on of Urban 
rans. IEICE, 
-Plane Image 
ructure from 
Vision, 1, 1, 
M.Sakauchi: 
ised on Self- 
Vol.J82-D-II, 
  
METHOD OF THE CURRENT PITCH ANGLE VALUE CALCULATION 
ON THE BASIS OF PASSIVE SENSORS DATA PROCESSING 
IN INTELLIGENT DRIVING SYSTEMS 
V. M. Lisitsyn, K. V. Obrosov, A. A. Spiridonov, S. V. Tikhonova, 
State Research Institute of Aviation Systems, Moscow, Russia 
7, Victorenko Street, 125319, Moscow, Russian Federation - lvm@gosniias.msk.ru 
Commission V 
KEY WORDS: Image processing, Real-time vision, Algorithms, Robotics, Stereo vision. 
ABSTRACT 
One of the tasks of car intellectual driving systems is determination of own orientation of the vehicle and, in particular, of pitch angle 
relative to a near zone of a road surface. It is supposed, that the system is equipped with looking forward passive stereo cameras. The 
method of determination of a pitch angle current value is considered on the basis of a sequence of the stereoimages processing. 
The parallel, independent calculation of the pitch angle based on road marking lines and on the remote objects correlation tracking is 
made, with the subsequent decision about a choice of a true value of the pitch angle. 
The estimation of an increment of the pitch angle frame by frame is formed by remote objects correlation tracking. For stable work 
some fragments of the previous image are used as the etalon, as well as appropriate fragments of the current image. 
The determination of the pitch angle on road marking lines is based on calculating of a position of road marking lines point of 
convergence on the stereoimages. To increase accuracy of the pitch angle calculation the method of dynamic correction of direct 
measurements is developed. This method consists in that for determination of road marking lines declination on the stereoimage the 
coordinates of the bottom points of the detected lines are used, and the expressions, defining angles of declination of straight lines, 
are expressed by constants and slowly varying parameters. 
1. INTRODUCTION 
One of the tasks of collision warning and car intellectual driving 
systems is tracking of traffic lane road marking lines. Here to 
detect road marking lines it is necessary to process the 
significant part of the image. Besides it is necessary to have an 
estimation of a mutual disposition (positional relationship) and 
orientation of the cameras relative to the road (Broggi, 1999a, 
Broggi, 1999b). While processing large fragments of the image 
here exists a significant probability of detection of false 
formations instead of road marking lines (misoperation). 
The most interesting are systems with dynamic data processing, 
which allow in dynamics to estimate car own position on the 
road and to predict a position of road marking lines to be 
detected on the images at any exterior perturbations of the 
vehicle (Lisitsyn, 1998). 
While solving a problem of optimization of distribution of 
search gains at road marking lines tracking there is a natural 
inconsistency between the wish to reduce zones of searching 
and possibility of line tracking failure. 
Information about car’s own orientation relative to a road is 
necessary to prevent road marking lines tracking failure in case 
of exterior perturbations and lack of stabilization. In this case 
(Here) deciding factor is correct determination of current value 
of a pitch angle relative to near zone (NZ) of the road surface. 
It is supposed, that the system of intellectual driving is equipped 
with looking forward passive stereo cameras, similarly to a 
system considered in (Lisitsyn, 1998). In this paper the method 
of a pitch angle current value definition is considered on the 
basis of processing a sequence of the stereoimages. 
In case of driving on highways one of the functions of system is 
tracing of road marking lines in a near zone. In the elementary 
case the car moves on a straight section of road and all road 
marking lines are detected on a stereo image. In this case pitch 
angle, as well as an angle of a roll and course angle, of system 
relative to NZ can be calculated by results of road marking lines 
detection. For road marking lines detection and their direction 
finding some methods can be used, for example, Hough 
transform. However actually situation usually is more complex. 
In particular: 
e detected fragments of the right and left road marking lines 
could be not parallel on road turns, especially in case of not 
continuous lines; 
e extraneous objects, including the cars moving on adjacent 
traffic lanes or ahead on the same traffic lane, can 
influence correctness of road marking lines detection and 
their direction finding; 
e variety of road and weather conditions could became the 
case of absence of marking lines on the road surface or its 
detection failure for a long time. Besides there are 
situations when either left, or right marking lines only are 
detected; 
e inclination angles of straight fragments of the detected 
marking lines often are determined with significant errors 
because of insufficient accuracy of determination of their 
tops position. This error occurs because tops of marking 
lines fragments are observed on much bigger distances and 
under more flat angles, than their bottom parts. 
—573—