next up previous
Next: Experiment 1. Up: icra02HTML Previous: Action Fusion Approach


Experimental Evaluation

A series of tracking experiments has been performed to experimentally evaluate the proposed approaches. The sequences are obtained in an everyday environment with changing lighting conditions and background clutter. The original size of the images was $ 320\times240$ and a standard CCD camera with 6mm focal length was used. Tracking runs at frame-rate on a PC 333MHz Pentium using a standard Matrox Meteor frame-grabber card. The main objectives of the experiments were: i) evaluation of fusion approaches with respect to the weighting methods, and ii) performance of the fusion approaches versus the performance of each cue.

The results are discussed through accuracy and reliability. The accuracy is expressed using an error measure which is a distance between the ground truth (chosen manually using a reference point on the object) and the currently estimated position of the reference point. The results are summarized through the mean square error and standard deviation in pixels. The measure of the reliability is on a yes/no basis depending on if a cue (or the fused system) successfully tracks the target during a single experiment. The tracking is successful if the object is kept inside the window of attention during the entire test sequence.



Subsections
next up previous
Next: Experiment 1. Up: icra02HTML Previous: Action Fusion Approach
Danica Kragic 2002-12-06