeling
;
Pattern Recognition
;
Regression
;
Shape Representation
;
Software Engineering
;
Theory and Methods
Abstract:
Previous works on reconstruction of a three-dimensional (3D) point shape model commonly use a two-step
framework. Precisely localizing a series of feature points in an image is performed on the first step. Then the
second procedure attempts to fit the 3D data to the observations to get the real 3D shape. Such an approach
has high time consumption, and easily gets stuck into local minimum. To address this problem, we propose a
method to jointly estimate the global 3D geometric structure of car and localize 2D landmarks from a single
viewpoint image. First, we parametrizing the 3D shape by the coefficients of the linear combination of a set
of predefined shape bases. Second, we adopt a cascaded regression framework to regress the global shape
encoded by the prior bases, by jointly minimizing the appearance and shape fitting differences under a weak
projection camera model. The position fitting item can help cope with the description ambiguity of local
appearance, and provide m
ore information for 3D reconstruction. Experimental results on a multi-view car
dataset demonstrate favourable improvements on pose estimation and shape prediction, compared with some
previous methods.
(More)