HONS 03/07
Tracking Object Trajectories Relative to Planar Surfaces Using Stereo
Matthew Elliot
Department of Computer Science
University of Canterbury
Abstract
This project proposes a methodology for 3D tracking of objects in relation to a planar surface,
with trajectory accuracy enhanced using applied statistical analysis. Planar surface extraction,
with camera position and orientation invariance, is achieved by finding limiting regions
established by graph-based segmentation and mapping the resulting segments to disparity data
from a stereo camera. Secondly, object detection and tracking is performed using a
combination of adaptive background subtraction and least squares linear regression for
calculating object trajectories. The accuracy of bounding planar surface extraction is shown to
be accurate to within 1.4% and tracking has shown similar high correlations between the
calculated and actual positions.