The goal of this work is to provide tasks for the workshop 'Digital Image Processing', in the field of image segmentation. The object to work with is the board of a maze game, of which images are taken with an industry level camera system. The setup and required modifications of the task itself are discussed briefly. In order to obtain optical undistorted images, the required theory of mathematical modeling is explained and a method for the calibration is shown. For a possible use in a real-time system, attempts were made to speed up the process of geometric correction, which results in a significant performance gain.

Afterwards the clean image is segmented using a pixel based local threshold filter, which turned out to be a suitable solution. The resulting segments are analyzed for their properties as size, perimeter, symmetry and some others. An easy possibility to classify different types of objects is shown, so that it is possible to intuitively get information about the path, holes, start- and endpoint. An automated labeling of the order of the holes is done by masking and multiplying the image information, in combination with a simple path finding algorithm.

Keywords: digital image processing, labyrinth game, camera calibration toolbox

February 28, 2013