Abstract
This thesis was conducted at the Fraunhofer Chalmers Centre for Industrial Mathematics in collaboration with the Fraunhofer-Institut für Techno- und Wirtschaftsmathematik. The aim of this thesis is to develop an imaging system for the automated detection of holes in images of supermarket shelves. The proposed approach uses an unsupervised segmentation method to presegment the image into homogeneous regions. Each of those image regions is then classified separately using a support vector machine. Finally, suitable bounding boxes are found for image regions that are likely to represent holes. Apart from the SVM classifier also an AdaBoost classifier and a structural classifier based on conditional random fields are implemented and tested. This thesis describes the implementation and performance characteristics of the resulting imaging system, which is implemented using the ToolIP graphical image processing framework and C++.