bration. 
ent methods in 
togrammetric Record, 
jision in. Close-Range 
ry Engineering and 
1982  pp.561-570 
nization and Design of 
> 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 5. Hakodate 1998 
DETERMINATION OF CAMERA'S ORIENTATION PARAMETERS BASED ON LINE FEATURES 
Tianen CHEN, R.S. SHIBASAKI (Associate Professor) 
IIS, University of Tokyo 
7-22-1 Roppongi, Minato-ku, Tokyo 106-0032 
E-mail: chen@skl.iis.u-tokyo.ac.jp 
Japan 
Commission V, Working Group V/1 
Keywords: exterior orientation, straight lines, space resection, robust. 
Abstract 
Determining camera’s orientation parameters is one of the most important tasks in photogrammetry and robot 
vision. The paper presented a new mathematical model to determine the position and attitude of camera. The 
technique employed the coplanar relationship of the projection center, 3D straight line and its 2D image to 
construct the basic constrain equations. In the processing, the position and attitude of camera were divided into 
two steps and computed respectively. Since amount of obvious known geometrical objects such as horizontal and 
vertical lines exist in man-made environment, they are all be able to be used as controls effectively in our new 
method. This made it possible to do the space resection with fewer absolute controls. The method was based on 
stronger theory of mathematics and geometric, so it’s robust, accurate and possesses semi or full automation. 
The proposed approach was tested with simulated and real data. The result verified the method robust, 
accurate and reliable automation. 
1. INTRODUCTION 
Determining the orientation parameters of camera is 
one of the most fundamental tasks in photogrammetry 
and machine vision. These parameters include 
camera's position and orientation (exterior parameters) 
in world coordinate system, the true image center, scale 
factor and the lens focal length (interior parameters) 
and the distortions of lens (radial and decentring). For 
the past decades, a lot of methods have been developed 
and served us well in many applications ( Fukui, 1981; 
Lenz and Tsai, 1988; Fischler and Bolles,1981; Tsai, 
1986; Maybank and Faugeras, 1992 ). However, among 
these methods almost all are point-based. They are 
more and more not enough for the need of the coming of 
digital photogrammetric era because of their several 
weaknesses: time consuming, error-prone, necessity of 
some sort of structured target or calibration range, and 
difficulty of automation. 
Straight edges are the most popular features in large 
scale images of man-made environment. From the view 
point of image processing, they are easier detected and 
extracted from a digital noisy image at subpixel 
accuracy than point features. This makes the 
measurement automatically In addition, amount of 
geometric constrains such as parallel, perpendicular, 
horizontal and vertical can be used as controls. These 
characters make them used in many applications such 
as determining the camera's orientation parameters 
automatically relative and absolute orientation 
( Mulawa and Mikhail, 1988; Tommaselli and Lugnani, 
23 
1988; Dhome et al, 1989; C.L. Tozzi, 1986; Chen et al, 
1989; Salari and Jong 1990; Liu et al 1990; Wang and 
Tsai, 1990; Lee et al, 1990; Echigo, 1990; Chen and 
Jiang, 1991; Chen and Tsai, 1991; Tommaselli and 
Tozzi, 1992). 
The paper presented a new mathematical model 
which is based on the coplanar condition constructed by 
the perspective center, straight line in object space and 
its projection on the image plane to determine the 
camera’s exterior parameters. In the method the 
exterior orientation parameters were divided into two 
groups (position and attitude) and calculated 
respectively. This character makes it possible to use 
some known information provided by GPS and INS and 
reduce the calculation. 
In this method the calibration targets include a set of 
parallel lines (horizontal and vertical) and one more 
known control lines. These straight lines are extracted 
with dynamic programming method firstly and then 
fitted in the least square from simulated and real 
images. The sub-pixel measuring accuracy of image can 
be ensured. 
2. MATHEMATICAL MODEL 
As other computer vision applications we take the 
interior parameters of the camera remain as stable in 
the solution and calibrated in advance. In addition, the 
lens distortion parameters are known.