| BUILDING OUTLINE EXTRACTION FROM AERIAL IMAGES USING CONVOLUTIONAL NEURAL NETWORKS |
| Paper ID : 1351-SMPR |
| Authors: |
|
Fatemeh Alidoost1, Hossein Arefi *2, Federico Tombari3 1School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Tehran, Iran 2University of Tehran 3Computer Aided Medical Procedures & Augmented Reality, Fakultat fur Informatik, Technische Universitat Munchen, Munich, Germany |
| Abstract: |
| Automatic detection and extraction of buildings from aerial images are considerable value in many applications, including disaster management, navigation, urbanization monitoring, emergency responses, 3D city mapping and modelling. However, the most important challenge is to localize buildings, precisely using monocular aerial images. In this paper, a Deep Learning (DL)-based approach is proposed to localize buildings, estimate the relative height information, and extract the buildings’ boundaries using single aerial images. In order to detect buildings and approximate localization, the ResNet architecture is trained for building and non-building object classification. We also introduce a novel Multi Scale Convolutional-Deconvolutional Network (MSCDN) including skip connections to predict the high resolution height values from a 2D single image. The extracted information including the 3D locations of buildings are then employed by an Active Contour Model (ACM) to provide precise boundaries of buildings. The experiments show that, even having noises in height values, the proposed method performs well on single aerial images with different complex background. The quality rate for building detection is about 86% and the RMSE of height prediction is about 4 m. Also, the accuracy of boundary extraction is about 68%. Moreover, since the whole procedure is based on single 2D images, the proposed method could be employed for real time applications. |
| Keywords: |
| Building Detection, Deep Learning, Active Contour Models, Selective Search, Depth Prediction |
| Status : Conditional Accept (Oral Presentation) |
