Review on 2D and 3D Building Change Detection Methods Based on Remotely Sensed Data

Land is rapidly changing at the local, regional, national, and global scales, with a significant impact on the environment. Some changes occur due to natural causes, while other changes due to human projects such as urban growth.

This article provides an overview on the categorization of different methods used in detecting urban changes with emphasis on building complexities. Advances in facilitating the acquisition of three-dimensional data have led to three-dimensional change detection methods with two concepts of geometric comparison (including height difference calculation, Euclidean distance and transition-based methods) and geometric-spectral analysis (including correction). The purpose of this review is to answer the question of whether advances in change detecion methods and converting them from two-dimensional methods to three-dimensional ones have been able to meet the challenges in this context. What is future research to improve the results of 3D change detection methods?

According to the results of research on different types of change detection methods, although two-dimensional change detection methods have considerable variation, they lack altitude information and estimation of changes in the third dimension and in the face of high spatial and spectral resolution and three-dimensional effects such as buildings face challenges. Therefore, just by relying on the results of these methods, it is not possible to get a proper assessment of damages during accidents or construction estimations and so on.

In this article, while discussing the concepts presented in the three-dimensional methods of detecting changes, the strengths and weaknesses and challenges of the existing research are compared with the two-dimensional methods. It is concluded that in most cases, three-dimensional change detection methods rely heavily on two basic issues: 1) the use of advanced image-matching algorithms to produce three-dimensional data; 2) high-level machine learning techniques based on geometric and spectral data.