Pose Estimation of Randomly Organised Stator Housings with Circular Features

This project deals with the development of a computer vision system for handling an instance of the bin picking problem. The system performs pose estimation of randomly organised stator housings containing circular features with a known CAD model. The implemented system obtains 3-D information through a binocular stereo setup. The individual stereo images are processed by a Canny edge detector followed by an edge segmentation and a least squares ellipse detection. The detected ellipses in each image are subject to a monocular pose estimation technique estimating two possible circular pose candidates for each ellipse. The correct pose candidates are determined by performing a matching between the circle pose candidates in each stereo image. The matching is done via an association graph search locating the maximally weighted clique through the use of simulated annealing. The final pose candidates are verified by back projection in a distance transformed edge image. System tests justified the choice of the circular feature due to the presence of detectable stator housings per bin configuration. Tests on simulated data showed acceptable error measures both for position and orientation estimation. The final conclusion was that the implemented method could form a useful part, handling certain pose situations, in a final bin picking system.