지적재조사사업에서의 무인비행장치와 Mask R-CNN을 이용한 건축물 현황 취득 방안 연구

Abstract

In the cadastral re survey project, the current status of buildings is assessed during the survey of the cadastral re survey area. This results in loss of time and cost because it is necessary to additionally conduct a building status survey during the cadastral re survey. To reduce this, we applied the unmanned aerial vehicle image and the Mask R-CNN algorithm of deep learning to extract the current status of the building. The performance of the extracted object and the boundary point of the ground survey was compared. As a result of comparison, RMSE X = 0.072m, Y = 0.062m, a value applicable to the cadastral resurvey project was derived. However, the extraction rate of the building status was about 50%, and there were also buildings with low accuracy of the extracted objects. It is judged that the building was not clearly recognized due to the lack of training data. If the extraction rate is further increased due to the accumulation of data and the development of deep learning algorithms in the future, it is judged that it can be applied to cadastral fields such as cadastral resurvey project in the future.