Self-supervised monocular depth estimation from oblique UAV videos

Unmanned Aerial Vehicles (UAVs) have become an essential photogrammetric measurement as they are affordable, easily accessible and versatile. images captured from UAVs applications in small large scale texture mapping, 3D modelling, object detection tasks, Digital Terrain Model (DTM) Surface (DSM) generation etc. Photogrammetric techniques routinely used for reconstruction UAV where multiple of the same scene acquired. Developments computer vision deep learning made Single Image Depth Estimation (SIDE) a field intense research. Using SIDE on can overcome need reconstruction. This paper aims to estimate depth single aerial image using learning. We follow self-supervised approach, Self-Supervised Monocular (SMDE), which does not ground truth or any extra information other than depth. video frames training model learns pose jointly through two different networks, one each pose. The predicted reconstruct viewpoint another utilising temporal videos. propose novel architecture with 2D Convolutional Neural Network (CNN) encoders CNN decoder extracting consecutive frames. A contrastive loss term is introduced improving quality generation. Our experiments carried out public UAVid dataset. experimental results demonstrate that our outperforms state-of-the-art methods estimating depths.