Integrated vision framework for a robotics research and development platform

Abstract

This thesis presents the design of a vision framework integrated into a robotics research and development platform. The vision system was implemented as part of the software platform developed by the Personal Robots Group at the MIT Media Lab. Featuring representations for images and camera sensors, this system provides a structure that is used to build robot vision applications. One application shows how to merge the representations of two different cameras in order to create a camera entity that provides images with fused depth-color data. The system also allows the integration of computer vision algorithms that can be used to extract perceptual information from the robot's surroundings. Two more applications show detection and tracking of human face and body pose using depth-color images.