VECTOR MAP GENERATION from AERIAL IMAGERY USING DEEP LEARNING

Abstract

We propose a simple yet efficient technique to leverage semantic segmentation model to extract and separate individual buildings in densely compacted areas using medium resolution satellite/UAV orthoimages. We adopted standard UNET architecture, additionally added batch normalization layer after every convolution, to label every pixel in the image. The result obtained is fed into proposed post-processing pipeline for separating connected binary blobs of buildings and converting it into GIS layer for further analysis as well as for generating 3D buildings. The proposed algorithm extracts building footprints from aerial images, transform semantic to instance map and convert it into GIS layers to generate 3D buildings. We integrated this method in Indshine's cloud platform to speed up the process of digitization, generate automatic 3D models, and perform the geospatial analysis. Our network achieved ∼70% Dice coefficient for the segmentation process.