It is the aim of this thesis to investigate the possibilities to digitally record pavement characteristics of highways in Germany with the support of machine learning methods. Furthermore, a model idea will be presented. This model idea shall enable the usage of various data streams that are generated as part of ZEB campaigns for a variety of tasks in the pavement maintenance management. As there has been hardly any research done in the area of automated analysis based on ZEB data, this thesis will prepare the foundation for this approach.

Therefore, an application has been programmed that will enable to mark characteristics on surface pictures of the ZEB and to save them in a database. Based on this, different Convolutional Neural Networks were developed using this data. A result has been that pavement features, that are very similar, e.g. sealed cracks and patches cannot easily be classified on pictures with a 128x128 pixel size format. Therefore, it was investigated whether a combination of different networks with different input will lead to in an improvement of the results. Hence, an additional network was developed to depict a bigger area of the pavement surface. The outcome was that this approach improves the results slightly. However, it will need to be investigated in additional thesis and projects, if there are other possibilities to improve the detection of pavement characteristics.

In addition, this thesis showed that a bigger set of data will be needed due to the high variation of pavement surfaces.