1305
DEVELOPING AND IMPLEMENTING LINE-BASED TRANSFORMATION MODELS TO
REGISTER SATELLITE IMAGES
Ahmed F. Elaksher
Faculty of Engineering, Cairo University, Egypt - ahmedelaskher@yahoo.com
KEYWORDS: Rectification, Satellite Images, Linear Features, Transformation Models
ABSTRACT:
Advances in remote sensing technologies demonstrated the capability to acquire high quality metric information from satellite
images. The IKONOS satellite image, for example, has a pixel size of one meter. The sensor model parameters are not yet released.
This motivate the use of empirical transformation models to represent the relationship between image space coordinates and object
space coordinates. Most of the employed transformation models are based on point features. However, these transformation models
could be driven using linear features. Linear features are easier to digitize than point features and they improve both the geometric
strength and the redundancy of the adjustment model. This research investigates the rectification of a single Geo panchromatic
IKONOS image using different line-based transformation models. Experiments are conducted using the 2D line-based
transformation models using 6 and 8 parameters and the line-based DLT model. Different sets of ground control lines are tested.
Results showed that less than two meters horizontal RMSE could be achieved. In addition, insignificant differences are noticed
between the transformation parameters computed using the point-based and the line-based transformation models. Results are
compared with those of point based transformation model and they are stable.
1. INTRODUCTION
The recent introduction of IKONOS and other high-resolution
commercial satellite imaging systems has initiated a new era for
large scale mapping. Space Imaging provides a number of
IKONOS image products with different processing levels
including the Geo, Reference, Pro, Precision and Precision Plus
products that have corresponding absolute positioning accuracy
Root Mean Square Error (RMSE) of 25, 11.8, 4.8, 1.9 and 0.9
meters, respectively [1]. Highly accurate products, such as
Precision and Precision Plus, are much more expensive than the
Geo products. Space Imaging has refused to release information
on the sensor model for the IKONOS imaging system, as well
the precise in-flight position and the attitude data of the imaging
sensor. The orientation information of IKONOS images is
available in the form of a so-called Rational Functions (RF)
model. Recently, several 2D and 3D transformation models
have been employed to register IKONOS stereo images using
different sets of ground control points. Different investigations
reported that the rectification of a single Geo panchromatic
IKONOS image using point features could achieve the required
accuracy for large scale mapping. Most of these investigation
are done using point features.
Recently, line based modelling has stimulated a great interest.
Different investigation have been done on the use of linear
features in digital photogrammetry. The advantages of
employing linear features in digital photogrammetry is
summarized in [2]. Linear features add more information, they
have higher semantic than point features, and they are easier to
detect than point features. Geometric constraints are more likely
to exist among linear features. This will eventually improve the
adjustment process. Moreover, linear features have the
advantage that they can be defined by segments. They can be
easily delineated in digital images either manually or
automatically. Corresponding ground space lines could be
identified from digital maps, GIS layers, or by GPS surveying
techniques.
This research investigates the potential of using straight lines to
rectify a single IKONOS Geo panchromatic image. The line-
based 6 parameters transformation model, the 8 parameters
transformation model, and the Direct Linear Transformation
(DLT) model are presented and examined. Different sets of
Ground Control Lines (GCLs) are generated and tested.
Checklines and checkpoints are used to evaluate the
rectification process. Results showed an RMSE of about one
meter using line based transformation models using either 6 or
8 parameters with 6 GCLs. The DLT model showed an RMSE
of about two meters. The results were stable and showed
insignificant differences between different sets of GCLs. In
addition, the results are compared with the results of the point-
based transformation models and insignificant differences
between the parameters are noticed.
2. BACKGROUND
Recent advances in mapping technology satellite have produced
high resolution satellite imaging systems. Mapping systems
based on high-resolution satellite images are increasing. The
accuracy of these systems is still under investigation.
Researchers mainly focus on the analysis and assessment of
using point based transformation models in the rectification of
satellite images. For example, the use of the Rational
Polynomial Coefficients (RPCs) to model the IKONOS sensor
using ground control is investigated in [3] and [4].
A number of investigations have been reported concerning the
accuracy attainable by various methods of processing IKONOS
stereo images. For example, a full suite of new methods and
software package SAT-PP (Satellite Image Precision Processing)
for the precision processing of satellite images is developed and