Shape Based Stereovision Assistance in Rehabilitation Robotics

A graphical user interface program was created along with shape models, which allow persons with disabilities to set up a stereovision system with off-theshelf hardware and detect objects of interest, which can be picked up using a sensor assisted telerobotic manipulator. A Hitachi KP-D50 CCD camera and an Imaging Source CCD camera were used along with two Imaging Source DFG/LC1 frame grabbers to set up a stereovision system. In order to use the stereovision system, the two main problems of correspondence and reconstruction are solved using subroutines of the program created for this work. The user interface allows the user to easily perform the intrinsic and extrinsic camera calibration required for stereovision, by following a few basic steps incorporated into the user interface program, which are described in this thesis. A calibration table required for these tasks can also be easily created using the program. In order to detect the object of interest, shape models, created by the user interface program, are used to solve the correspondence problem of stereovision. The correspondence problem is that of locating corresponding points in the left eye and the right eye, which are necessary to perform the calculations to obtain the location of the object of interest with respect to the endeffector. The shape models created for some commonly available items such as a doorknob and a door handle are included in the program and used to test the stereovision system. As expected, the error of detection decreases as the stereo system is moved closer to the object of interest in the x-, y- and z-position. The user interface created allows the user to detect the object of interest, approach it with a telerobotic manipulator using velocity-scaled assist functions, which scale the velocity of the telerobotic manipulator perpendicular to the desired movement, and finally grasp the object. The user can then move the object to any desired location