Image based automatic road detection and modeling

With the developing technology, individuals have become users who are not only consuming but also producing information. This is particularly the main source of information for many open source projects. OpenStreetMap (OSM) and OpenAerialMap (OAM) are very common projects created with this perspective. OSM is a project that has emerged to produce a free world map. OAM is a platform for storing and sharing high resolution satellite and aerial imagery of the world. On the other hand, significant developments and progress have been made in machine learning and artificial neural network (ANN) areas, especially with the rapid developing and cheaper prices of computer technologies after the 2000s. The concept of classical programming, especially in deep learning and convolutional neural networks, has completely changed. One of the most important factors affecting the success of machine learning and deep neural networks is the provision of sufficient data to be used as input for training the system. In this study, a data set for supervised classification was created by using high resolution aerial images obtained from OAM and vector information obtained from OSM. This data set was used to train and test the model for the detection of road networks using semantic segmentation technique. Different models such as customized fully convolutional network (FCN), SegNet and PSPNet were used in the study and the results obtained are given comparatively. According to the evaluation results, the most successful results were obtained with the FCN. The pixel accuracy of this model is 0.7102, the recall value is 0.7826, the precision value is 0.7949 and the F1 score is 0.7887. Keywords: Road network detection, Convolutional neural networks, Semantic Segmentation, Openstreetmap, Openaerialmap