Digital Terrain Modeling Using AKAZE Features Derived from UAV-Acquired, Nadir and Oblique Images

Hyojoo Son; Hyunsoo Kim; Changwan Kim

Abstract:

During construction, digital terrain models based on images from unmanned aerial vehicles (UAVs) are useful, as they rapidly provide data for objective volume calculations that can be used to monitor in-progress earthworks. As many curved objects are present on construction sites and as construction materials and equipment can provide partial obstructions, nadir images alone is not adequate to generate accurate digital terrain models during the earthwork phase, thus, it is necessary to acquire both nadir and oblique images. However, it is difficult to extract features from oblique images to determine accurate locations using the traditional method from photogrammetry software, the scale-invariant feature transform (SIFT) algorithm. This study proposes a method for generating accurate digital terrain models of construction sites based on the accelerated KAZE (AKAZE) algorithm using a combination of nadir and oblique UAV images. The proposed method consists of the following steps: 1) feature extraction and matching based on the AKAZE algorithm; 2) three-dimensional (3D) point cloud generation based on the results of the feature matching; and 3) digital terrain model generation based on the resulting 3D point cloud and on the GPS information for the corresponding locations of each ground control point (GCP). Validation of the proposed method involves 100 oblique UAV images of the actual construction sites in-progress earthworks. The experimental results indicate that the AKAZE algorithm can be applied to generate an accurate digital terrain model by extracting and matching features to ensure the accuracy of the edges and corners. Based on these results, the proposed method can be expected to provide accurate volume calculations from a generated digital terrain model, which can enable monitoring of the earthwork phase.