Object Tracking with Ratio Cycles Using Shape and Appearance Cues

M.E. Sargin, P. Ghosh, B.S. Manjunath and K. Rose
Department of Electrical and Computer Engineering, UC Santa Barbara
{msargin,pratim,manj,rose} [at] ece.ucsb.edu

Abstract

We present a method for object tracking over time sequence imagery. The image plane is represented with a 4-connected planar graph where vertices are associated with pixels. On each image, the outer contour of the object is localized by finding the optimal cycle in the graph such that a cost function based on temporal, appearance and shape priors is minimized. Our contribution is the particle filtering-based framework to integrate the shape cue with the temporal and appearance cues. We demonstrate that incorporating the shape prior yields promising performance improvement over temporal and appearance priors on various object tracking scenarios.
[PDF] [BibTex]
M.E. Sargin, P. Ghosh, B. S. Manjunath and K. Rose,
IEEE International Conference on Pattern Recognition, pp. 3649-3652, Aug. 2010.
Node ID: 544 , DB ID: 352 , Lab: VRL , Target: Conference
Subject: [Image and Video Segmentation] « Look up more