MONOPLOTTING - A SEMI-AUTOMATED APPROACH FOR 
3D RECONSTRUCTION FROM SINGLE SATELLITE IMAGE 
Xiaojing HUANG , Leong Keong KWOH 
Centre for Remote Imaging, Sensing and Processing (CRISP)National University of Singapore, Singapore 119260 
-(crshxj)@nus.edu.sg, -(crsklk)@nus.edu.sg 
Commission III, ThS-7 
KEY WORDS: Modelling, Reconstruction, Triangulation, Visualization, Rendering, Semi-automation 
ABSTRACT: 
In this paper, we present a semi-automated approach to high quality three-dimensional (3D) objects reconstruction from single high- 
resolution satellite image using monoplotting technique. The method requires high accuracy of satellite sensor model, such as 
Rational Polynomial Coefficients (RPC), and high resolution of Digital Elevation Models (DEM) to achieve the high precision 
geometry of the 3D reconstructed object. The RPC is used to determine the ray of the satellite to the object, while the azimuth and 
elevation angles of the Sun is used to determine the shadow of the object on the ground. The vertices of the polygon, which are 
manually extracted from the top surface of an object, are used for the prediction of the base and shadow positions on the 2D satellite 
image based on the height of the object. Once the predicted base and shadow positions match their positions in the image by 
observation, the height of the object is determined. As a consequence, the 3D object is reconstructed simultaneously. We have 
developed a full suite of the software for realtime extracting, editing, reconstmcting and visualising of the 3D objects from single 
IKONOS image. 
1. INTRODUCTION 
The high accuracy Rational Polynomial Coefficients (RPC) 
sensor models and stereoscopic images of IKONOS satellite 
allow us to extract 3D spatial information and build digital city 
models with high geometric accuracy. However, the demand for 
reconstructing the city model from single satellite image is 
increasing due to the data availability, weather condition, 
timing and cost. Monoplotting, a photogrammetric technique, is 
able to extract 3D spatial information from single image with 
the availability of high resolution digital elevation models 
(DEM). 
For the flat top of a 3D object, its position in a 2D image can be 
derived with the RPC, which defines the ray from satellite to 
the object. The ray intersects the ground and the height at this 
intersection point is given by the DEM. The object can be re 
positioned to its 3D coordinate along the ray direction with its 
base landed on the ground. Once the height of the object is 
measured, its 3D geometric position is unique. 
There are a few methods to derive the height of the building: (1) 
from satellite azimuth and elevation angles; (2) from sun 
azimuth and elevation angles; and (3) from RPC. Satellite 
angles show the building top relative to its base, while the sun 
angles show the shadow of the building top relative to its base. 
By measuring the distance from top to base or from top to 
shadow in image space (pixels), the height of the building can 
be obtained in ground space (meters). However, the accuracy of 
the obtained height depends on the accuracy, size or the 
difference of the satellite and the Sun angles, and these values 
given in the image auxiliary data are not always very accurate 
for the whole strip of scene. RPC on the other hand, generates 
the building height with higher accuracy by measuring from top 
to base, but loses its accuracy when the base is not clearly 
shown in the image. 
To achieve higher accuracy for the height extraction, we design 
an interactive platform for the user by combining the above 
methods. Instead of the conventional ways of measuring the 
length of an object, we use the reverse method by changing the 
height of an object to predict its locations of base and shadow in 
the image. With the polygon of the flat top of an vertical object 
drawn, the line from each vertex of the polygon to its base, and 
line from vertex to its shadow are also drawn according to the 
height of the object. Changing the height of the object will 
change the position of the polygon in 3D (map coordinates), as 
well as change the position and length of the lines in 2D (image 
space). Once the lines touch the ground base or the shadow 
edge by observation, the height is set, and the position is fixed. 
This paper presents a semi-automated method for height 
measurement and 3D object reconstruction from single image 
by combining the above methods using the building top, base 
and shadow information. We also design an interactive software 
system for realtime 3D reconstruction, multi-dimensional 
visualization based on our previous development of a 3-D 
visualization system (Huang et al., 2006) to provide a complete 
package. 
The software package has been implemented on a standard PC 
running the windows operating system. A user-friendly 
graphical user interface complete with drawing and editing 
tools and a 3-D visualization component is introduced. The 
geometric accuracy of the 3D objects reconstructed from single 
image is compared to those generated from stereo images due to 
the lack of 3D ground truth data. Fortunately, the 3D 
information from stereo images is achieved quite reliably.