Automatic road detection from data taken by unmanned aerial vehicles

With the development of technology, data collection methods offer the opportunity to learn more about the world's topography. The density in the data obtained made it necessary to infer the meaningful ones in these data. Previously, the extraction of the requested details was done manually by the operators. This situation is considered an adverse reaction in terms of time and cost. Thanks to the developing detail extraction methods, automatic detection of the needed objects have become faster and easier. We need to use the spatial and spectral information of an object and its surroundings to obtain information from high-resolution images obtained by the Unmanned Aerial Vehicle (UAV). If pixel-based approaches are applied to obtain information from such remotely sensed data, only spectral information is used. Therefore, pixel-based approaches cannot meet high-resolution image classification, and information extraction is based only on gray-level thresholding methods, which requires the user to deal with big data. This situation makes it very difficult to extract sensitive data from urban areas containing precise boundaries in buildings, roads, parks, etc. in the pixel-based classification method applied in high-resolution images with rich information content. This shortcoming has brought along advanced image analysis and classification approaches depending on current computing technologies. Therefore, object-based classification approaches are being used extensively in today's image processing and remote sensing technologies. In this study, automatic detection of roads and similar lines with determined rules and processes are aimed by using an object-based classification method using orthophoto images with similar spatial resolution. In this thesis study, the data collected with UAV was evaluated by photogrammetric techniques and road determination was made by using an object-based classification method in Definiens eCognition software. With the object-based classification method; It has been shown that it is possible to remotely classify the perceived data according to different object characteristics such as spatial, spectral, and contextual information. Then, vector data drawn by the geodetic method were compared with the detected roads. Within the scope of the thesis, five different studies have been done using the images obtained in the visible and infrared region. The raw data obtained by UAV were evaluated and automatic path detection and classification were made in a digital environment with an object-based classification approach. As a result of the study, statistical analysis was done by using the paths determined by the object-oriented approach from the orthophoto image and the paths obtained by the current map and manual digitization. User accuracy ranges from 85% to 91% in five different studies. General accuracies were calculated as the lowest 0.7520 and the highest 0.8768. As a result of the comparison, it has been determined that the results obtained are within the confidence interval. This situation shows that the object-based classification method can contribute positively to the automatic determination of the road details when the spatial resolution increases with the spectral reflection value in the pictures.