struction. 
for various 
nce system 
liction. For 
dimensions 
solved by 
uction and 
ies showed 
) results in 
1. This fact 
ind relative 
igh degree 
| high-way 
| in spatial 
| length for 
basis of the 
This results 
parameters 
of three- 
orientation 
ed on using 
reasonable 
; estimation 
coordinate 
ene. 
was a low 
he method 
automated 
etric system 
are used as 
n based on 
and special 
essing. The 
t procedure, 
re of image 
' calibration 
| targets as 
reference points markers for automatic reference point 
identification and precise determination of image coordinates 
of reference points. It allows to calculate image coordinate of 
reference points with sub-pixel accuracy, using elliptic 
approximation for target image. The correspondence problem is 
solved automatically due to target code recognition. 
Below the description of the proposed technique along with 
some numerical results of stereo vision system calibration and 
application are presented. 
24. SYSTEM LAYOUT 
The developing system of stereo vision is designed as driver 
assistance system. It is aimed for the solution of such tasks as 
lane markings detecting and own lane recognition, obstacle 
detecting in own lane and estimating obstacle parameters. The 
obstacle parameters to be determined are distance from a 
vehicle, wide of the obstacle, position of the detected obstacle 
relating to the own lane. 
For methods developing and testing in real road condition a 
laboratory vehicle based on *Volga GAZ-3110" automobile 
was built up. The hardware of the laboratory vehicle includes: 
e two CCD video cameras, located on the windshield of 
the vehicle; 
e personal computer (with frame grabber and special 
real-time image processing cards), located in the 
vehicle boot; 
e two DC power sources, located in the vehicle boot; 
e monitor, located in front of a driver to the left of the 
steering wheel. 
The scheme of hardware location and photographs of camera 
and personal computer installed in laboratory vehicle are shown 
in Figure 1 and Figure 2 respectively. 
  
Figure 1. Scheme of hardware location 
  
Figure 2. Personal computer and camera setup 
The strict requirements to designed system of detecting an 
obstacle with 10 cm elevation over the road at the distance up 
to 80 meters force to choose telescopic lenses with focal length 
of 18 mm to provide sufficient image scale. The maximal 
possible stereo basis of system is 1.05 m. So the aim of the 
work was to develop the procedure of system calibration, which 
could be performed in on-line mode and could give high 
accuracy of 3D measurements. 
For calibration system simplicity and automation possibility the 
planar test field is used both for system calibration and for 
relative orientation. 
3. CALIBRATION 
The proposed method of system calibration does not require 3D 
test field with high accuracy of knowing reference points 
coordinate. The basic requirements to test field is its planar 
surface and knowing of some reference distances between some 
reference point with given accuracy. For automated calibration 
procedure a pre-determined set of images is acquired at given 
test field positions. Then calibration procedure is performed 
automatically resulting in estimation of unknown vector 
parameter including parameters of camera interior orientation: 
principal point, scales in x and y directions, the radial 
symmetric and decentering distortion. 
During the calibration procedure the next parameters are also 
estimated: exterior orientation parameters of all the images, 
coordinates X, Y of all the object reference points excluding 
two points determining object space coordinate system. With 
the plane test field assumption Z coordinates of all points is 
taken to be zero. 
The unknown parameters are found as least mean square 
solution by iteration process. The original method for initial 
approximation for exterior orientation parameters is used. It 
allow automatically to determine reliably first approximation 
for iteration procedure, thus resulting in short unknown vector 
estimation time. The developed method for initial 
approximation determination is given below. 
3.1 Automated initial approximation determination 
The method for one-step initial approximation for exterior 
orientation parameters is developed for planar test field. It 
supposes that spatial coordinates of the test field and 
corresponding image coordinates are known. 
The general equations of central projection are: 
.(X 7 X5)a;; +(Y — Yo )a12 +(Z — 20 )a;3 
(X— X9)a31 +(Y — Yp )A32 +(Z — Z JA 33 (1) 
(X— X09)a2; +(Y —Y) Ja 72 +(Z-Zg)az; 
(X - X,)a5; * (Y - Y;)aj; € (Z - Zo)as; 
x, == 
Va=-f 
where 
f - camera constant, 
Xa Va — Image coordinates, 
Xo Yo, Zo — camera spatial coordinates 
X, Y, Z — object spatial coordinates, 
A11,, d33 — elements of transition matrix A. 
> 
-do—