Motion Segmentation Using Differential Geometry of Curves and Edges

Jain, Vishal

Full Metadata

Overview

Year:: 2009
Contributor:: Jain, Vishal (creator); Kimia, Benjamin (director); Taubin, Gabriel (reader); Cooper, David (reader); Brown University. Division of Engineering. Electrical Sciences and Computer Engineering (sponsor)
Genre:: theses
Subject:: Differential Geometry; flow; edges; affine motion; Curves; Geometry, Differential
Extent:: xiii, 147 p.
DOI: https://doi.org/10.7301/Z0FQ9V20

Files

Description

Abstract:: This thesis proposes to solve the problem is of segmenting independently moving objects which is specifically useful in applications for compression, initialization for tracking, input for recognition of the target and others. The existing approaches which can be broadly classified into feature-based and dense image flow-based approaches. A central tenant of this thesis is to propose that curves act as complimentary representation in regard to features and pixels as most of the features (mainly corners) lie on the curves and a pair of curves do bound the regions of pixels. The approach proposed comprises of two stages. The first stage is to obtain segmentation from a single/two views and two different approaches are discussed for case of stationary camera and moving camera. In case of moving camera, the correspondence of the curves is extracted from two views. The alignment for the correct pair of corresponding curve is used to fit a 2D parametric model such as affine and the curves with similar parametric models are grouped to obtain segmentation. In case of stationary camera, the position and orientation of sub-pixel edges of the background are modeled and any deviation from this model is detected as foreground. This modeling of edges compared to traditional intensity based approaches is more robust to sudden changes in illumination and requires fewer frames to build the model. The second stage of this approach enriches the impoverished segmentations obtained from first stage by using multiple frame consistency. First, edges of the segmentations from different frames (5 or 7) are aligned on a central frame using a Thin Plate Spline model. After the edge maps have been aligned, the edges which fit a smooth local curve supported by edges from different frames are kept and the rest are discarded leading to removal of noise and filling in the gaps even in case of significant occlusions. Finally, this work also gives a theoretical analysis to estimate 3D translation direction which shows infeasibility to obtain reasonable estimate. However, some aspect of 3D translation which is ratio of translation parallel to the image can be recovered.
Notes:: Thesis (Ph.D.) -- Brown University (2009)

Content

Access Conditions

Rights: In Copyright
Restrictions on Use: Collection is open for research.

Citation

Jain, Vishal, "Motion Segmentation Using Differential Geometry of Curves and Edges" (2009). Electrical Sciences and Computer Engineering Theses and Dissertations, Engineering Theses and Dissertations. Brown Digital Repository. Brown University Library. https://doi.org/10.7301/Z0FQ9V20

Relations

Collections:

Electrical Sciences and Computer Engineering Theses and Dissertations

Theses and Dissertations for the Electrical Sciences and Computer Engineering department.
...
Engineering Theses and Dissertations

Theses and Dissertations for the Engineering department.
...