The information technology for automated recognition building

Abstract

The share of space images in the total quantity of Earth remote sensing data is increasing every year. The quality of the data received from the satellites is improving, the accuracy is increasing, the price is being reduced, and the productivity of the new satellites is significantly improving. Automatic building recognition helps save time and system resources for updating geoinformation databases and maintaining up-to-date municipal geodetic data. The analysis of modern methods of building recognition revealed the imperfection of the use of some methods and algorithms, the low number of information technologies for automatic recognition of buildings and the needs to develop a comprehensive approach to address this issue. This paper is devoted to the description of the developed information technology for the tasks of recognizing buildings on high resolution remote sensing images of the Earth and verifying the recognition results. Modern approaches to building recognition are analyzed. Automatic recognition of high resolution satellite images is not effective due to the high heterogeneity of spectral, textural and spatial characteristics. In addition to "preferred" objects (homes and roads), high-resolution images include objects that interfere with spatial recognition (such as trees, cars, and most importantly shadows). The proposed technology is based on five steps: splitting the image into plots with the elimination of vacant plots and determining the type of plot histogram analysis; feature segmentation in image space; validation based on size analysis, geometric and shadow analysis; finding and generalizing the outline of the building. Methods based on geometric analysis, shadow analysis and metadata usage have been developed to verify the recognition results. The result is a vector file that contains recognized polygonal objects. An analysis of commercial buildings has found that a complete run, which includes all stages, improves overall recognition accuracy and reduces manual post processing. Geometry analysis is a crucial step in the overall process with a significant impact on the end result. Size analysis and shadow analysis also play an important role and should be integrated into the process to reduce post-processing efforts.