METHOD FOR ON-LINE CALIBRATION FOR AUTOMOBILE OBSTACLE
DETECTION SYSTEM
V. A. Knyaz *,
State Research Institute for Aviation System, 7, Victorenko str., Moscow, Russia
knyaz@gosniias.msk.ru
Commission V, WG V/4
KEY WORDS: Close-range photogrammetry, camera orientation, coded targets, calibration, automation, accuracy, reconstruction.
ABSTRACT:
The recent progress in digital photogrammetry creates a backgroun
applications and in particular for intelligent transportation system. I
d for applying close-range photogrammetric technique for various
ntelligent vehicle is often equipped with driver assistance system
which helps to solve various drivers task including such an important problems as own lane detection and collision prediction. For
collision prediction driver assistance system has to detect an obstacle in own lane and to determine the position and the dimensions
of the detected obstacle. The problem of obstacle geometrical parameters determination could be accurately solved by
photogrammetric system providing high accuracy of 3D measurements for adequate road surface 3D model reconstruction and
reliable estimation of obstacle parameters. The theoretical analysis and investigation of the images of typical road scenes showed
that the accuracy of common relative orientation procedure (eliminating vertical parallaxes for corresponding points) results in
unreliable estimation of relative orientation parameters for poor stereo condition of automobile photogrammetric system. This fact
caused a low precision of spatial measurements for such a system.
orientation which provides required precision for tasks of obstacle
The paper presents methods for on-line calibration and relative
detecting and estimating. The proposed method has high degree
of automation and gives high accuracy of 3D measurements approved by investigations in tested area and in real high-way
conditions.
1. INTRODUCTION
Last years vision techniques are widely involved in more and
more areas of human activity due to such factors as great
amount of information in vision flow inputting from a video
camera and a significant progress in image processing. One of
the challenging fields for vision techniques applying is an
intelligent transportation system. Cameras installed on an
automobile are used for various tasks solution beginning with
parking assistance and finishing with complicated obstacle
recognition problem. Among the tasks solving by vision system
are lane departure analysis, ego-motion estimation, road
geometry reconstruction and analysis, obstacle detection and
recognition.
Applying a stereo vision instead of mono-camera system for an
intelligent vehicle seems to be rather attractive because of
increasing data flow and possibility for accurate measurement
performing. One of the important tasks which an intelligent
vehicle has to solve is obstacle detection in a road and
estimation of the position and dimensions of the detected
obstacle. For resolving this problem one needs as possible high
accuracy of 3D measurements which provides adequate road
surface 3D model reconstruction and reliable estimation of
obstacle parameters. So when designing vision system for
obstacle detection one has to reach contradictory requirements
of high accuracy of 3D measurements and geometrical
restrictions caused by vehicle dimensions.
On the one side for obtaining required precision in spatial
measurements one has to use lenses with long focal length for
suitable image scale. On the other side the possible basis of the
stereo-system is restricted by vehicle dimensions. This results
in bad condition for relative orientation parameters
determination and therefore in low precision of three-
dimensional measurements.
Some methods for mobile system calibration and orientation
were developed (Knyaz V., 1999), which were based on using
spatial test field. This technique has demonstrated reasonable
accuracy of a stereo system orientation parameters estimation
providing required accuracy for spatial ^ coordinate
determination of objects belonged to an observed scene.
The main disadvantage of the proposed technique was a low
level of automation, this factor decreasing the method
applicability for an obstacle detection system.
This paper presents the method for on-line automated
calibration and relative orientation for photogrammetric system
based on an automobile. Two CCD video cameras are used as
image sensors for an obstacle detection system based on
personal computer equipped with frame grabber and special
image processing cards for real time image processing. The
proposed method is based on bundle adjustment procedure,
using special 2D test scene and special procedure of image
acquisition for calibration. The automation of calibration
procedure is provided by using original coded targets as
* Correspondence: Email: knyaz@gosniias.msk.ru, Telephone: 7-095-1573127, Fax: 7-095-1573900
SAS
ju aaa s.
ONS IY Rd
— Pp E. "ms m FANS 11:
ln
11
