SEMIAUTOMATIC 3D MAPPING OF BUILDINGS FROM 
MEDIUM SCALE (1:40,000) AERIAL PHOTOGRAPHS 
Yair Avrahami* *, Yuri Raizman ^, Yerach Doytsher“ 
Department of Transportation and Geo-Information Engineering, Faculty of Civil and 
Environmental Engineering, Technion — Israel Institute of Technology. 
Technion City, Haifa 32000, Israel. (yaira, doytsher)@tx.technion.ac.il 
8 Survey of Israel, 1 Lincoln St., Tel-Aviv, 65220 Israel .yurirg@mapi.gov.il 
ThS12 
KEY WORDS: digital photogrammetry, semi-automation, building extraction, mapping, small scale, aerial images, GIS 
ABSTRACT: 
In the last decade, much of the research dealing with 3D building extraction from aerial photographs has been carried out by the 
photogrammetry and computer vision communities. The increased usage of 3D City Spatial Information Systems and National GISs 
for control, management and planning, necessitated development of fully or semi-automatic methods for establishing and updating 
these systems. Most research tries to reconstruct the building space from large (~1:4,000) scale photographs, mainly for establishing 
or updating the 3D city model systems. However, the research presented in this paper focuses on 3D mapping from medium scale 
(~1:40,000) aerial photographs, specifically for establishing and updating the building layer in nationwide GIS databases. The 
algorithm developed for semi-automatic building mapping is based on a 2D approach to solving the 3D reality. The algorithm 
consists of five consecutive stages: pre-processing, left image operations, height extraction, right image operations and final 
mapping of the buildings. The first stage is the only stage performed manually in order to achieve specific goals: model solution, 
photograph preparation and designating the desired building roof. From the second stage onwards, the process is fully automatic. 
This algorithm can be employed in two ways, either as part of a fully automatic mapping of all buildings in the overlapping arca, or 
as stand alone, enabling a new technology for semiautomatic mapping within a non-stereoscopic environment, without using 3D 
spectacles. Based on the algorithm, a system for semi-automatic mapping of buildings was developed in order to test its efficiency 
and accuracy. The results are satisfactory with an accuracy of 0.5m for planimetric measurement and 1m for altimetric measurement. 
1. INTRODUCTION the building contours simultaneously in both images and 
finally, 3D mapping of the building. Implementation of this 
1.1 General Background algorithm enables mapping a building, while relying on a 2D 
building vote. This process can be referred to in two ways: 
either as part of a fully automatic mapping of all the buildings 
in the overlapping area, or as a separate part that facilitates a 
new technology for semiautomatic mapping within a non- 
stereoscopic environment and without using 3D spectacles. 
Based on this algorithm, a system for semiautomatic mapping 
of buildings was developed in order to test its efficiency. The 
experiments were conducted on two residential buildings areas 
in Tel-Aviv using medium scale images (~1:40,000) scanned at 
a pixel size of 14 gm . This article presents the algorithm, the 
When a human eye looks at an aerial image it is usually easy to 
detect what is the building and location of its corners. This 
ability consists of the reticular reception stage and information 
processing by the brain. In automatic mapping procedures, 
human vision is replaced by the computer. There is an analogy 
between human vision and computer vision. The reception stage 
parallels the scanning and saving of data in the computer 
memory. The information processing stage parallels a set of 
topology, geometry and radiometry rules, which enable the 
computer to detect and extract the building. In 3D mapping, 
these rules are required to simultaneously detect and extract 
from both images. Formulating these rules is a complex 
procedure. This complexity stems from the difficulty in 
formulating stable rules that remain valid for a large variety of 
images taken under diverse conditions (different orientation, 
camera, time photographed, season, etc) and suitable for all 
buildings. In order to surmount this difficulty, it is possible to 
divide the problem into two independent sequential stages. The 
first includes an automatic process to find votes (pointers) for 
the buildings in a single image. The second includes an 
automatic process to extract the 3D contour building for every 
vote. The current research focuses on developing a detailed 
algorithm for the second process and deals with three issues: 
Identifying the contour building in a single image, extracting 
experiments and the results. 
1.2 Related Works 
Nowadays 3D mapping of buildings is carried out manually 
employing a digital photogrammetric workstation (DPW) or an 
analytical stereoplotter. The advantage of the DPW 
environment is the ability to develop automation for 
photogrammetric assignments. However, full automation of 
object space mapping is still far from being implementable. 
Various methods are available in the scientific community, for 
mapping object space at different automation levels. These 
methods are limited and designed for specific mapping 
categories (i.e., specific areas, scales, objects etc.). In the last from aerial images executed by the photogrammetry and 
decade, much of the research dealt with 3D building extraction computer vision communities. The increased use of 3D City 
774 
r^ 
es er a ~~ qe e—— "rm Es — — 
beo i 
ee m ii rmt