Abstract

One of the most challenging problems in rehabilitation robotics is the design of an efficient   Man-Machine interface 
allowing the user with disabilities considerable freedom and flexibility. The newly proposed multi-modal user direc
tion approach combining command and control methods is a very promising way to achieve this goal. This design is 
motivated by the idea of minimizing the user's burden of operating a robot manipulator while utilizing user's intelli
gence and available mobilities with his/her head. With this design, the user with physical disabilities simply uses ges
ture (pointing) to indicate the location or the desired object; and uses voice to activate the system. The rest of the 
manipulation, for instance, picking an object interested and placing it where the user wants, will be handled by the 
robot system automatically. It can be seen that one of the major tasks for this system is to gather the information that 
describes the interested object, i.e. size, height, orientation, etc. as well as the destination location the user points. 
Since the system is designed to be used in an unstructured environment, an active three dimensional information 
acquisition method is required. This report presents the development of a stereo vision interface that provides the 
required 3D information for the robot system in real-time. Topics discussed are camera calibration, the design of a 
simple yet effective line segment pair-match scheme of lines in two images, 3D recovery and segmentation, and object 
orientation determination. Steps of operating this vision interface are provided. An experimental example of using 
this interface (picking and placing a handbook) is presented. Also, the source code of the C programs used to run this 
system is listed in the appendix for interested readers.