Registration of objects in FRIEND environment using Disparity Map Computation (Abstract)

This is my bachelor's thesis which I did at Institute of Automation, Universitaet Bremen. All the reference material is acknowledged back IAT, Uni-Bremen.

FRIEND (Functional Robot with dexterous arm and user-frIENdly interface for Disabled people) system employs a pair of stereo cameras (Bumblebee- Point grey research) and currently uses stereo triangulation algorithm to do the 3-D reconstruction of the surroundings. This method uses the Disparity Map for the 3-D reconstruction. Disparity Map gives the polar distance of a feature point pertaining to a specific object in the scenario. Construction of Disparity Map is based on an algorithm by Birchfield and Tomasi.

We use other parameters of the Bumblebee camera system such as ‘Focal length’ and ‘Field angle’ along with the position of feature point in the image to reconstruct in 3-D. Position of feature point within an image is estimated using feature segmentation. Using the inverse sine of the ratio of horizontal and vertical separation of the feature point from the centre of an image to that of focal length, we get the tilt angle and span angle of feature point with respect to the centre of an image. Physical tilt angle of camera can be measured using sensors in Bumblebee. Combining these two parameters we can get three co-ordinates of a feature point in spherical polar system namely Distance (r), Tilt angle (θ) and Span angle (φ). This can be then converted to Cartesian co-ordinate system to give the values of X, Y and Z (Camera co-ordinate system)

The attachments to this page are my actual thesis report, and the paper based on the same which was presented at ICARUS, 2010 conference at IIT Kanpur in India.