In factory automation cameras and image processing algorithms can be used to inspect objects. This can decrease the faulty objects that leave the factory and reduce manual labour needed. A vision sensor is a system where camera and image processing is delivered together, and that only needs to be configured for the application that it is to be used for. Thus no programming knowledge is needed for the customer. In this Master’s thesis a way to make the configuration of a vision sensor even easier is developed and evaluated.

The idea is that the customer knows his or her product much better than he or she knows image processing. The customer could take images of positive and negative samples of the object that is to be inspected. The algorithm should then, given these images, configure the vision sensor automatically.

The algorithm that is developed to solve this problem is described step by step with examples to illustrate the problems that needed to be solved. Much of the focus is on how to compare two configurations to each other, in order to find the best one. The resulting configuration from the algorithm is then evaluated with respect to types of applications, computation time and representativeness of the input images.