Developing a Novel Method for Fast and Automatic Detection of Damaged Buildings from Post-Earthquake Images and Point Cloud (Case Study: Sarpol-e-Zahab)

In the recent three decades, building damage map production after an earthquake has been one of the most important challenges in disaster management. For decades, researchers have utilized different kinds of sensors, including airborne LiDAR, RADAR, and optical satellite imagery, in order to create earthquake-induced damage maps. LiDAR and RADAR sensors are disadvantageous due to their high cost and difficulty of acquisition. Additionally, optical satellite images are limited by the lack of timely availability of high-resolution imagery, a low signal-to-noise ratio, and the inability to reconstruct 3D data. On the other hand, the recent development of low-cost and flexible-resolution UAVs along with high accessibility has opened up a great opportunity in this area. Furthermore, UAV data make detecting damage levels possible according to the well-known EMS-98 damage standard. However, there are still a few researchers that have considered international damage standards. When developing a strategy for detecting damage, three important factors should be considered: accuracy, efficiency, and cost. Recently, a lot of attention has been paid to deep-learning models in the field of damage detection. These algorithms require enormous amounts of data, which takes a lot of time and energy. Moreover, such models are unlikely to be general enough due to the sophisticated and numerous types of damage that can occur to a building. So, these issues pose a serious limitation for their practical application in a disaster situation. Since this research put forward a novel and rapid method for detecting the damage to buildings on four levels using a designed removal decision tree. For testing the proposed method, orthophoto and DSM were derived from raw and post-event UAV images of two regions of Sarpol-e-Zahab. An overview of the presented framework for generating the damage map of buildings is depicted in figure1. In summary, our original contributions are described as follows:Taking into account three crucial parameters of crisis management including accuracy, cost, and time efficiency.Presenting a height anomaly index that distinguishes minor damage from major damage. In short, this index represents the percent of negative values in clustered DMS based on DBSCAN.Damage level detection according to EMS-98 standard in four levels, including, “no visible damage to slight damaged”, “Minor damaged”, ”major damaged”, and “collapsed”.Fig.1. The proposed method for detecting damage to buildings in 4 levelsThe results for two study areas of "Sarpol-e Zahab" reveals achieving an accuracy of 86% and a kappa coefficient of 81% for processing 142 buildings within just 15 seconds which indicates the efficiency and time efficiency of the proposed method. In addition, Because of being fully automatic (being unsupervised), the cost of this method is very low. As a consequence, our findings in this research revealed the high potential of UAV data for timely damage detection and rescue after an earthquake. However, the negative effects of shadows and trees are some of the challenges that should be considered in future research.