MONOCULAR DEPTH ESTIMATION OF GOOGLE EARTH IMAGES USING CONVOLUTIONAL NEURAL NETWORKS

Abstract. Depth estimation from images is an important task using scene understanding and reconstruction. Recently, encoder-decoder type fully convolutional architectures have gained great success in the area of depth estimation. extraction aerial satellite one topics photogrammetry remote sensing. This usually done image pairs, or more than two images. Solving this problem a single still challenging has not been completely solved. Several neural networks proposed to extract image, which act as encoders decoders. In article, we use these networks, performed well for estimation, order height Our main goal investigate performance Google Earth data produce digital elevation model. At first, extracted model target ISPRS benchmark data, then did same thing The paper presents network computing high-resolution map given RGB image. results show proper extraction. We achieved values 2.07 m 0.36 RMS REL metrics, respectively, are very comparable acceptable 2.04 0.39 obtained