CNN-based statistics and location estimation of missing components in routine inspection of historic buildings

Abstract

Abstract Ancient buildings have a high cultural and historical value. In the process of their protection and maintenance, it is crucial to conduct regular routine inspections on them. During the regular routine inspections, the identification and statistics for components on historic buildings are of great significance to conservators, managers and visitors. However, the current identification and statistics work is almost carried out by human eyes, which is time consuming and labor intensive. Actually, this work can be done by artificial intelligence. In order to promote the intelligent development of routine inspections of historic buildings, this paper proposes a methodology in the case of the Forbidden City to identify and count the numbers of intact and impaired components based on Convolutional Neural Network. The applied algorithm is Faster R-CNN, which is an effective object detection algorithm for 2D images. In addition, the positions of the missing components can be inferred and marked in the images as their regularity of the positional arrangement. This methodology can lay the foundation for the subsequent intelligent inspection system of the historic buildings.