Ground Object Recognition from Aerial Image-based 3D Point Cloud

Abstract

Recently, several attempts have been made to grasp 3D ground shape from a 3D point cloud generated by aerial vehicles, which help to fast situation recognition. For example, in case of earthquake disasters, we may detect building collapse and inclination by comparing the height of buildings in 3D models before/after the disasters. However, identifying such objects on the ground like buildings, vehicles and trees from a 3D point cloud, which consists of 3D coordinates and color information, is not straightforward due to the gap between the low-level point information (coordinates) and high level context information (objects). In this paper, we propose a ground object recognition method from a 3D point cloud that captures the heights of ground surface. Basically, we rely on some existing tools to generate such a 3D point cloud from aerial images, and our method tries to give semantics to each set of clustered points. In the proposed method, firstly, such points that correspond to the ground surface are eliminated using the elevation data from Geographical Survey Institute. Next, we apply an inter-point distance-based clustering and noise filtering method according to the point density of each cluster. Then such clusters that share some regions are merged to correctly identify a point cluster that corresponds to a single object. Finally, a filtering method is applied based on the knowledge on the sizes of objects. We have evaluated our method in several experiments conducted in real fields. We have confirmed that our method can remove the ground surface within 5% error, and can recognize most of the objects.