Experimental setting:
The following results aims to provide a qualitative evaluation of the performance of our proposed method. The left panel reports the set of detection provided by the face detector (in blue), the middle panel the tracking results (in green) produced with color histogram-based appearance model, and on the right the results produced by our proposed method.
First experiment:
This experiment is designed to reproduce the context of a casual discussion among 4 people, with the robot among them. Due to the limited field of view of the robot, it needs to move the camera to follow the main speaker (detected by using a sound source localization method).
Second experiment:
This experimental setting aims to evaluate specifically the ability of the tracking algorithm to handle identity switches in the context of mutual occlusions. In this context, the camera is fixed, and 4 people are moving in circle front of the robot to maximize the number of self-occlusions.
