Abstract

This paper presents a flexible framework to build a target-specific, part-based representation for arbitraryarticulated or rigid objects. The aim is to successfully track the target object in 2D, through multiple scalesand occlusions. This is realized by employing a hierarchical, iterative optimization process on theproposed representation of structure and appearance. Therefore, each rigid part of an object is describedby a hierarchical spring system represented by an attributed graph pyramid. Hierarchical spring systemsencode the spatial relationships of the features (attributes of the graph pyramid) describing the parts andenforce them by spring-like behavior during tracking. Articulation points connecting the parts of theobject allow to transfer position information from reliable to ambiguous parts. Tracking is done in aniterative process by combining the hypotheses of simple trackers with the hypotheses extracted from thehierarchical spring systems.

Reference

Artner, N., Ion, A., & Kropatsch, W. (2010). Multi-scale 2D tracking of articulated objects using hierarchical spring systems. Pattern Recognition, 44(4), 800–810. http://hdl.handle.net/20.500.12708/167563