Automatic 3D Building Footprint Extraction from High Resolution Satellite Image using OSM

Abstract

Building impression data is fundamental for 3D building demonstrating. Customarily, in remote detecting, building impressions are removed and outlined from ethereal symbolism and LiDAR point cloud. Adopting an alternate strategy, this paper is devoted to the advancement of OpenStreetMap (OSM) building impressions misusing the shape data, which is gotten from profound learning-based semantic division of angled pictures procured by the Unmanned Aerial Vehicle (UAV). Initial, an improved 3D building model of Level of Detail 1 (LoD 1) is instated utilizing the impression data from OSM and the height data from Digital Surface Model (DSM). In parallel, a profound neural system for pixel-wise semantic picture division is prepared so as to extricate the structure limits as shape proof. Thusly, a streamlining incorporating the shape proof from multi-see pictures as a requirement brings about a refined 3D building model with improved impressions and stature. This technique is utilized to advance OSM building impressions for four datasets with various structure types, exhibiting hearty execution for both individual structures and different structures paying little respect to picture goals. At long last, the contrast the outcome and reference information from German Authority Topographic-Cartographic Information System (ATKIS). Quantitative and subjective assessments uncover that the first OSM building impressions have enormous balanced, yet can be fundamentally improved from meter level to decimeter level after enhancement.