Інтелектуальна технологія побудови плану будівель за аерофотозйомкою їхніх дахів

Abstract

The article is devoted to the development of an intelligent technology for the construction of building plans based on the data of remote sensing of the Earth. Such data can be the data of aerial photographs or high-quality satellite images. A detailed review of the types of roofs was carried out by analyzing the most common classifications, and the most typical of them were determined, on the example of which, the improvement of the intelligent technology of building plan construction based on aerial photography data can be carried out. Modern and traditional methods of image analysis that can be used to solve this problem are characterized. The methods are chosen, which are the most advanced and can be effective for this class of tasks. Generalized algorithm for the class of single-pitched, gable, flat and hip roofs has been developed. It is proposed to improve the intelligent technology of constructing a plan of buildings based on the data of aerial photography of their roofs, by integrating the DETR detection model (“DEtection TRansformer”) together with the segmentation based on ViTs (Vision Transformers) for a comprehensive solution to the problems of finding and identifying roofs, in a first approximation, for further improving the construction of building plans. Combined approach is proposed that takes advantage of the strengths of both models by using the DETR model to localize groups of roofs in large-scale images, then using ViTs to accurately segment similar types of roofs. Comparison of the accuracy of models for image segmentation and object detection in images was made. The results of the approbation of the improved technology of construction of the plan of buildings based on the data of aerial photography of their roofs are characterized: the algorithm, approaches and software on the test data of aerial photography of the public dataset, which proved their effectiveness. Possible improvements of the proposed technology due to the use of pseudo masks are suggested. The results of the work can be extended to other types of building roofs, on the condition of proper adaptation in accordance with the characteristic features of specific types of roofs.