Building detection on lidar and multispectral images with graph-cut method

In today's world, determining the boundaries of buildings on earth is a field where remote sensing specialists work extensively because of social, economic and military significance. The automatic detection of the distribution of land and buildings from aerial photographs has become a field that has been challenged by many scientists in the last 25 years. The increase in interest in this area is due to abilities of achieving high resolution, low cost images quickly. Another factor that leads to increased work on this area is the widespread use of high-resolution LIDAR data. Due to the LIDAR height data, pixel groups belonging to objects other than land can easily be obtained due to the elevation information. In this study, new methods for building boundary detection are developed using Lidar data. The seed points for the building detection algorithm are obtained by morphological operations. By using these pixel groups, a map is obtained that contains the computed geodesic distance and the areas that are candidates for buildings. By using this map, classification of buildings and non-building areas are obtained by using the graph-cut method. It has been experimentally demonstrated that the performance of the proposed method is more successful than the other building methods applied today.